Abstract
While scholars have for some time debated the role of refugee flows in the international spread of conflict, most evidence has been indirect due to the scarcity of systematic data on refugee-related violence. The Political and Societal Violence By And Against Refugees (POSVAR) dataset addresses this lacuna by providing cross-national, time-series data on refugees’ involvement in acts of physical violence in their host state, either as the victims or the perpetrators of violence, individually or collectively, in all countries between 1996 and 2015. In this article, we provide an overview of the main features of the dataset, identify its limitations, and trace variation in reported levels of refugee-related violence over time and across different types of actors. We emphasize that the data may be helpful to both researchers and policymakers for more accurate understanding of the prevalence of refugee-related violence and the design of more optimal policies to mitigate it.
Introduction
One of the most studied aspects of forced migration is the potential security consequences of refugee groups for host countries. Refugee flows have been associated with an increase in the risk of interstate wars, civil wars, terrorism, communal violence, and crime rates, especially if host states are unable or unwilling to address potential negative externalities (Choi & Salehyan, 2013; Fisk, 2018; Lischer, 2003, 2005; Salehyan & Gleditsch, 2006; Salehyan, 2008, 2009; Whitaker, 2003; Zolberg, Suhrke & Aguayo, 1989). The policy implications of this research program are consequential not only for the security of states hosting refugees but also, and more importantly, for the well-being of millions of individuals who leave their homes to escape violence and persecution. With the global increase in the number of forced migrants and the rise of far-right parties and anti-immigrant rhetoric, it is increasingly imperative that scholars and policymakers have an accurate understanding of the prevalence and dynamics of refugee-related violence.
While recent research has made important strides in advancing our understanding of refugee-related violence, the relevant empirical literature has yet to overcome two major empirical and theoretical challenges. First, refugees are often portrayed as violent actors, when strong anecdotal evidence suggests that they are generally the victims of abuse. In the absence of systematic empirical research on refugee victimization, scholars are at risk of producing a biased and potentially misleading picture of refugee-related violence. Second, the extent to which some refugees may be intentionally responsible for the spread of conflict remains unclear. Although empirical evidence suggests that larger refugee populations are associated with higher levels of violence in host states, further systematic research is required to understand whether the diffusion of violence is intentional on the part of refugees.
The purpose of the Political and Societal Violence By And Against Refugees (POSVAR) dataset is to help scholars overcome these challenges by offering new and systematic data on individual and collective acts of refugee-related violence. POSVAR provides information on a wide range of violent activities targeted at or perpetrated by refugees, including government victimization, societal violence, and terrorism, in all countries between 1996 and 2015. The data may be used by researchers to analyze variation in refugee-related violence over time and space as well as investigate new and/or untested questions such as: does humanitarian aid shape states’ treatment of refugees? Can education reduce societal violence against refugee populations? And under what circumstances do states fail to uphold their responsibility to protect refugees from terrorist attacks?
The article proceeds as follows. The next section further explains why a new global dataset on refugee-related violence would enrich our understanding of transnational political violence. The third section defines the main concepts as well as the scope of the dataset. Next, we discuss our methodology, including our coding rules and the sources we used to collect and code the data. In the fifth section, we present a series of descriptive statistics. We then use the POSVAR dataset to conduct preliminary tests of some of the arguments found in the empirical literature on refugee-related violence. We conclude by briefly discussing the policy implications of the prevalence of refugee-related violence we observe in the data.
Why a new dataset?
While previous research on forced migration has noted a positive relationship between refugees and the likelihood of political violence, only recently have scholars begun to test the conditions under which refugee-related violence is likely. For example, Rüegger (2019) demonstrates that refugees increase the risk of civil war when they share ethnic ties with excluded minority groups in host states. Similarly, Fisk (2019) provides subnational-level evidence for the increased risk of communal violence when the communities hosting refugees are politically marginalized and have ethnic ties with refugee groups. In both cases, refugees do not affect the risk of civil war or communal violence unless there is a strong pre-existing political discrimination in refugee-hosting states. In a similar vein, Böhmelt, Bove & Gleditsch (2019) show that the capacity of host states attenuates the effect of refugees on non-state actor violence. Detailed country-case studies on anti-refugee violence also raise important questions but are yet to be tested in a cross-national setting: to what extent does the excessive use of force by government agents trigger refugee violence (Fiske, 2013) and can good governance advance refugees’ right to be protected from sexual violence in their host states (Ho & Pavlish, 2011)?
