Home, Again: Refugee Return and Post-Conflict Violence in Burundi

In the mid-1990s, hundreds of thousands of refugees returned to Bosnia and Herzegovina, the war finally over. Amid their homecoming, a song became popular in Sarajevo: “Sarajevan [people] / While Bosnian cities burned/You were far away/When it was difficult/You left Sarajevo… When you come back one day I shall greet you/But nothing will ever be the same / Don't be sad then, it is nobody's fault/You saved your head, you stayed alive.”¹ The song speaks to a climate of resentment, discrimination, and marginalization prevalent in Bosnian society, just not along the lines one might expect.² After a war largely characterized by ethnic and religious rivalries, the lyrics exemplify how the legacy of forced migration and subsequent refugee return created new group divisions in Bosnia between those who stayed and those who fled.

This kind of volatile animosity following return migration is not a phenomenon exclusive to Bosnia. In fact, conflict between returning and nonmigrant populations after civil war is a nearly ubiquitous issue for societies recovering from such wars. In Iraq, after years of displacement, the government urged refugees and internally displaced persons (IDPs) to come home. These returning families often faced violent backlash from those who stayed behind, causing many to flee again.³ In South Sudan, as tens of thousands of refugees and IDPs returned in anticipation of independence in 2011, tension emerged among individuals who had lived as IDPs in Khartoum, those who had been refugees in East Africa, and those who had stayed in southern Sudan during the war. Hostility between these groups played out in competition for employment, struggles for land, and youth gang violence in urban centers.⁴ In El Salvador, return migration from the United States changed the nature of violence against the state. In the 1980s and early 1990s, civil war was the primary threat to peace. During this time, roughly 2 million Salvadorans fled to the United States.⁵ In the past ten to fifteen years, however, transnational criminal organizations have become the main source of insecurity in the region. Experts have linked this rise in gang activity to the increase in U.S. deportations of Central American migrants to their countries of origin.⁶

The prevalence of insecurity related to return migration is especially puzzling considering that repatriation is often thought to be both the solution and end point to migration crises.⁷ For example, peace agreements have called for the facilitation of return migration as a way to “undo” the negative impact of war, and international observers have used refugee repatriation as an indicator of postwar stability.⁸ Although scholars and practitioners alike have pushed back against the conception of return as the most natural and pragmatic solution to forced migration crises—pointing to the political incentives behind promoting repatriation,⁹ documenting cases in which return migration disrupted local power structures,¹⁰ and demonstrating that many migrants do not have a home into which they can reintegrate—repatriation remains the de facto preferred solution to refugee crises.¹¹

Still, with notable exceptions,¹² the security implications of return migration are undertheorized in political science. This is a stark contrast to the myriad connections scholars have established between out-migration and civil war. For example, research demonstrates that refugee flight can regionalize conflict and open channels to transnational violence;¹³ refugee camps can prolong wars in sending countries by providing safe haven for rebels;¹⁴ diaspora congregating abroad can influence conflict outcomes in their countries of origin by lobbying for, organizing, and financing conflicts at home;¹⁵ and political elites can manipulate population flows into other countries to gain strategic leverage in war.¹⁶ Although studying the consequences of out-migration is important in its own right, they often cannot be fully understood without reference to previous cycles of return and repeat migration.

Moreover, there may be differences between how out-migration affects security in receiving states and how mass refugee return shapes conflict dynamics in countries of origin. Repatriated populations enter a fluid environment in their home states, as those in power (often alongside international interveners) attempt to reform governing institutions in an effort to restructure political and economic competition. Their return may aggravate old rivalries or even alter the nature of social divisions.¹⁷ Without acknowledging how social processes such as mass displacement may have changed underlying community structures, state-building processes may inadvertently, or intentionally, favor returnees or nonmigrants. Such biases, real or perceived, can exacerbate tensions over property rights, access to public goods, or citizenship rights, and potentially create new sources of insecurity. Yet, much of the existing theory on post-conflict state building implicitly assumes that causes of violence do not change over the course of the war, and therefore focuses on how to build institutions to contain these issues.¹⁸ By failing to take into account how migration may affect the operation and legitimacy of government institutions after wartime, institutions designed to build peace are likely to miss—or, worse, exacerbate—new sources of conflict.

To address these lacunae, this article examines the connections between mass refugee return, peace, and security in countries of origin. I offer a theory to explain both the prevalence and character of returnee-nonmigrant divisions in post-conflict societies, arguing that return migration creates new, situational identity cleavages based on where individuals lived during the war. Because forced displacement is one of the most common consequences of civil war, the potential to categorize groups based on collective experiences of staying or fleeing is common across post-conflict settings. I then argue that in each civil war context, these latent divisions become salient and antagonistic when post-conflict institutions (such as property rights, cultural traditions, or language laws) intentionally or unintentionally provide differential dividends to individuals based on their displacement history.

I lay out my theory and empirical evidence in the following five sections. First, I present my theory of return migration and conflict and then outline the methodological approach, case selection, data collection methods, and analysis of positionality and potential bias in the data. Second, I demonstrate how mass refugee return to Burundi after the country's 1993–2005 civil war created a culture of hostility between returnees and nonmigrants that manifested in widespread local-level conflict over land. Third, I analyze how local institutions in Burundi, namely informal and formal land governance, hardened the country's migration-related divisions. Fourth, I demonstrate that because return migration fomented new local conflict, when a new national-level electoral conflict arose in 2015, many returnees were already primed to leave the country and fled (again) to Tanzania. Thus, previous experiences of return migration shaped both the character and timing of renewed population flight in 2015. The final section discusses how understanding the connections between return migration and local conflict is critical for policymakers and humanitarian actors engaged in peacebuilding.

In this section, I offer a theory to explain both the prevalence of return migration-related divisions in post-conflict societies and their variation in form and intensity. I then outline my research design and ethnographic approach.

The theory draws on two key findings from the anthropological and sociological literatures on political violence. First, scholars have demonstrated that experiences of forced migration can alter conceptions of identity and nationalism, change individuals' priorities in the postwar period, and create new social networks both within and across borders.¹⁹ Second, research shows that the experience of violence during civil war—who it is directed at and why—is frequently tied to local-level social dynamics, such as rivalries over land rights, power-brokering, clan competition, and inequality between villages. These local cleavages often differ significantly from the national division viewed as the broader cause of the war, such as religion, ethnicity, or national political grievances.²⁰

Starting from these premises, I offer a two-part theory. First, I argue that return migration after civil war creates new local divisions based on where individuals lived during the war. These group categories may be as simple as “those who stayed in-country” and “those who left,” or they can be further delineated by the type of displacement (internal vs. international),²¹ host-country characteristics (region, political relationship to country of origin, language, etc.), or time period (e.g., era of flight and/or duration of exile). For individuals living abroad, shared experiences of adapting to new environments, combined with the very act of leaving, help create new networks and signals of group likeness. Some characteristics that define these networks are discrete and observable—language, accent, way of dress, religion. Others are more nuanced, based on perceptions of differences in national ideology or patriotism, roles in the prior conflict, access to wealth and education, or deservedness of peace dividends. Because most civil wars result in some form of forced displacement, the opportunity to categorize individuals according to those who stayed and those who left and returned is common across post-conflict contexts.

