Civilian victimization and ethnic attitudes in Africa

Abstract

Previous research shows that violence is an important factor driving ethnic identification and grievances, but most works that explore micro-level effects focus on specific cases and have limited external validity. This article looks at the individual-level consequences of civilian victimization in a large sample across Africa. Combining georeferenced survey data from several rounds of the Afrobarometer, victimization events from the UCDP-GED, and data on collective targeting from the ethnic one-sided violence dataset, it studies the effect of exposure to violence on ethnic identification and self-reported ethnic grievances. Results show that violence increases ethnic identification and ethnic grievances particularly when it is committed by state forces and among individuals who belong to an ethnic group that was collectively targeted in the past.

Introduction

What is the impact of civilian victimization on ethnic identities? The idea that violence hardens ethnic identities and feelings of ethnopolitical discrimination has long been present in sociopolitical research (Horowitz, 1985; Khawaja, 1993; Bar-Tal, 2007). A more recent body of research on the consequences of violence has provided ample evidence for this argument, but its external validity is limited (Davenport et al., 2019). With a few exceptions (e.g., Grosjean, 2014; Bauer et al., 2016), most of these works have focused on finding robust, micro-level research designs that prioritize causal identification but are contextually limited. As a result, although violence has been found to have an impact on political attitudes across several contexts, we know much less about how generalizable these effects are, or whether they depend on the type of perpetrator or other contextual variables. ¹

In this study, I contribute in filling some of these gaps. Using data from several countries in Africa, I analyze the individual consequences of exposure to violence on two measures of ethnic attitudes—ethnic identification and self-reported ethnic grievances. I focus on how this effect varies depending on two conditioning variables: whether violence was committed by the state or by rebels, and whether individuals belong to a group with a prior history of collective targeting. These two variables are usually specific to each context and do not vary internally within a conflict. Including them helps to understand the generalizability of case studies. In particular, I try to answer two main questions. Do we see consistent effects of exposure to violence on ethno-national identities across several contexts? And what is the importance of group-level processes for understanding the consequences of violence?

I argue that violence should have a larger impact on ethnic attitudes when it is committed by the state and when individuals belong to a group with a prior history of collective targeting. In these cases, violence produces a stronger emotional reaction given the supposed role of state authorities in protecting their citizens, and a group-based history of violence helps in the interpretation of these events and their translation into ethnic mobilization.

Empirically, I test this argument combining georeferenced survey data from Afrobarometer (BenYishay et al., 2017) with victimization events from the UCDP-GED dataset (Sundberg and Melander, 2013). Rather than coding personal victimization, I create a measure of exposure to violence based on spatial and temporal distance across 20 different countries. Using information on the ethnic identity of respondents, I match them to data on ethnic-based collective targeting (Fjelde et al., 2021) to develop the group-based measure of prior targeting.

Results support the expectation that the effect of violence is stronger among individuals who belong to an ethnic group that has suffered collective targeting and when it is perpetrated by the state. Additional analyses suggest that the effect of violence is dependent on group-level processes. Overall, this article contributes to the literature on the consequences of civilian victimization, showing how the effect holds across a larger set of countries and opening up the debate about the mechanisms that explain these effects.

Exposure to violence and ethnic attitudes

During the last few years, research on the consequences of violence, including its effect on ethno-nationalism, has grown considerably. We know that exposure to violence increases hostility towards the outgroup and makes ethnic reconciliation more difficult (Bakke et al., 2009; Beber et al., 2014). For example, exposure to wartime violence decreased political tolerance among Tamils in Sri Lanka (Rapp et al., 2019) and increased ethnic identification and ethnopolitical preferences in Bosnia (Hadzic et al., 2020). Research on civil war dynamics also suggests that violence against civilians increases grievances among the victims, which translates into group-specific processes of collective blaming (Wood, 2003; Condra and Shapiro, 2012; Lyall et al., 2013). This body of research is coherent with those works that focus on the legacies of political violence, which point out that individuals exposed to violence are more likely to reject the political identity of the perpetrator (e.g., Balcells, 2012; Lupu and Peisakhin, 2017; Rozenas et al., 2017). Thus, an increase in ethnic identification can be a response to violence in contexts where ethnicity is highly politized.