While these studies are important steps in the right direction, data limitations have hampered scholars’ ability to test some crucial questions on refugee-related violence. Fisk (2018:10), for instance, argues that ‘attacks on self-settled refugee communities may […] be part of an overall attempt by governments fighting insurgencies to bolster support by punishing refugees – as an out-group – for the host country’s security problems.’ Yet, in the absence of data on anti-refugee violence, the claim made by Fisk has yet to be tested in a systematic manner. Similarly, Choi & Salehyan (2013: 57) acknowledge that ‘the presence of refugees and foreigners in general may prompt right-wing anti-immigrant groups to attack people who are ethnically and culturally different’. Their quantitative analysis does not, however, test whether or when this phenomenon is at play. In a recent study, Wright & Moorthy (2018: 20) call for a disaggregated measure of state repression and state that to better understand the processes that underline the positive association between refugees and state repression, ‘it would be helpful to know if the state is targeting incoming refugees […], or if they are targeting their local population’.
The POSVAR dataset will not only allow scholars to test some of these theoretical claims, but will also be of great use to answer diverse questions on political violence, including the diffusion and onset of civil conflicts (Forsberg, 2014; Miller & Ritter, 2014; Salehyan & Gleditsch, 2006; Salehyan, 2009), one-sided violence and repression (Danneman & Ritter, 2014; Fisk, 2018; Wright & Moorthy, 2018), insurgent recruitment strategies (Eck, 2014; Hegghammer, 2013; Weinstein, 2005), terrorism (Choi & Salehyan, 2013; Milton, Spencer & Findley, 2013), pro-government militia behaviors (Carey, Mitchell & Lowe, 2013), and scapegoating and domestic diversion (Klein & Tokdemir, 2016). Research on refugee integration (Strang & Ager, 2010; Valenta & Bunar, 2010) can also benefit from the dataset.
The POSVAR dataset also distinguishes between violence perpetrated by individuals and collective violence. Unlike acts of violence perpetrated by isolated individuals, acts of group violence require some level of coordination and may be particularly likely to find their roots in deeply rooted collective grievances. Furthermore, collective violence by or against refugees is likely to have a greater impact on the host government’s stability than violent events caused by isolated individuals. POSVAR will enable researchers to gain new insights into actors’ motivations for resorting to violence, individually or as a group. It will also shed light on brutal forms of collective violence, including group violence by local residents against refugees.
Definitions and scope
Refugees
Following Salehyan & Gleditsch’s (2006: 341) pioneering study on the international dimensions of civil war, and in line with the existing international treaties governing refugees, we define refugees as ‘anyone who flees a country of origin or residence for fear of politically motivated harm’. This definition explicitly excludes internally displaced persons (IDPs), who are unable or unwilling to flee their country of origin or residence, and returnees. 1
Violence against refugees
We define violence as the intentional use of physical force against another person, or against a group, that results in injury, death, or the destruction of the victim’s property. Violence is coded ‘collective’ when at least two individuals are identified as the perpetrators.
Government violence against refugees
We define anti-refugee government violence as any act by which state agents seek to physically harm refugees or damage/confiscate their property. Government agents include military personnel, police, prison personnel, other security officers, and any other individuals employed by the state.
Civilian violence against refugees
Civilian or societal violence occurs when local residents of the host country resort to violence against at least one refugee and/or her property. Civilian violence may be perpetrated by individual civilians or by groups of civilians. Violence may or may not be motivated by political considerations. An example of civilian violence includes violent acts perpetrated by neo-Nazi protesters against refugees in Finland in 2015.
Non-state actor violence against refugees
Non-state actor violence can be perpetrated by individual members or by several members of a terrorist, insurgent, or pro-government militia groups. Militias such as the Janjaweed and terrorist groups such as the Islamic State of Iraq and the Levant (ISIL) often engage in violence against refugees.
Terrorism against refugees
Terrorism is the ‘premeditated use or threat to use violence by individuals or subnational groups to obtain a political or social objective through the intimidation of a large audience beyond that of the immediate noncombatant victims’ (Enders & Sandler, 2011: 4). Terrorist incidents may involve one or several perpetrators and qualify as terrorism against refugees if at least one of the targets/victims is a refugee.