Second, these displacement-based cleavages become more salient when post-conflict institutions create real or perceived differential outcomes for individuals based on their migration history. Institutions include both formal bodies and regulations, or informal practices at the national, regional, and community levels. Intentional institutional design choices or ambiguities in interpretation and implementation can produce different outcomes for returnees and nonmigrants. For example, national language laws may affect returnees who spent protracted periods of time in host countries with a different predominant language by impeding access to jobs, education, and health care. Informal land inheritance practices may create new sources of conflict when family members return from exile seeking to resettle in their home areas. Interaction with these formal and informal institutions can create an endogenous cycle whereby institutional biases shape and reify migration-related divisions: as individuals begin to understand their position in society as connected to their migration history, they adjust their future political and social behavior accordingly. For instance, if national language laws preclude Arabic-speaking returnees from pursuing jobs in the government, as they did in South Sudan, and narratives exist about returnees being less patriotic than those who stayed and fought for their country, returnees may interpret their inability to access jobs as the nonmigrants deliberately keeping them from power. These return migrants may then change their behavior in other aspects of political, economic, and social life accordingly. Of course, as with other types of identity divisions, elites may manipulate migration-based cleavages to their strategic advantage. Societies where institutions do not provide differential dividends to returnees and nonmigrants, or where policymakers quickly remedy disparities and ambiguities rather than exacerbate them, are less likely to see violent tension between these groups.

Importantly, the theory does not predict that divisions between returnees and nonmigrants will be stagnant. Rather, it allows for change depending on interaction with institutions and elites over time. Tensions between returnees and nonmigrants are therefore best understood on a scale from absence, to peaceful delineation of categories, hostility, and violence. Widespread tensions between returning and nonmigrant populations after civil war can shape future large-scale conflict dynamics, should additional issues arise at the national level. Thus, the actual manifestation of violence during a renewed crisis may occur along migration-related lines, as individuals use the cover of national politics, or ally with elite actors, to follow through on the local returnee-nonmigrant rivalries.

I developed the theory inductively based on observations in South Sudan from 2011 through 2013. I then use an in-depth ethnographic case study of refugee return following Burundi's 1993–2005 civil war to evaluate the theory against evidence in a second case.²² In a case study approach, theory development and theory testing are often intertwined, and this project is no exception.²³ But by developing the primary theoretical constructs in South Sudan first, and then conducting the ethnographic study in a different context, I am able to get better analytical leverage to assess whether the expected dynamics appear outside the conditions in which they were originally developed.

Both the methodology and case selection offer a number of advantages. Process tracing and thick description using ethnographic data are effective tools for assessing evidence against theory-generated expectations and eliminating alternative explanations.²⁴ This type of inquiry is especially suited to documenting and evaluating complex meso- and micro-level dynamics over time such as the intricate interactions between return migration, community identity, institutions, and violence, which are central to my theory.²⁵ Whereas survey work often requires researchers to impose external categorizations to develop survey instruments, by using an ethnographic approach, I am able to gather data on the meaning-making processes themselves and explore how individuals' ideas, beliefs, values, and preferences are embedded in power relationships in their communities without externally framing the narratives at the outset.²⁶ Additionally, semi-structured interviews, repeat interactions, and metadata allow researchers to identify and apprehend why certain political behavior makes sense when considered in context. This metadata is especially important when working with communities that have recently experienced the type of violence that leads to mass forced migration. Individuals in these areas are likely to be skeptical of outsiders, and key informants will likely have participated in or witnessed violence. Analyzing metadata, such as dissimulation or nonverbal communication (e.g., silences, gestures, and tone of voice) can reveal useful information about how social and political dynamics inform what interviewees are willing to say and how they prefer to portray themselves to outsiders.²⁷

Burundi is a particularly useful case for studying refugee return. The country has experienced repeated cycles of forced migration, which allows for the investigation of my theoretical expectations about differentiation among subsets of return migrants who fled to different destinations, at different times, or to different types of host countries. The timing was such that return migration had been completed no more than ten, but no less than two, years from when I began data collection. This timing allowed for migration-related divisions to develop, ebb, or persist, and thus provided the opportunity to explore change over time.

Burundi is also a hard case for my theory. Unlike in South Sudan, migration from Burundi was highly correlated with preexisting ethnic cleavages: the majority of refugees in Tanzania were Hutu, whereas Tutsi were more likely to stay in-country during the war. Therefore, in expectation refugee return would be more likely to exacerbate these preexisting ethnic divisions rather than produce new, cross-cutting, migration-related divisions.

Finally, the 2015 electoral crisis in Burundi provided a unique opportunity to evaluate how the experience of return migration affects individuals' future behavior. Faced with political uncertainty, many Burundians were forced (again) to make a decision whether or not to flee the country. I was thus able to explore whether prior experiences of return affected individuals' future decisionmaking in a particularly high-stakes context.

Having developed the core aspects of this theory in South Sudan, I outlined a set of indicators for evaluating how well the argument held in other environments. A monthlong exploratory trip to Burundi allowed for both preliminary deductive evaluation of the theory and additional inductive refinement. In particular, on this preliminary trip I observed that land governance was the likely intervening institution at play reifying identity categories in Burundi (as opposed to my observations about language laws in South Sudan). I then returned to Burundi (and later Tanzania) to conduct the primary data collection for the ethnographic case study.

Based on my theory, I expected to see new group categorizations in Burundi between returning and nonmigrant populations, potentially cutting across preexisting divisions such as ethnicity. Because there were two distinct waves of out-migration, and markedly different characteristics between host countries, there could be additional delineation of the returnee group according to the era in which they fled or country to which they fled. Further, if these divisions were particularly strong, I expected that competition between these groups would converge around institutions that were perceived to provide different outcomes to migrants and nonmigrants.

If my theory did not fit the Burundian context, respondents would be more likely to focus on general malaises of the community (poverty, health) or to highlight preexisting divisions, such as ethnicity, religion, political party, or other local-level rivalries, as the primary cleavages in their community. Evidence that returnee-nonmigrant labels were used simply to couch references to preexisting divisions, such as ethnicity, would also suggest that migration-related divisions were less important.

