Political Institutions and Perceived Political Representation Before, During, and After Identity-based Conflict: Comparing Views from Rwandan and Burundian Citizens

Abstract

We study who perceives gains and losses in political representation in Rwanda and Burundi and why. We do so in the run-up to and during violence, but also in its aftermath characterized by radically different institutional approaches to manage a similar ethnic divide in both countries. We rely on quantitative and qualitative analyses of over 700 coded life histories covering the period 1985–2015. We find convergence in perceived political representation across ethnic groups in Rwanda, but divergence in Burundi, and argue how this relates to the postwar institutional remaking, legitimization strategies, and their impact on descriptive and substantive representation.

Résumé

Nous examinons qui perçoit les gains, les pertes et pourquoi en matière de représentation politique au Rwanda et au Burundi. Notre étude porte sur les périodes qui ont précédé la présente violence, mais également au lendemain de celle-ci, caractérisées par des approches institutionnelles radicalement différentes pour encadrer une fracture ethnique similaire dans les deux pays. Nous nous appuyons sur des analyses quantitatives et qualitatives de plus de 700 histoires de vie codées couvrant la période 1985-2015. Nous constatons une convergence dans la représentation politique perçue entre les groupes ethniques au Rwanda, cependant, une divergence au Burundi, et nous discutons comment cela est lié à la refonte institutionnelle d’après-guerre, aux stratégies de légitimation et à leur impact sur la représentation descriptive et substantielle.

Resumo

Neste artigo, estudamos a perceção, no Ruanda e no Burundi, dos ganhos e das perdas na representação política, e os motivos dessa perceção. O estudo incide sobre o período crescente das tensões até à violência e durante a própria violência, mas também sobre os períodos de rescaldo, caracterizados por abordagens institucionais radicalmente diferentes para gerir conflitos éticos semelhantes em ambos os países. Tomamos por base análises quantitativas e qualitativas de mais de 700 histórias de vida codificadas, abrangendo o período entre 1985 e 2015. Concluímos que há uma convergência na representação política percebida em todos os grupos étnicos no Ruanda, mas que no Burundi há divergência, facto que, em nosso entender, se relaciona com a reconstrução institucional do pós-guerra, com as estratégias de legitimação e o seu impacto na representação descritiva e substantiva.

Introduction

Africa accounts for a predominant proportion of post-Cold War conflicts. Politicization of ethnicity is one of the main drivers of these conflicts, thriving on a history of political exclusion and associated horizontal inequalities that fuel an “all or nothing” competition for power (Esteban, Mayoral, and Ray 2012). The lack of (perceived) political representation lies therefore at the origin of many so-called identity-based conflicts and should be addressed to pave the way to sustainable peace. Consequently, postwar initiatives to address ethnic conflicts revolve around institutional remaking, with power-sharing and integrative state institutions being the most popular approaches (Simonsen 2005).

Yet, the conceptualization and operationalization of these institutional solutions are generally elite-centered, and we have limited understanding of how this institutional remaking is perceived by ordinary citizens (see Blais, Singh, and Dumitrescu 2014). From the micro-level conflict literature, we know that war and mass categorical violence polarize social identities, reshape prosocial behaviors, and affect political participation of citizens (Guariso, Ingelaere, and Verpoorten 2018; Ingelaere and Verpoorten 2020). There is a need to extend this micro-level research to the post-conflict period, to understand how institutional reforms influence individual perceptions, attitudes, and behaviors, thereby affecting the prospect for peace. We contribute to this goal by studying perceived political representation by different ethnic groups in Burundi and Rwanda before, during, and after identity-based violence. Concretely, we ask who perceives gains and losses in political representation, and how these perceptions relate to the postwar remaking of institutions.

Rwanda and Burundi are often referred to as “false twins”: they have some clear similarities, but also striking differences. Their shared characteristics include their ethnic composition (Hutu, Tutsi, and Twa ethnic groups), a relatively long tradition of statehood, the political salience of ethnicity, and a long history of political and ethnic violence (Curtis 2015; Uvin 1999). As they emerged from conflicts in the 1990s, former rebel groups in both countries conquered the political power and became hegemonic parties. However, their paths to power and the policies they adopted to accommodate the ethnic divide differed fundamentally (Vandeginste 2014). Burundi has adopted consociational power-sharing, a political model that recognizes ethnic fragmentation and provides for a joint decision-making polity to ensure the effective representation of majorities and minorities (McCulloch 2020, 85). Rwanda has, instead, opted for an integrationist approach that promotes Rwandan citizenship, de-emphasizes the political salience of ethnicity, and criminalizes reference to ethnicity (Blouin and Mukand 2019). Given their similarities, yet different postwar institutional remaking, a comparative study is well suited to study the impact of such remaking on key political outcomes (here on perceived political representation).