Violence by refugees
Refugee-on-refugee violence
Refugee-on-refugee violence occurs when both the perpetrator(s) and the victim(s) of a violent event are identified as refugees. Refugee-on-refugee violence may be perpetrated by individual refugees or by groups of refugees. It may be motivated by political, ideological, ethnic, or criminal considerations. Instances of refugee-on-refugee violence include interclan violence among rival Somali clans in Kenya and sexual violence perpetrated against female refugees within Tanzanian refugee camps.
Refugee violence against civilians
This type of violence occurs when refugees resort to violence against at least one local civilian and/or her property, regardless of motivations. It may be perpetrated by individual refugees or by groups of refugees. Violent protests in Pakistan in 2014 involved refugees’ use of violence against civilians.
Refugee violence against the government
This type of violence occurs when refugees resort to violence against at least one government official, regardless of motivations. It may be perpetrated by individual refugees or by groups of refugees.
Terrorism by refugees
Terrorist incidents qualify as refugee-related violence if at least one of their perpetrators is a refugee.
Refugee riots
We regard riots as a form of refugee violence if at least some of the participants are refugees. They may entail the use of violence against state officials or local residents. Recent acts of collective violence in Turkey, Germany, and Hungary are instances of refugee riots.
Refugee recruitment by non-state actors
The recruitment of refugees into non-state armed groups occurs when refugees join a terrorist organization, a rebel group, or a pro-government militia, either voluntarily or as a result of coercion. Groups such as the Kamajors in Sierra Leone, Liberians United for Reconciliation and Democracy (LURD) rebels, and the Sudan People’s Liberation Army (SPLA) were responsible for the voluntary as well as forcible recruitment of refugees.
Coding procedure and data sources
The unit of analysis is the country-year. Each country has observations for each year between 1996 and 2015 irrespective of number of hosted refugees. Following previous work on political violence (Cohen & Nordås, 2014, 2015), we use an ordinal scale to create all but three variables included in the dataset. 2 The use of an ordinal scale helps ease concerns about under-reporting bias, which we discuss in greater detail below. For instance, while sexual violence against refugees in Kenya is likely to be under-reported, its systematic nature is well documented by news sources and human rights reports. An ordinal scale is therefore less likely to suffer from under-reporting bias than a count measure of refugee-related violence. Since information is often available to distinguish between isolated, common, and systematic acts of violence by and against refugees, we choose an ordinal scale of violence to provide a more nuanced picture of refugee-related violence. 3
Summary of coding rules for refugee-related violence and recruitment indices
In order to identify instances of refugee-related violence, we first searched for the keyword ‘refugee’ in each of the following sources: US State Department Country Report on Human Rights Practices (1999–2015); Armed Conflict Location and Event Data Project (ACLED) (1997–2015); Social Conflict in Analysis Database (SCAD) (1996–2015); and Sexual Violence in Armed Conflict (SVAC) Dataset (1996–2009). Each sentence that included the term ‘refugee’ and, if necessary, adjacent paragraphs were carefully read to find evidence of refugee-related violence and understand its context. We then searched for variable-specific keywords in the following sources: the USCRI World Refugee Surveys (1996–2008); the United Nations High Commissioner for Refugees (UNHCR) website (1996–2015); LexisNexis (1996–2015); ReliefWeb (1996–2015); and Proquest (1996–2015).
Government violence against refugees
Descriptive statistics: main ordinal variables
Descriptive statistics: main dichotomous variables
Descriptive statistics
Civilian and non-state actor violence against refugees
We adopt the same methodology and rely on the same sources and keywords to create Civilian violence against refugees. This category encompasses all forms of anti-refugee violence perpetrated by civilians of the host state. Using the same coding procedure, we also create two additional variables, Individual civilian violence and Collective civilian violence. Civilian violence is collective if at least two civilians are involved as perpetrators. Next, we create an index of non-state actor (NSA) violence against refugees. If an armed organization is identified as the perpetrator of violence against refugees, and is described as a rebel group, terrorist organization, or militia, then the variable Non-state violence takes the value of 1, 2, or 3 depending on the highest reported level of NSA violence against refugees. 5
Refugee-on-refugee violence
We conduct a systematic analysis of the aforementioned sources to assess whether refugees engaged in any type of violence against other refugees (Refugee-on-refugee violence), as isolated individuals (Individual refugee-on-refugee violence) or collectively (Group refugee-on-refugee violence).