I then used the 2015 national electoral crisis to further examine the salience of return-related divisions. Evidence from my two trips to Burundi prior to the crisis established that tension along the return migration-based cleavage centered on land ownership. Therefore, if these divisions were particularly salient, I expected Burundian refugees to cite land conflict between returnees and nonmigrants as among their reasons for renewed flight. I also anticipated that some Burundians would have tried to exit Burundi before the opening of the borders following the 2015 conflict. If the migration-related divisions were less salient, I expected Burundian refugees to cite push factors related to the national-level political conflict (fighting between political parties, targeting for recruitment into armed groups, repression of perceived opposition voices) or general fear of war as their primary reason for fleeing. An absence of migration, or attempted migration, prior to 2015 would also indicate that the local-level issues related to migration were less important compared to national political party divisions.

I gathered data for this project over the course of nine months of research on both sides of the Burundi-Tanzania border from 2014 to 201 6.²⁸ Overall, I conducted 258 semi-structured interviews with Burundian civilians, international humanitarian organization staff, Tanzanian and Burundian government officials, and Tanzanian villagers, in addition to countless hours of field observation.

I conducted the research in three primary areas: (1) villages in Makamba Province in Burundi; (2) the Nyarugusu Refugee camp in Tanzania; and (3) Ilagala village, a small farming town in Tanzania's Kigoma region. Makamba Province received the most returnees from Tanzania to Burundi after the civil war. I selected villages in Makamba to get a diversity of settings and experiences (e.g., rural hillsides, town centers, Peace Villages, and areas with known issues of land conflict or relatively fewer known issues of land conflict.). Nyarugusu was the first camp to receive refugees from Burundi in Tanzania when violence broke out in Burundi in 2015. Finally, I included a non-camp field site in Tanzania to explore the nature of migration after the Burundian civil war, but prior to the renewed conflict in Burundi that began in April 2015, commonly referred to as the “third mandate crisis.” I chose the village of Ilagala because of its reputation as a clandestine destination for Burundians wishing to live in Tanzania, and because it was the site of a recent International Organization for Migration pilot program to register Burundian “irregular migrants.” Through the program, Burundians living illegally in Ilagala who could prove they were refugees from the 1993 civil war were given a type of Tanzanian identity card that allowed them to live legally in the area for two years. Individuals outed as illegal migrants could face imprisonment or deportation, and would be wary of publicly identifying as Burundian or speaking to outsiders. Therefore, by choosing a village that had already experienced a registration program, I was, by my presence, less likely to jeopardize their safety in the community.

I adapted my sampling methods per the demands in each site. In Makamba Province, I used a “random-walk method” to collect semi-structured interviews in the villages, in addition to key-informant and expert interviews. In Nyarugusu, I devised a random-walk procedure, stratified by residential zone. I conducted both randomly selected and nonrandom informant interviews in all the Burundian zones open in camp at the time.²⁹ In Ilagala, I used snowball sampling, rather than random walks, to better protect the safety and anonymity of interviewees.

The primary aspect of my positionality that could have biased the data was that because I am a young, white, American, female, many of my interviewees assumed I was a humanitarian aid worker.³⁰ I took a number of precautions to combat this perception. I hired my own driver most days while in Makamba and Ilagala, rather than riding in a nongovernmental organization (NGO) vehicle. The camp managers in Tanzania, however, required that I enter in an NGO vehicle. I therefore limited my association with the NGOs in other ways: I walked through the camp to conduct interviews rather than traveling in the residential zones by car; I never wore an NGO badge or a shirt with a logo (which all NGO staff were required to do); and I repeatedly emphasized in my informed-consent process that I was not with an NGO or a government agency, and I could not offer respondents any immediate assistance in exchange for speaking with me.³¹

Still, because most of the white women in the area were aid workers, I am conscious in analyzing interview data that participants may still have thought that I worked with an NGO, and therefore framed their stories to better elicit aid. I analyzed interviews in which the interviewees placed themselves and their families as the sole victims with a measure of caution (for example, in a land conflict, the interviewees may also have been perpetrators of violence). Most important for my theory, however, was not culpability in violence or victimhood, but who interviewees identified as the competing sides in conflicts: Did they say they were attacked by other Hutus? By the government? Or, did they frame the conflict as occurring between returnees and nonmigrants? Therefore, even responses intended to establish interviewees' deservedness of humanitarian aid are unlikely to have biased the analysis of the existence and salience of return migration-related situational identities.

Although there have been many instances of forced migration in Burundi since independence, I focus on three primary cycles. First, in 1972 a selective genocide in Burundi, in which the predominantly Tutsi government organized the killing of 200,000–300,000 Hutu civilians, sent at least 217,000 Burundians across the border to refugee camps in Tanzania (known as the “Old Settlements”).³² Policymakers refer to this group of refugees as the “1972 caseload.” Some of these 1972 refugees returned to Burundi in the late 1980s, whereas others remained in Tanzania.

The second mass displacement began in 1993 with the assassination of President Melchior Ndadaye, and spiked again in 1996 through 1997 at the start of the civil war. Again, hundreds of thousands of Burundians fled to Tanzanian refugee camps and villages. Most of these refugees were housed in a different set of refugee camps, known as the “New Settlements,” and the existing camps that housed primarily the 1972 refugees were renamed the “Old Settlements.” The international community and Burundian policymakers refer to this cohort as the “1993 caseload.” Like in 1972, refugees fleeing in the 1990s were primarily Hutu, with the majority congregating in Tanzania. Tutsi civilians were more likely to stay in-country, though many were displaced internally, and some fled to Rwanda or elsewhere in the region. Some of the so-called 1993 caseload had also fled in 1972, spending only a handful of years back in Burundi before fleeing again. The “1993 caseload” label is therefore a misnomer: it more accurately refers to refugees who lived in the New Settlements.

Finally, in 2015, amid a renewed political crisis more than 413,000 Burundians fled to neighboring countries.³³ More than half of these refugees fled to Tanzania, while the other half escaped to Rwanda, the Democratic Republic of Congo, and Uganda.

Importantly, during each of these cycles, government agents and civilians who remained in-country expropriated land left behind by Burundians in exile.³⁴

Starting around 2002, with the civil war drawing to a close, hundreds of thousands of Burundians living abroad returned to Burundi. Some returned by choice, others by force. Tens of thousands of the Burundians living in the New Settlements did not want to go back. Many feared that their land in Burundi had been taken and they would have no place to which to return.³⁵ In delaying their return to Burundi, however, refugees ran the risk of being associated with the splinter rebel group/opposition party, the FNL, which maintained a presence in refugee camps in Tanzania. Despite their common origins as Hutu-nationalist rebel groups formed in exile, the FNL continued fighting after the primary insurgent group (turned political party), the Conseil National Pour la Défense de la Démocratie-Forces pour la Défense de la Démocratie (CNDD-FDD), took power in 2005.³⁶

In 2007, Tanzania announced that it would begin closing the New Settlements housing these refugees. Some of the Burundians in these camps found ways to remain in Tanzania illegally rather than repatriating, often living in small towns such as Ilagala and farming for Tanzanian villagers. In 2012, Tanzania closed down the last of the New Settlements, forcing the remaining 37,000 refugees to return to Burundi through a process called “orderly repatriation.”³⁷