I focus on the effect of violence on ethno-nationalist identification and ethnic-based grievances. While there are several works exploring these or related questions, most of the evidence comes from specific contexts (Davenport et al., 2019). Thus, we do not know how this effect varies across contexts, and particularly whether it depends on the type of perpetrator or the previous history of group-level conflict.

I argue that the effect of violence on ethnic identification and grievances should be particularly strong when violence is committed by state authorities and when individuals have been collectively targeted because of their ethnic identity. First, state violence should have a larger effect because forms of victimization that are more likely to be perceived as morally wrong have a bigger impact on ethno-national identification and grievances. This idea is linked to the concept of ‘moral shock’ in social movement studies, when an unexpected event or information produces moral outrage and ‘helps a person think about her basic values and how the world diverges from them in some important way’ (Jasper, 1997, 106). Violence against civilians should have a similarly strong emotional reaction. Focusing on the attitudinal effects of criminal violence, García-Ponce, Young and Zeitzoff (2022, 6) argue that ‘violence that violates moral principles should be particularly likely to induce anger.’ Civilian victimization carried out by state authorities is likely to violate moral principles in a more profound way that violence by non-state actors, if anything because state forces are expected to protect citizens.

Previous works show that wartime state violence produces moral outrage and grievances (Goodwin, 2001; Wood, 2003). This effect can easily reinforce group identity because of group-based processes of blame formation and collective coping (Lyons et al., 1998; Bar-Tal, 2007), which should be more intense when facing violence by state authorities. As Khawaja (1993, 66) argues, ‘repression can strengthen collective identity, the sense of belonging to a group, by operating as a symbolic reminder of a group’s shared circumstance vis-a-vis authorities and their agents of control.’ In the context of ethnic politics, the effect of state violence on ethnic identification and grievances will be stronger because moral principles of ethnic inequality are usually defined on the basis of institutional relationships with the state. Previous research shows that state-led collective targeting against ethnic groups increases the risk of conflict escalation (Cederman et al., 2020).

Second, the existence of a history of collective targeting bolsters moral reactions to violence. This group-based background also strengthens group divisions, which should be mirrored in the attitudinal effects of victimization. More specifically, we should expect civilian victimization to reinforce the cleavages along which it takes place, increasing social polarization (Wood, 2008). The fact that an ethnic group has already suffered collective targeting strengthens the effect of violence along the ethnic cleavage. For instance, Mironova and Whitt (2018) show that individuals in war-torn societies show signs of in-group parochialism in contexts that are marked by those divisions. Moreover, where ethnic divisions have been polarized by previous rounds of violence, local social structures help to articulate an attitudinal response to violence. We could expect these structures to be particularly defined by group boundaries if there is a history of ethnic-based violence, contributing to the hardening effect of violence on ethnic divisions.

An important caveat in this argument is that ethnic attitudes are endogenous to the occurrence of violence against civilians. First, state violence can be a response to existing rebel activity or even a preventive strategy in the face of increasing mobilization (Moore, 2000). Relatedly, previous research also highlights that actors can use violence against the civilian constituency of the opposite group for strategic reasons, with the goal of weakening the enemy’s base of support (Fjelde and Hultman, 2014). Second, rebels themselves can also use violence as a mobilization device. Laitin (1995) argues that nationalist violence in the Basque Country was part of a strategy to increase support for a minority movement. Finally, the characteristics of civilians could also influence the use of rebel violence against their own constituencies, particularly when rebels rely on more heterogenous constituencies (Ottmann, 2017).

The direct implication is that any relationship of violence with ethnic attitudes could suffer from reverse causality. In other words, any increase in ethnic identification or grievances linked to state violence could actually be caused by rebel violence in the first place. Yet, even if the role of prior rebel activity is important, it is also difficult to say whether we would observe the same effect on ethnic attitudes without the use of violence by the state. Indeed, De la Calle (2015) offers a related argument where rebel violence in nationalist contexts is also influenced by the incentives of local political elites to prompt the central state into the use of repression, increasing the likelihood of conflict. Rather than identifying the effect of specific events of violence, the goal of this study is to analyze the contexts in which we should observe a stronger association between violence and ethno-nationalism. This implies acknowledging that violence, and individual events in particular, are influenced by the local conflict history. Taking this into account, my expectation is that we should observe a stronger link in those cases where the perpetrator is the state and where there is a history of group-based targeting.