Refugee violence against civilians
Using the same methodology and sources, we code Anti-civilian violence, an index of refugee violence against the civilian population.
Refugee violence against the government
Anti-government violence includes violence (such as riots) perpetrated by refugees against the host state’s agents, that is, military forces, police, and other security forces. 6
Terrorist attacks by refugees and terrorist attacks against refugees
We searched for the keyword ‘refugee’ in the Global Terrorism Database (GTD) from 1996–2015 to identify attacks in which refugees were involved, as targets, victims or as perpetrators. We also relied on information from the US State Department Country Report on Human Rights Practices (1999–2015), LexisNexis (1996–2015), and Proquest (1996–2015) to identify additional terrorist attacks perpetrated by or against refugees. If refugees were described as being involved in a terrorist attack, either as victims/targets or as perpetrators, then we verified whether the violent event described in the report was also categorized as a terrorist attack by the GTD, based on information on the date and location of the violent event. Every violent attack described as a terrorist attack by or against refugees is coded as a refugee-related terrorist attack as long as it is also categorized as a terrorist attack in the Global Terrorism Database. Terrorism against refugees and Terrorism by refugees are count variables.
Refugee riots
Riots are ‘distinct, continuous and violent action directed toward members of a distinct “other” group or government authorities, and whose participants intend to cause physical injury and/or property damage’ (Salehyan et al., 2012: 2). We code refugee riots as 1 if there are any reports of riots involving refugees in the host country, otherwise as 0.
Refugee recruitment by non-state actors
We rely on a variety of sources to create NSA recruitment, an index of refugee recruitment by non-state armed groups. The Codebook and User Instruction Guide provides the list of keywords and sources we used to identify refugee recruitment by non-state actors.
If an armed organization is identified as conducting recruitment activities among a host country’s refugee population, we assess whether that group is a terrorist group, a rebel organization, or a pro-government militia. Furthermore, we code whether recruitment occurred on a voluntary basis or under coercion. Forcible recruitment takes the value of 1 if at least one source describes the recruitment of refugees as ‘involuntary’, ‘forced’, ‘coerced’ or caused by ‘abduction’. 7
Number of countries affected by refugee and anti-refugee violence proportional to total number of countries hosting at least 10,000 refugees (1996–2015)
Descriptive statistics
Tables II, III, and IV provide descriptive statistics of the variables in the dataset. To trace variation in reported refugee-related violence, we first identify whether refugees are more likely to be the victims or the perpetrators of violence in their host states. Figure 1 provides information on the percentage of host countries affected by refugee violence and anti-refugee violence among all countries hosting at least 10,000 refugees. It shows that, over the past two decades, violence against refugees was systematically reported in a larger number of host states than violence by refugees. This overall pattern seems to have picked up in recent years. Figure 2 provides information on the size of the largest total refugee population hosted by each country between 1996 and 2015, while Figures 3, 4, and 5 show the highest reported levels of government, civilian, and refugee violence by country.
Figure 6 confirms that reports of violence against refugees have become more prevalent. Among countries hosting at least 10,000 refugees, a larger number of host states experienced terrorist attacks against refugees than terrorist attacks by refugees. The share of host countries experiencing anti-refugee terrorism has increased considerably in recent years.
Figure 7 provides further information on the perpetrators of anti-refugee violence. States and their agents are the primary perpetrators of violence against refugees. Civilians are relatively less prone to anti-refugee violence, although this pattern appears to have shifted in recent years (Figure 8). And while refugee victimization by non-state actors (NSAs) is the least common form of anti-refugee violence, NSAs are generally responsible for the most systematic forms of violence against refugees, such as the 2004 massacre of at least 166 Congolese refugees by the Forces for National Liberation (FNL) or the killing of 107 Tutsi refugees from the Democratic Republic of the Congo (DRC) by Rwanda Hutu militiamen in 1997.