Exact data on how many Burundians fled in which period, where they fled, and if and where they returned do not exist. Estimates suggest, however, that in 1972 at least 217,000 Burundians fled to Tanzania. By 2003, ten years after the 1993 civil war had begun, the United Nations (UN) estimated that a total of 500,000 Burundian refugees were living in official camps in Tanzania from both the 1972 and 1993 out-migrations. The total number of Burundians living in Tanzania was likely much greater, as an estimated 300,000 Burundian refugees lived illegally in the Tanzanian countryside and urban centers.³⁸ The UN estimated that close to 500,000 refugees returned to Burundi from 2002 to 2012.³⁹ Although it is unclear exactly how many from each caseload returned, of the 1972 population still living in the Old Settlements, approximately 160,000 applied for naturalization in Tanzania and 55,000 expressed desire to repatriate. This number, however, likely underestimates the total number of 1972 caseload returnees who may have returned prior to the UN's organized efforts to close the camps, did not live in camps, or returned during the two interwar periods.⁴⁰

After the 1993–2005 civil war, experts worried that the mass return of Hutu refugees from Tanzania to Burundi would provoke ethnic tensions and destabilize the peacebuilding process.⁴¹ Return migration did indeed incite widespread local-level violence in Burundi, but these conflicts were not simply the result of reignited ethnic rivalries. Rather, refugee return also created new cross-cutting divisions in Burundi between returnees and nonmigrants.

In this section, I use ethnographic evidence of group labeling, attribution of group characteristics, and narratives of group competition to demonstrate the existence of displacement-related divisions in postwar Burundi. These divisions were distinct from, though often operated in alliance with, preexisting ethnic cleavages.

GROUP LABELING. By 2008, Makamba Province (Burundi's southernmost district, which shares a border with Tanzania) housed the largest concentration of refugee returnees in the country.⁴² In 2014, several years after the return had completed, villagers in Makamba still identified groups in their community by their previous migration history. Informants described two primary groups: first there were the Abahunguste (Kirundi for “those who came back”), also known by the French term les rapatriés (the repatriates). Second were the Abasangwa (Kirundi for “those who were here and welcomed others”), or les résidents (the residents) in French.⁴³ Within the rapatrié group, there were further subdivisions according to the era that returnees initially fled (1972 or in the 1990s), and in some cases by country of asylum (Tanzania, Congo, and Rwanda). International actors contributed to solidifying the time-related subdivisions, as they treated the two groups differently and held strongly to the narrative that the 1972 returnees needed more aid reintegrating because they had been away for so long, whereas returnees who originally fled in the 1990s did not have as hard of a time coming back, as they had lived outside Burundi for only a short time (just twenty plus years).

Certain names were also interpreted as pejorative. For example, Burundians frequently used the Swahili term “Sabini na mbili,” meaning “72,” to derogatorily refer to returnees as Tanzanian (Swahili is the Tanzanian national language). This term was often used regardless of whether the returnees in question actually fled in 1972. Individuals also called returnees “Tanzanians” or “Congolese,” referencing where they lived during the war. Many returnees, however, felt that these terms were insults insinuating that returnees were less Burundian than their peers who stayed in-country.

ATTRIBUTION OF GROUP CHARACTERISTICS. Respondents also distinguished rapatriés from résidents by certain alleged characteristics. Having lived abroad for decades, some rapatriés spoke only Swahili upon return, as opposed to Kirundi, the national language of Burundi. Others spoke a mix of Kirundi and Swahili, or spoke Kirundi with an alleged Swahili accent. Many Burundians claimed that they could tell if someone was a returnee by sight, claiming that returnees and nonmigrants dressed differently. Returnee women were said to cover their hair in a style different from that of women in Burundi and to carry their babies “like the Tanzanians”—wrapped in a cloth diagonally across their backs rather than horizontally, as is done in Burundi. It was also common to hear people comment that only returnee women rode bicycles, something women in Burundi had never done. Notably, the perception that returnees and nonmigrants looked and sounded different persisted despite evidence to the contrary.

The differentiation between these groups fueled narratives about which group had better claims of national legitimacy. People would use labels such as “the Tanzanians” to differentiate returnees from Tanzania (as opposed to those from the Democratic Republic of Congo), but also to deride returnees' claims of belonging in the homeland. For example, as one returnee said, “I thought when I came to Burundi, I would face many problems due to loss of culture. We don't speak Kirundi. They say people who don't speak Kirundi are not Burundian. They call us not Burundian.”⁴⁴

Importantly, the rapatrié-résident division was not simply another way to talk about prior ethnic relations. The two categories certainly overlapped, as rapatriés were primarily Hutu and résidents were thought to be primarily Tutsi. Therefore, in some cases politicians used these stereotypes of returnees as Hutu and internally displaced persons or résidents as Tutsi to provoke prior ethnic rivalries to their own advantage.⁴⁵ In actuality, although returnees from Tanzania were primarily Hutu, there were Hutu, Tutsi, and Twa résidents—they just had to be in Burundi to “receive” those returning from abroad. Many of the violent conflicts between résidents and rapatriés pitted Hutu against Hutu.⁴⁶ In fact, migration-related divisions not only cut across ethnicity, but frequently divided families where members had lived on either side of the border during the war. As such, migration-related categorizations existed independent of, though sometimes associated with, ethnic categorizations.

Thus, return migration to Burundi had created a new set of group categories based on where individuals were during the war. These delineations by label and perceived visible, auditory, and cultural characteristics are indicative of the creation of new migration-based situational identities. As a returnee from the 1990s put it, “[It is] us, the residents, and the 72s.”⁴⁷

SOCIAL AND POLITICAL TENSION. Although the creation of new identity groups does not automatically entail conflict, many Burundians described segregation in their communities along rapatrié-résident lines or said that they felt discriminated against based on their migration history. A common complaint was that if there was a death in a returnee's family, only other returnees would come to the family's home to mourn with them. For example, one woman explained with much derision that when her husband died, the résidents in her neighborhood went to drink beer instead.⁴⁸ Other informants in Burundi reported that a local administrator would call a meeting but would not inform returnees about the gathering. Some 1993-caseload returnees claimed, though it was not confirmed, that they could not access health care and that the national identification cards for returnees after 2012 were different from those of other citizens. These discriminatory practices, real or perceived, furthered the 1993 returnees' in-group narrative that the government (and other Burundians) treated them as a lesser class of citizens.