Data and methods

I test these expectations using individual-level data that includes information on whether individuals were exposed to events of violence by either state authorities or the rebels, and whether they belong to a group with a history of collective targeting. To build such data, I combine three different datasets.

First, I use survey data from the first six rounds of Afrobarometer, fielded between 1999 and 2015, which also include georeferenced information (BenYishay et al., 2017). Using this survey, I create two binary outcomes measuring ethno-national attitudes: 1) ethnic identification, which indicates that the respondent identifies more by her ethnic groups than by her national identity, and 2) ethnic grievances, which is coded as 1 when the respondent feels her ethnic group is often or always treated unfairly by the government. I include these variables in binary form to maximize the number of compatible survey rounds. ² In the sample, 16% of respondents identify primarily with their ethnic group (varying from 33% in Nigeria to 4% in Burundi), and 21% express ethnic grievances (varying from 35% in Nigeria to 2% in Madagascar). These two variables are the two outcomes in the analyses.

Second, in order to build the variable on individual-level exposure to violence, I use the georeferenced information in Afrobarometer to match respondents to events of violence against civilians. I rely on the Georeferenced Event Dataset (Sundberg and Melander, 2013), a data project run by the Uppsala Conflict Data Program (UCDP) which codes events of violence across the world since 1989 from both news and secondary sources, and includes information on the actor, location, and date, among other variables. Knowing the location and timing of each event of violence allows me to build a measure of exposure based on spatial and temporal distance between the Afrobarometer respondents and violence events. I focus on one-sided violence (OSV), or events when civilians were killed by an armed actor, distinguishing between violence by state authorities or by rebel groups. I match each respondent to OSV events that took place within the same country and before the survey was fielded, and calculate the spatial and temporal distance to each of them. I then code a binary measure of exposure based on distance thresholds, detailed below.

The full sample includes around 150,000 respondents across six survey rounds in 20 countries, although the number of observations varies depending on the model. ³ Figure 1 maps these two data sources. The right panel, which shows the events of violence against civilians, includes both state and rebel violence, but the variables used in the model distinguish between these two actors. ⁴

Figure 1. Survey respondents and civilian victimization.

Finally, the third relevant variable is whether an individual belongs to an ethnic group that has a history of collective targeting. To code such variable, I rely on the Ethnic One-Sided Violence (EOSV) dataset (Fjelde et al., 2021). The EOSV dataset expands previous data by UCDP on violence against civilians by identifying each of the victims and linking them to an ethnic group from the ethnic power relations (EPR) dataset (Vogt et al., 2015). Using this information, the EOSV then codes, for each perpetrator-ethnic group dyad and year, whether there was evidence of ethnic-based collective targeting, indicating that there is evidence that at least half the victims for a given year were killed because of their ethnic identity. This is different from just identifying the ethnicity of victims that could have been targeted for other reasons.

In order to match survey respondents to the EOSV data, I use the LEDA R package (Müller-Crepon et al., 2021), which uses information from different variables in Afrobarometer to match them to EPR groups. I then build a binary indicator for respondents who belong to a group that has suffered collective targeting at some point before the survey was fielded. When looking at exposure to events of state violence, I focus on collective targeting by state authorities, and when looking at exposure to rebel violence, I focus on collective targeting by rebels. ⁵ I explain below the details of the victimization variables and how I combine them.

Individual and collective exposure to violence

Combining these datasets allows me to create a measure of individual-level exposure to events of civilian victimization and probe how its effects vary depending on whether there is prior collective targeting at the group level.

First, I code a binary measure of individual-level exposure when events of violence against civilians took place within 10km of a respondent’s location. I distinguish between two versions of this variable depending on when events of violence took place: 1) exposure to events that took place in the 5 years previous to the survey, and 2) exposure to events that took place before the 5 years previous to the survey, measuring longer-term effects. ⁶ In all cases, I also distinguish between state and rebel violence. Figure 2 illustrates these different measures of exposure to violence. This measure of exposure to violence is independent from any form of ethnic-based collective targeting. The base models just include these individual-level exposure variables.

Figure 2. Defining exposure based on temporal distance to OSV events.