Figure 9 shows the prevalence of refugee violence against the host government, civilians, and other refugees. Refugee-on-refugee violence is the most frequent Highest number of hosted refugees (1996–2015) Violence by government against refugees, highest value of index (1996–2015) Violence by civilians against refugees, highest value of index (1996–2015)
Application
One of the strengths of the POSVAR dataset is that it allows us to test some of the mechanisms purported but not directly tested by the literature linking refugee flows to political violence. Due to space limitations, we briefly discuss one application here. According to pioneering work from Salehyan & Gleditsch (2006) on refugees, one of the mechanisms through which refugee flows Violence by refugees against civilians, highest value of index (1996–2015) Number of countries affected by refugee and anti-refugee terrorism proportional to total number of countries hosting at least 10,000 refugees (1996–2015)
However, due to the lack of available data, scholars have yet to demonstrate whether there is indeed a Prevalence of anti-refugee violence, by perpetrator (1996–2015) Number of countries affected by anti-refugee violence proportional to total number of countries hosting at least 10,000 refugees (1996–2015) Violence by refugees, by target (1996–2015)
Logistic regressions of civil conflict in host state (1996–2015)
Standard errors clustered by country in parentheses. **
Limitations of the dataset
The POSVAR dataset is the first systematic effort to code cross-sectional and time-series data on the prevalence of violence perpetrated by and against refugees in host states. Despite its virtues, the dataset suffers from a number of limitations that need to be acknowledged. First, similar to other large-scale data collection efforts on political violence, such as sexual violence (Cohen & Nordås, 2014) and terrorism (Drakos & Gofas, 2006), under-reporting bias is a problem. Violence against refugees is likely to be under-reported. Reasons for refugees not reporting violence may include the fear of deportation, shame, a lack of trust in host states’ institutions, or the fear of violent retribution by fellow refugees. In addition, one of our main sources of information, the United States Department of State Human Rights Country Reports, may be biased in reporting violations of physical integrity rights in US allies. We attempt to minimize the risk of political bias in State Department reports by using a variety of sources, including international news sources. 8 Finally, we acknowledge that under-reporting bias may exhibit a temporal pattern, as advances in modern technology may have facilitated the diffusion of information on refugee-related violence in recent years.
Second, due to lack of consistent data in country reports and news sources on the exact count of refugee-related violence, the dataset describes the prevalence of violence in an ordinal scale (0–3) rather than a continuous measure. While the exact count would be ideal, the risk of under- and over-reporting poses a greater challenge for continuous measures. It is also important to note that the absence of reports of refugee related violence (index = 0) does not necessarily mean that no refugee-related violence took place in a country-year. As Cohen (2013: 467) suggests, ‘although the four-point scale is a blunt instrument, it makes a contribution by allowing a systematic comparisons of relative levels of violence across time and space’.
Another important limitation of the POSVAR dataset is that it does not, at this stage, provide information at the refugee group level. Unlike other datasets such as the Minorities at Risk (MAR) project, the POSVAR dataset is not designed to provide information on the behavior or living conditions of ethnopolitical groups. While recent research has focused on refugee groups as the relevant unit of analysis (Rüegger & Bohnet, 2018), further efforts will be required to disaggregate refugee-related violence based on the national origin of refugees. Additional data collection will also be necessary to disaggregate refugee-related violence at the subnational level of analysis.
Conclusion
In this article, we introduce a new global dataset on refugee-related violence from 1996 to 2015. This new dataset will allow researchers to test previously untested questions on refugee-related violence and explore new areas of research on the impact of refugee movements. Our data indicate that refugees are generally more likely to be the victims of violence than its perpetrators, a trend that is likely to persist in future years. The data also show that while reports of refugees involved in terrorist attacks are relatively rare, the number of reported terrorist attacks against refugees has increased in recent years. These data patterns suggest that current migration debates in the Western democracies, in particular in the United States, may be ill-informed. While there has been undue emphasis on the perceived risk of violence by refugees, our dataset suggests that the protection of refugees should be of greater priority for host governments. Understanding why some states fail to uphold their responsibility to protect refugees should thus become a research priority for scholars of international law, human rights, forced migration, and political violence.
Footnotes
Replication data
Acknowledgments
We would like to thank Li Chen, Dong Ju Lee, and Kelly Morrison for their superb research assistance with the data collection and Alex Braithwaite, Idean Salehyan, and the editor, Gudrun Østby, for valuable comments.