For their part, résidents claimed that rapatriés would gather in the market to drink coffee and discuss current events, but would exclude résidents from joining. Many résidents also claimed that the government and international community preferred the returnees when distributing aid. Said one résident, “It's because the government is not fair. They always want to help the people who repatriated and those who remained inside they see as meaningless.”⁴⁹ The trope of government favoritism reflected the fact that the ruling party, the CNDD-FDD, was previously a rebel group formed in exile, and many of its leading members had relatives who died or were forced to flee during the 1972 genocide. The party therefore tended to favor returnees from 1972. On the other hand, out-groups stereotyped the 1993/New Settlement returnees as FNL supporters, given that they did not return to Burundi when the CNDD-FDD took power in the 2005 elections. Many returnees from the 1990s therefore believed that the Burundian government discriminated against them or ignored their needs.⁵⁰

LAND CONFLICT. In addition to these sources of segregation and animosity, tensions between rapatriés and résidents were most evident in conflicts over land. For many Burundians, national political and social identity is entwined with a physical connection to ancestral land.⁵¹ Therefore, returning to the homeland was not simply about crossing the border into Burundi, but also about reclaiming part of their family's ancestral land. Moreover, most Burundians rely on the land—an estimated 90 percent depend on smallholder agriculture for a living.⁵² Therefore, combined with rapid population growth, the mass population return put immense pressure on the country's most desired resource. Because land in the interwar years had been occupied, expropriated, bought, or sold, both rapatriés and résidents often claimed the same land as rightfully theirs. As such, simply by showing up in villages—even if not directly claiming a plot—returnees posed a potential threat to nonmigrants. The threat of losing land bred distrust, conflict, and violence between returning populations and nonmigrants. Conflict between rapatriés and résidents over land manifested in everything from harassment, to destruction of crops, threats of future violence, physical assault, and murder.⁵³

Relations did not have to be this bad. As I demonstrate in the following section, conflicts between rapatriés and résidents became as severe and widespread as they did in part because institutions governing land and property rights provided different dividends to individuals depending on their migration history. By the time I arrived in Burundi in 2014, returnees from 1972 were faring better in land conflicts than both résidents and the 1993 returnees, as the ruling party promoted property restitution for crimes committed during the 1972 genocide, and the international community had prioritized helping 1972 returnees find or recoup land. Returnees from the 1990s were least advantaged in land competition. Denigrated as FNL sympathizes, and subject to the international community-endorsed narrative that they did not need as much assistance in claiming land upon return, 1993 rapatriés had little power to evict résidents from a disputed property.

For their part, résidents claimed that having lived on the land for more than twenty years, and investing time and money into planting crops and building houses on the property, it was rightfully theirs. Résidents often questioned the legitimacy of returnees' claims by asserting that those coming back were not “truly Burundian” or that the returnees, especially those from 1972, were claiming land that was never theirs. (See figure 1 for a summary of these group divisions and in-group/out-group narratives).

The primary institutions contributing to violent conflict between rapatriés and résidents in Burundi were informal and formal land-governance practices. An informal tradition of patrilineal inheritance of land set the stage for competition over ownership. Consequently, conflict over land was frequently between male members of the same family (and ethnicity) who had different migration histories—brothers or uncles who fled during the war and those who stayed behind. Considered alongside disputes between returnees and nonmigrants who were not related, rapatrié-résident divisions within kinship groups show the power of refugee return in delineating new lines of conflict in Burundi.

If informal inheritance practices created a permissive environment for intra-familial rapatrié-résident conflict, the formal institutions governing land and property rights, including a newly created federal land commission called the Commission Nationale des Terres et Autres Biens (CNTB), had an even greater impact. Although the CNTB was not the only formal institution governing land, by 2014 it had become the focal institution in the politics of return and resettlement.⁵⁴ The CNTB was widely known in rural areas, and my interviewees regularly cited the commission by name without prompting. Therefore, to understand the role of institutions in the construction of migration-based situational identities, I focus on the institution most salient in respondents' accounts of returnee-nonmigrant relations.

The peace agreement ending Burundi's civil war, known as the Arusha accords, stipulated that the government encourage the repatriation and reintegration of refugees and create a special commission to adjudicate land disputes arising from return—specifically those from “old caseload” returnees.⁵⁵ Thus came the CNTB. The first iteration of the CNTB was a relatively independent commission, led by a Tutsi clergyman named Father Astère Kana. This early CNTB promoted a policy of sharing land between rapatriés and résidents. Although this was not fully satisfactory to either party, some villagers accepted the practice.

This sharing policy was not universally upheld, however. People could appeal CNTB rulings through the regular court system, which often reversed the CNTB's decisions. In other cases, returnees, especially those from the 1990s who had less political and economic capital, were too scared to take their claims to court for fear of retribution. Résidents, having earned livelihoods from the land for decades, tended to have more resources that allowed them to get by in a stand-off or buy off officials. Moreover, it was generally easier for résidents to maintain occupation than for rapatriés to enforce an eviction. This opacity in implementation precluded the CNTB from assuaging the hostility between the two groups.

When Father Kana died, the CNTB was placed under the presidency, and a party loyalist, Sérapion Bambonanire, was appointed as head of the commission. Under Bambonanire, the CNTB revised its policy to require full restitution of all land and property to returnees, particularly those from 1972. The CNTB justified this policy as a form of transitional justice: the land was wrongly taken as a result of the 1972 genocide; therefore, it was only right that all property be returned. Although true in part, this was an oversimplification of the issue, as occupants on the land may not have illegally appropriated the land themselves, but bought the plots from the government or inherited them indirectly. While the government's line was that résidents had other land to which they could return if forced to leave the disputed property, résidents disagreed, claiming that, having lived for decades in the area, they maintained no other home. Taking matters to the extreme, Bambonanaire's CNTB applied this policy retroactively and reopened some cases where disputants had already resolved to share the land.

The new approach worsened relations between returnees and nonmigrants, exacerbated violence between community members and against the CNTB itself, and reified the groups' separation. As one respondent explained, “Before we used to share the land. Back then we used to live together, could go to neighbor and ask for fire [for cooking]. Now there is no sharing.”⁵⁶

Soon, communities began rebelling against the commission. Wielding machetes, stones, and farming tools, villagers attempted to prevent CNTB vehicles from entering their town to investigate cases and implement rulings. In March 2015, about one month before the third mandate protests in Bujumbura would shake the nation, thousands of citizens in Makamba took to the streets protesting the land commission. The governor of Makamba responded by halting the implementation of CNTB rulings in the province “to avoid a blood bath,” and President Nkurunziza announced a nationwide suspension of CNTB activity.⁵⁷

The economic consequences of restricted access to land were also dire. Because the vast majority of villagers were subsistence farmers, without property they were left in a desperate financial situation. Forced to farm for others and use wages to rent houses, Burundians who were most disadvantaged in accessing land were often unable provide adequate food, clothes, or care for their families.