Second, in order to probe whether the effect of violence is different on individuals who belong to an ethnic group with a prior history of ethnic-based victimization than on those who do not, I interact the exposure variables with the indicator of group-level prior collective targeting. As I explain above, I use an indicator of collective targeting by the same actor as the events of violence against civilians (namely, state or rebels).

Models

All models are logistic regressions, including the respondent’s gender and age, whether the respondent lives in an urban area, her employment situation (employed, partially employed, or fully employed) and the distance to the national capital (in logged km) as control variables. Data for other control variables are missing in some country-rounds, which would reduce observations available, and thus they are not included to maximize the number of observations. To account for this, I also include country-round fixed effects, effectively comparing only individuals within the same country and survey wave. Additionally, every model restricts the sample to those countries where there is variation in the main independent variables (either individual exposure or individual exposure and collective targeting).

Results

Table 1 shows the results for the models using exposure to violence in the 5 years previous to the survey. For each of the outcomes and for both state and rebel violence, I estimate a base model that just includes the individual-level measure of exposure to violence based on temporal and spatial distance, and a model that includes an interaction with the group-level measure of prior collective targeting. Figure 3 shows the results graphically. ⁷ The base models without any interaction show that violence only has a significant and positive effect on ethnic attitudes, in particular on ethnic identity, when it is perpetrated by state actors. Exposure to state violence increases ethnic identification regardless of group history, but in the case of ethnic grievances, state violence only has such an effect when there is a group-based history of collective targeting. In such cases, the effect is significant and substantive, with an increase of around 5 points in the probability of expressing ethnic grievances. In the case of rebel violence, the results again show that violence does not have any consistent effect.

Table 1. Civilian victimization and ethnicity (exposure last 5 years, within 10km)

Note: + P < 0.1, * P < 0.05, ** P < 0.01, *** P < 0.001. Control variables (not shown): urban, gender, age, employment situation, distance to national capital (in km, log). Country-round fixed effects not shown. Both violence exposure and collective targeting refer to the actors defined above (state or rebels).

Figure 3. Simulated effects of exposure to victimization during the previous 5 years to the survey.

Table 2 repeats these analyses but using the measure of exposure to violence before the 5 years previous to the survey, analyzing longer-term effects. Figure 4 shows the results graphically. Results are not very different to the previous ones.

Table 2. Civilian victimization and ethnicity (exposure before last 5 years, within 10 km)

Note: + P < 0.1, * P < 0.05, ** P < 0.01, *** P < 0.001. Control variables (not shown): urban, gender, age, employment situation, distance to national capital (in km, log). Country-round fixed effects not shown. Both violence exposure and collective targeting refer to the actors defined above (state or rebels).

Figure 4. Simulated effects of exposure to victimization before the previous 5 years to the survey.

The largest effect of violence is found again for ethnic grievances, with a marked increase for individuals who belong to previously targeted groups. Rebel violence does not show consistent effects.

How large are the effects of violence? Following Model 1 in Table 1, the probability that someone who has not suffered exposure to violence and does not belong to an ethnic group with a history of collective targeting and reports ethnic grievances is 0.40. That number increases to 0.47 when both victimization variables are positive, an increase of around 18%. In the case of Model 1 in Table 2, the probability increases from 0.37 to 0.42, or an increase of around 14%. To put these coefficients in context, comparing them to other control variables in the latter model, the average increase in the likelihood of reporting ethnic grievances in urban areas is 16% higher than in rural areas. In the case of distance to the national capital, there is an increase of 14% between someone who lives in the capital and someone who lives 500 km away.

I include a number of robustness tests and alternative specifications in the Online Appendix. The results remain consistent when using linear models on the outcome variables in continuous form (Appendix E) or when using multilevel models (Appendix G). Results using 5 km and 25 km distance thresholds are also consistent with the main results (Appendix D), even if the short-term effect of violence on grievances stops being significant when the time window is reduced from the last 5 years to the last 2, suggesting that some form of collective coping might be needed before these events trigger any change in ethnic attitudes.

An additional concern is that the wording and coding of the outcome variables was different in Rounds 1 and 2. In particular, ethnic identity was coded in a binary form in these rounds. To account for this, I include in Appendix I results excluding Rounds 1 and 2, which show that the main results do not change when limiting the sample to later survey waves.