In addition to land governance, respondents described several other institutions as discriminating against individuals based on their migration history. For example, because many returnees had studied in English and Swahili in Tanzania, they often claimed to be at a disadvantage in the Burundian system, which taught in French and Kirundi. In other cases, individuals assigned this narrative of discrimination without the same concrete evidence of differential treatment. For example, Bonifax⁵⁸—a returnee—had completed some university education while in Tanzania, but struggled to get Burundi to recognize his certificates of program completion so he could pursue employment in Burundi. In this instance, Bonifax's qualifications earned in Tanzania may not have been equivalent to the diploma he sought in Burundi. But according to Bonifax and others, discrimination against returnees was to blame: “I think it's because we are repatriates.”⁵⁹ Bonifax later went on to explain that issues between rapatriés and résidents persist in his community because they were still having land conflicts, and in some cases having to give back land.⁶⁰ Bonifax's use of rapatrié-résident land conflict as the lens to interpret other community interactions (being denied a diploma) illustrates the endogenous cycle proposed in my theory: new local divisions spurred by population return are aggravated by institutions that provide differential dividends based on where individuals lived during wartime such that individuals view future community interactions through the same returnee-nonmigrant frame.

One could argue that resource scarcity, poor land governance, or both are sufficient to explain the emergence of widespread local violence in post-civil war Burundi. Land in Burundi is scarce yet extremely valuable. So, regardless of institutions, one may expect competition for property rights. I agree that resource competition was an essential contributing factor. Resource competition alone, however, is not a sufficient explanation for the character and persistence of returnee-nonmigrant conflict in Burundi, as the context of the land governance regime appears to have played a role. Given the same resource scarcity, returnee-nonmigrant tensions were worse under the second iteration of the CNTB than they were under the initial policy of land-sharing between repatriates and residents.

An alternative land governance-based explanation would suggest that long-standing structural issues concerning land rights in Burundi explain the local violence in Burundi after the civil war rather than return migration.⁶¹ To this end, Mathijs Van Leeuwen points to the fact that return migration did not exacerbate ethnic conflict between Hutu rapatriés and Tutsi résidents. Instead, violence in Burundi after the civil war largely occurred between family members, or it was the result of state expropriation.⁶² Therefore, analyzing the violence through the lens of land conflict, not return migration, sufficiently explains the observed dynamics.⁶³ I agree that land governance is one of the primary factors that contributed to the violence in postwar Burundi. In fact, this is central to my theory. My data suggest, however, that one cannot discount the role of return migration in this process. In my interviews, conflicts between family members were often also characterized as conflicts between returnees and nonmigrants and some interviewees used the language of migration identity (e.g. “the one who stayed”) to describe problems within their family.⁶⁴ Moreover, although the violence between returnees and nonmigrants most frequently centered on property rights, narratives around these group identities also existed outside the realm of land conflict or even economic competition; they were also evident in descriptions of social segregation, claims of legitimate citizenship, and perceived discrimination in education. Limiting scholars' understanding of the rapatrié-résident divide to land conflict would preclude the identification of a wider pattern of conflict between returnees and nonmigrants in Burundi and in other countries where the institutional environment rendered a different issue as the primary source of competition between migrants and nonmigrants.

In 2014, given the violent conflict over land and extreme poverty in Makamba, there was a significant population—especially among the 1993-caseload returnees—who wanted to leave Burundi. Some rapatriés interviewed in Burundi before the third mandate crisis even reported that their neighbors had threatened them by saying that if war came back to Burundi, the rapatriés would be the first killed: “Those who didn't run away [the résidents], they are trying to scare us that if the fighting starts up during the elections that they will come and kill us. But really [I] think this is their way of trying to scare us, chase us off the land.”⁶⁵

At the same time, Burundi was approaching another nationwide conflict. The country was scheduled for a presidential election in 2015, the first in which a president would face the constitutional two-term limit envisioned by the Arusha accords. In April 2015, the ruling CNDD-FDD Party declared that President Nkurunziza would seek a third term in office, or third mandate. This announcement set off mass protests in Bujumbura, an attempted coup, the formation of an armed rebellion, and a government crackdown on anyone perceived to be a member of an opposition party or critical of the regime. The crisis also spurred mass displacement: by May 2015, more than 100,000 Burundians had fled to neighboring countries.⁶⁶ Refugees continued to flee in droves in the coming months.

Observers initially thought that the difference between who stayed and who fled in 2015 reflected the political nature of the conflict—Nkurunziza and his allies against anyone associated with the opposition.⁶⁷ But, in the following section, I argue that prior experiences of return also influenced who stayed, who left, and when they left in 2015.

Refugees' explanations for what forced them to flee in 2015 elucidate the existence of two types of security issues that operated both separately and in combination: the national electoral crisis and local conflicts related to previous return migration. Accordingly, reasons for flight fell into three categories.

Those in the first category (25.6 percent) emphasized land conflict or related issues as the preponderant danger. Many in this category were repeat refugees who previously fled in the 1990s and then returned to Burundi. For example, in an interview with two women who fled early on in the crisis, immediately, and without my prompting, one of the women told me her husband had died in 2013, “after we returned to Burundi by force. Because of land problems.”⁶⁸ As we continued talking, she explained that she was among those Burundians who had fled in 1972, and then again in 1993. She had not wanted to return to Burundi because she knew that her land was occupied, and there would be problems if she went back. She was forced to return, however, when Tanzania closed the New Settlements in 2012. According to her, “The ones who had not fled do not want to see us in the country, because they have taken lands. Whenever they see us, they feel bitter.”⁶⁹

Both women went on to tell me about how, when they returned to Burundi, family members who had stayed in-country during the war were now occupying their land. Those on the land threatened to kill the women and their families if they tried to stay. In the first woman's case, they had already murdered her husband. This is why they had to leave in 2015. When I asked them about the third mandate crisis, the second woman looked at me with an expression completely absent of recognition and said, “I do not understand.” I explained about Peeta (as President Nkurunziza is colloquially known) running for a third term and the subsequent electoral violence, and she replied “[We have] nothing to hear about Peeta.”⁷⁰

These women were exceptional in their absolute rejection of the relevance of the national-level conflict. Most refugees knew about the third mandate crisis. Still, the consequences of their previous return were preponderant in their decisions to flee. As one refugee explained, “We had fled and those who remained on the land said it was theirs. Conflicts followed. [T]hey can kill each other based on land.”⁷¹ Another woman said she left simply because, “We had no lands and nowhere to cultivate.”⁷²

Refugees in the second category (28 percent) highlighted issues related to the national conflict as their primary reason for flight, such as political repression, torture, attempted recruitment into government or opposition forces, and fear of return to full-scale war. For example, a man whose eyes were badly injured after he was tortured by members of the ruling party recounted, “Well first, you see what they've done to me (pointing to his eyes). I was told with the insecurity (meaning the 2015 electoral crisis) they were coming to finish me. At 22.00h that night I made the decision to leave.” He went on to describe how the main problems in Burundi were with the ruling party, which he had been asked to join and had refused.⁷³ Notably, refugees' characterization of the national conflict centered on pro- or anti-Nkurunziza allegiances rather than ethnic dynamics.⁷⁴

Other interviewees in this category responded that they left because they were afraid the protests were a harbinger of full-scale civil war. As one woman stated, “The main reason to come here—I saw that life was bad since I was a child. I saw that what I fled before comes back. So, I thought some members of my family died, even my father, because of war. And even now [it's] starting to be the same as it had [been].”⁷⁵

The majority of refugee respondents fell into the third category (46.4 percent), citing both local-level issues related to their previous return and the national-level political conflict as reasons for fleeing. Some expressed these as distinct issues; for example, they faced land conflict or other adverse consequences of return and, separately, were afraid that the country was about to descend into war. More commonly, however, respondents saw the two as intertwined: given the elevated political chaos, they felt it would be easier for those with whom they were already fighting to act with impunity. For many refugees, the person they had a land conflict with had a network connecting them to the ruling party. For example, their nephew may have been in CNDD-FDD's feared youth militia, the Imbonerakure, and therefore able to use the cover of the ruling party's crackdown to access arms and follow through on existing vendettas in the name of party allegiance. As one man explained to me, “The first reason, our land was taken. [They] tried to kill us … [They] said once this conflict begins (meaning potential renewed war), you would be among those killed, so [they] could take the land permanently.”⁷⁶ This response is consistent with those of rapatriés interviewed in Burundi before April 2015 who worried that the elections would exacerbate existing land conflicts.

Many interviewees further explained that their family members who remained in-country during the war often accused returnees of being with the opposition political party/rebel group, the FNL, as a way to leverage the political climate and scare the returnees off the land. In fact, many current Nyarugusu residents described how those in Burundi would accuse them of being affiliated with the FNL even before the 2015 crisis broke, using the logic that only FNL loyalists would have waited so long to come back after the war was over.

This complicated interaction between political and personal conflict is most evident in the following interviewee's description of the CNTB: “There is this organization called CNTB, they came to [allow] people to be back in their land, but [they] were to be killed in that land. Because they [the CNTB] could say ‘the repatriates tak[e] that land,’ and they let you. But then they [the disputing party] come back and kill you with your family and say they were killing murderers and FNL. You are FNL because of the land they want. So they can kill you.”⁷⁷

The ways in which refugees spoke about these different divisions operating in alliance reflects a common pattern of civil war violence: local actors work with national elites or appropriate national-level discourses—such as religion, class, or political allegiance—for private or supra-local purposes.⁷⁸ In turn, local politics can shape the outlines of contestation at the center. In this case, the local cleavages were clearly linked to prior cycles of forced migration and return from the previous wars. This is evident in how respondents across all three categories delineated between threats coming from “those who didn't run away” and “repatriates” as opposed to (or in conjunction with) “the ruling party.”

Local conflict resulting from return migration after the 1993–2005 civil war not only affected who fled amid the 2015 crisis, but also when they fled.

EARLY FLIGHT. Although the violence at the outset of the crisis centered in the capital city, Bujumbura, the majority of Burundian refugees arriving in Tanzania came from Makamba Province.⁷⁹ As previously discussed, Makamba was one of the regions with the greatest density of returnees after the previous war. The first available reports show that, as of July 10, 2015, 64.5 percent of the refugees in Tanzania were from Makamba. Only 5.6 percent had come from Bujumbura and Bujumbura Rural combined—the area where protests and violence centered at the outbreak of the national conflict.⁸⁰ International organizations staff also confided that the vast majority of the refugees they met arriving in first few months of the influx were women and children, many of whom were repeat migrants—this was their second, third, or even fourth time fleeing Burundi. In my interviews, these “early arrivers” often emphasized issues resulting from their previous return (like land conflict) as one of their primary reasons for flight. Many explained that they had wanted to leave Burundi well before the 2015 crisis, but had not found the means to do so. As one refugee explained, “The first problem was land. The thought of leaving was there before. Those who had tickets went to Uganda [before]. But I missed because I did not have the means.”⁸¹ The electoral crisis in Bujumbura provided the opportunity they were looking for to flee to Tanzania with the expectation that they would be allowed to cross the border and seek UN protection.

Given that so many of the early arriving refugees in camp cited issues that predated the third mandate crisis, it is unsurprising there were also Burundians who fled prior to the 2015 crisis after unsuccessfully returning. Take the case of a young construction worker, James, who returned from the New Settlements to Burundi in 2012. When he and his siblings arrived on their parents' land in Burundi, someone else was occupying it. A dispute ensued, and James's brother was mysteriously killed after destroying a fence their neighbor had constructed to divide the land. Seeing his brother killed, and worrying for his other siblings, in 2013 James decided to leave Burundi again. He and his family tried to flee to Kenya, but wound up settling as irregular migrants in Ilagala, a small Tanzanian town near the Burundi border. James's experience was common among Burundians living in Ilagala. Many were former rapatriés who lived in Burundi for only a year or two before deciding to re-migrate. Like many refugees in Nyarugusu, land and/or family conflicts upon return were among the primary reasons that new arrivals in Ilagala fled Burundi again—and why they were desperate not to return.

LATER FLIGHT. Refugees fleeing to Tanzania from approximately mid to late July 2015 onward often highlighted political persecution as their primary reason for fleeing. These refugees were also more likely to be first time migrants (former résidents), and originate from a more diverse set of locations in Burundi: the percentage of refugees arriving in Tanzania from Makamba fell from 65 percent in July 2015 to 46 percent in January 2016.⁸² In addition, later arrivals were more likely to be men who had stayed on land they maintained or reclaimed after the previous war until they felt it was too risky (vis-à-vis the national conflict) to stay.

In my interviews in Nyarugusu, it was not until I reached the zones housing a greater proportion of refugees who arrived later in the crisis that I saw outward signs of heightened political engagement indicative of flight based on the threats posed by party politics. It was there where I met a refugee who proudly showed me his FNL (opposition party) flag, which he had gone to great lengths to keep with him throughout his journey, and where I saw young men playing a local board game that was labeled with four teams—three were Champions League football teams and the other was FNL.

This pattern in the timing of flight was also evident in the different reports of security issues refugees would bring to international NGO staff responsible for protection in the Tanzanian camps. Nyarugusu was the first camp to receive refugees in Tanzania, but as the refugee crisis continued, the international community opened two new camps, Nduta and Mtendeli. Nduta tended to house later arrivals in Tanzania. Mtendeli tended to house earlier arrivals who were directly transferred from Nyarugusu to relieve overcrowding. NGO and UN staff reported that refugees in Nyarugusu and Mtendeli would complain about different security risks than those in Nduta: In Nduta, refugees reported that their safety in camp was at risk because of potential cross-border forced recruitment or infiltration of Imbonerakure into Tanzania— in other words, political party–related conflict. In Nyarugusu and Mtendeli camps, NGO staff received safety complaints from refugees citing land conflict back in Burundi that had emerged during their previous repatriation. These refugees feared that family members or neighbors would send someone to come and hunt them to prevent them from returning to Burundi and reclaiming land (and there was at least one such attempt in Nyarugusu while I was there).