Finally, another problem with these analyses is that they include completely different contexts, where the political trajectory of each country is distinct. For example, regime change can complicate the meaning of state and rebel violence and how it relates to individuals’ experience or the collective memory of an ethnic group. This limitation is the consequence of this research design, which tries to zoom out to see if the general effects of violence hold. In order to explore these differences in a more detailed way, Appendix H shows how the effect of violence, both individual-level exposure and group-level collective targeting, varies by country-round. I also discuss below a series of analyses using sub-samples, in connection to the potential mechanisms behind the main results.

Exploring the mechanisms

I run a series of complementary models in order to explore the mechanisms behind the main results. First, the measure of individual exposure used in the main analyses is defined in terms of time and distance. It cannot register individual-level variation within a same locality, and is more geared to measure the local or regional incidence of civilian victimization. In order to get closer to an individual measure of exposure to violence, I use an alternative coding that indicates whether a respondent was exposed to violence while she was between 15 and 25 years old. This measure offers variation in exposure to violence between individuals in the same geographical location and helps testing, even in an imperfect way, whether the effects of violence are contextual. In these models, I include region-round FE, thus comparing individuals within smaller geographical areas. ⁸ The results (Appendix F) do not show any significant effect of violence, which suggests that the relationship between violence and ethnic attitudes is probably more dependent on group-level dynamics that shape the response to them.

Second, if the positive effect of violence on ethnic attitudes depends on individuals having lived through these previous violent events, as opposed to second-hand experiences transmitted at the group level, we should see stronger effects among older individuals. I run the main analyses using three subsamples for individuals between 18 and 29 years old, between 30 and 49, and 50 or older. The results (Appendix J) show that it is actually among younger individuals (between 18 and 29 and between 30 and 49) where the effect is stronger and significant, and no effect is found in the oldest group. Even if further work is necessary to confirm this, it suggests that group mobilization or second-hand experiences could be more important than personal exposure.

Finally, the argument states that the strenghtening of ethnic attitudes as a result of violence, particularly when committed by the state, depends on perceptions of injustice vis-a-vis the state. If this is true, the effect should be weaker for ethnic groups that have direct representation in the central power. To test this, I run the main analyses on a subsample that only includes individuals who belong to an ethnic group that had access to central executive power at the time of the survey, using data from the EPR dataset (Vogt et al., 2015). Results show that there is no effect in these cases (Appendix K), which suggests that the effects of violence probably add up to general grievances about power inequalities.

Relatedly, I also test if the results change when the sample is limited to individuals who belong to minority ethnic groups, namely, those that make up 20% or less of the country’s population (Appendix L). This variable can be related to the likelihood of suffering collective targeting but also to a feeling of overall insecurity towards the state, both when they are targeted by the state itself or by rebel groups. Results show that the combined effect of individual-level exposure and prior collective targeting on ethnic grievances is much larger among minority groups.

Conclusion

This study combines survey data and data on violence from several African countries to show how exposure to civilian victimization influences ethno-national attitudes. Results show that exposure to violence increases both ethnic grievances and ethnic identification, particularly when violence is perpetrated by the state and among individuals who belong to an ethnic group with a prior history of collective victimization. I also offer tentative evidence on the importance of contextual processes in explaining the effect of victimization on political attitudes.

These findings speak to previous research on the positive effect of violence on ethno-nationalism. They show that the effects hold when tested on a larger sample across many different countries, but also that their magnitude is conditional on the type of perpetrator and larger in cases where there is a prior history of violent conflict, suggesting a few theoretical takeaways. First, moral reactions to violence are important in understanding its consequences, which probably add up to more general grievances about political inequalities. Individuals are particularly aggrieved by state violence because it violates basic social norms. Second, they also point out that the translation of violence into a political response, which in this case is an increase in ethnic attitudes, is probably influenced by group-level processes. The interpretation of violent events takes place at the level of local communities and builds on previous collective narratives.

Future research should explore in a more detailed way the mechanisms by which violence increases ethno-nationalism, developing a better way to measure both individual-level and group-level exposure to prior violence and exploring how collective narratives influence the attitudinal consequences of violence. These effects can be an important source of political instability in many post-conflict contexts.