As discussed in the previous section, Burundians who fled later on during the crisis were often first time migrants (former résidents), those who recovered land during the return period, or both; they stayed as long as they felt it was still safe in order to protect those assets. Similarly, those who stayed in-country tended to be résidents or 1972 rapatriés who were less adversely affected by prior return migration in their community.⁸³

PERCEPTIONS OF SECURITY. On the Burundian side of the border, state officials in Makamba and villagers alike asserted that there were no security issues in area. These claims of relative safety reflect the direct inverse of the early-leaving refugees' reasons for flight: those who fled early felt their safety was threatened due to land conflict, and those who stayed (or waited to leave) claimed that these issues did not threaten their safety. However, whereas those who fled later on in the conflict admitted the national security issues had finally reached a breaking point, those still in Burundi as of February 2016 denied that political unrest was creating insecurity. Indeed, respondents in Burundi explained away the refugee flight by claiming that these Burundians fled only rumors of war, not real violence. Others derided refugees for leaving by saying they only left to seek handouts from the UN, or because they thought they would be resettled in the United States. Many respondents I interviewed in Burundi claimed that although there had been issues between rapatriés and résidents over land before, this had not been an issue for some time.

Despite respondents' claims that there was nothing to fear in Burundi, the fact that the governor now traveled in an extended motorcade armed with rocket launchers would suggest that the security situation was more precarious than these interviewees, including the governor himself, cared to admit. Analyzing why some respondents had incentives to lie about the security situation provides important evidence about the political atmosphere in Burundi. Where the refugees in Tanzania pointed to violent conflict over land with their neighbors as forcing them to flee, if you felt safe in Burundi in 2016, it was quite possible that you had been on the winning side of a violent land conflict or had threatened to use violence in such a conflict.

In addition to these incentives to de-emphasize their own role in creating insecurity, interviewees may have avoided characterizing the situation as insecure because they were scared of retaliation from the ruling party. The state security apparatus was deeply embedded, and interviewees may have worried that their answers in an interview would leak to party officials. Indeed, respondents' frequent attempts to discredit the refugees mirrored the CNDD-FDD's rhetoric that the Burundians fleeing were overly fearful, unfaithful to Burundi, or opposition supporters.

This evidence suggests that individuals who stayed in Burundi had incentives to dissemble about the real sources of violence in their community—both local and national. Although it is more difficult to parse out for each respondent whether their interests in covering up the insecurity resulted from their participation in local land conflict or from their alliances with and/or fear of the ruling party, triangulating this evidence with interviewees' economic status provides further indication that local-level conflict and access to land are at least part of the explanation.

ECONOMIC INCENTIVES. Many of the Burundian villagers I interviewed who were still in-country as of February 2016 had economic assets to protect, stating that they would wait until it was absolutely necessary to leave if they had land or a job. For example, some of the villagers I spoke to owned or had access to land; others had a business, job, or other source of income that allowed them to get by. These interviewees' decision to stay makes sense if one considers both the local and national sources of insecurity pushing people to leave. Those who were on the winning side of a local-level land conflict, and therefore had economic assets to protect, would be more likely to risk staying behind to protect those assets until they felt that the insecurity posed by national crisis outweighed the benefit of protecting the land they had recovered or maintained during the previous period of return migration.

The mass return of refugees to their countries of origin is often thought of as a sign of increased peace and stability. Return migration, however, is actually a frequent source of insecurity in countries recovering from civil war. Processes of out-migration and return can aggravate old rivalries and create new divisions between populations who were displaced across borders and those who remained in-country. New migration-related group identities are more likely to harden and become violent when post-conflict institutions intentionally or unintentionally favor individuals based on where they were physically located during wartime.

These dynamics were clearly at play in postwar Burundi. In a country plagued by ethnic and political divisions, refugee return to Burundi after the 1993–2005 civil war created new community divisions between so called rapatriés and résidents. Exacerbated by land governance policies, competition along return migration-related cleavages led to widespread violent conflict over land. Rivalries between returnees and nonmigrants were a powder keg ready to explode if the opportunity presented itself. That opportunity came in the form of the 2015 third mandate crisis. Thousands of Burundians fled the government crackdown on anyone perceived to be in an opposition political party or critical of the regime. Who left and who stayed did not, however, reflect a simple distinction between ruling-party and opposition supporters. Rather, the electoral crisis activated divisions at both the local and national level. Rapatriés who had not yet been able to recover land worried that the national conflict would allow local actors to carry out personal vendettas with impunity, and were therefore among the first to flee to Tanzania. As the conflict took hold, later arrivals had more direct connections to opposition groups or experience of being targeted by the CNDD-FDD.

This pattern accords with existing theories that explain the origins of violence during civil war as the result of alliances between local and national divisions, but also highlights a crucial dimension missing from those explanations. Seemingly idiosyncratic local conflicts may have a common source: return migration. Because forced migration is one of the most common sequelae of civil war, the process of return is likely to spark local hostilities across a variety of post-conflict contexts. The differing manifestations are related to institutional conditions in the country of origin.

Understanding the process through which return migration leads to violence is critical for policymakers engaged in humanitarian responses to refugee crises. In cases of protracted forced migration, plans to orchestrate voluntary return to countries of origin must be treated as a potential source of new conflict. Simply (re)entering home areas creates the opportunity to differentiate groups based on their migration history. In cases of repeat migration, this may mean that individuals who experienced issues upon their previous return because they were seen as returnees will be unwilling to endure that process again, even if national peace processes succeeded. Instead, these refugees may seek to stay in host countries or resettle abroad.

More broadly, the implication of institutions in the emergence of displacement-based identities has important consequences for how scholars and policymakers think about successful peacebuilding. Creating strong institutions is frequently cast as one of the strongest tools in the peacebuilding arsenal, especially in cases of ethnic conflict. Yet, the case of Burundi demonstrates that although institutional reform may help to mitigate against preexisting tension, it can also create venues to reify social divisions and intentionally or unintentionally fuel new conflicts in postwar environments. Peacebuilding interventions must therefore balance the need to address the distinct needs of different groups in the population without institutionalizing new migration-related divisions.

The author extends her deepest thanks to Séverine Autesserre, Jonathan Blake, Kate Cronin-Furman, Page Fortna, Tonya Putnam, Jack Snyder, the anonymous reviewers, and participants at the Columbia University International Politics Seminar for their feedback over the course of this project. Geoffrey Basesa and Achel Niyinkunda were invaluable interpreters and interlocutors in Burundi and Tanzania. The author also thanks Beatrice Edmonds and Zeina Laban for their research assistance. The research for this project was made possible by generous grants from the National Science Foundation, the U.S. Institute of Peace, the Columbia Global Policy Initiative, and Columbia University's Department of Political Science.