To analyze changes over time in perceived political representation (denoted as “PPR” hereinafter), we combine quantitative and narrative dimensions of life histories of over 700 ordinary citizens. For both countries, the stories cover the period from before the start of war and mass violence in the 1990s until up to fifteen years after the formal end of the violent conflicts. To allow for a quantitative analysis, we structured the life history interviews by a ranking exercise in which the respondents systematically ranked their PPR on a scale of −5 to +5 for every year in their adult life, providing us with 22,546 observations of self-reported rankings of political representation. The narrative dimension includes reasons for changes in these self-reported rankings. To systematically analyze the narratives and understand the drivers of changes in PPR, we coded them according to the four dimensions theorized by Hanna Pitkin (1967): descriptive representation, substantive representation, formalistic representation, and symbolic representation.

Our findings point to common trends as well as differences across Rwanda and Burundi. The clearest similarity is that, prior to violence, the ethnic group deprived of power (Tutsi in Rwanda and Hutu in Burundi) has a relatively low PPR, and this PPR reverses across ethnic groups in both countries, reflecting the postwar ethnic power shifts, and thus indicating the sensitivity of PPR to descriptive and symbolic representation, or the “standing for” dimension. The most striking difference is the postwar convergence in PPR of Hutu and Tutsi in Rwanda over time, compared to the inter-ethnic divergence in Burundi. Our analysis reveals that the convergence in Rwanda is largely driven by substantive representation—the “acting for” dimension, which is most often mentioned by Rwandan Hutu, who are relatively deprived of descriptive political representation. This empirically supports the argument made by Chemouni (2016) and Mann and Berry (2016) that the minority Tutsi regime in Rwanda seeks its legitimation partly by adopting policies that appeal to the Hutu majority. In contrast, the recovery of PPR in Burundi does not continue after the 2005 election, and instead stagnates at a relatively low level. The stagnation is especially pronounced for Burundian Tutsi, who lost power throughout the decades under study, but is also the case for Burundian Hutu, who gained in terms of descriptive representation, but without these gains being sustained by substantive representation. Our results suggest that postwar institutional remaking should look beyond its usual focus on formalistic and descriptive representation, and take substantive representation equally seriously, that is, put in place sound strategies and mechanisms that will ensure that state institutions deliver on key governance outcomes.

Political representation and legitimacy

Political representation and political legitimacy are two of the most important concepts in political studies. Political representation refers to “mental perceptions, conceptions, and processes whereby a representative or a group of representatives stands for a represented person” (Daloz 2017, 8; authors’ translation from French). In her landmark contribution, Pitkin (1967) distinguished four types of political representation: formalistic, symbolic, substantive, and symbolic representation. The formalistic form of representation relates to the formal and institutional arrangements, while symbolic representation is understood as the “power to evoke feelings or attitudes” (Pitkin 1967, 97). Substantive representation is about congruence between policy outputs and policy preferences by citizens, and descriptive representation concerns “the extent to which a representative or legislative body resembles a given constituent and her social or demographic identities” (Hayes and Hibbing 2017, 33). The four forms of representation cannot always be univocally distinguished from one another, and often work in tandem (Hayes and Hibbing 2017).

Political legitimacy is considered as “the soil that nourishes the seeds of civic membership” resulting in “shared subscription to a single authoritative state governing a defined territory, and the sense of membership, fellow-feeling and loyalty to a national community” (Weatherford 1991, 253). As such, legitimacy appears as a property or perception (Suddaby, Bitektine, and Haack 2017). Legitimacy as a property speaks to normative considerations, that is, “when politics is legitimate” (see Netelenbos 2016, 5), while legitimacy as a perception alludes to an empirical perspective which considers political legitimacy as a social construct that should be analyzed in its historical context (Netelenbos 2016). Researchers make the distinction between input legitimacy and output legitimacy (Taylor 2019). According to Caby and Frehen (2021, 226), input legitimacy “results from policy decisions based on citizens’ preferences, and output legitimacy derives from the achievement of policy goals in line with citizens’ interest.”

Political representation and legitimacy are two “closely related theoretical constructs” (see Arnesen and Peters 2018; Hayes and Hibbing 2017; Weatherford 1991, 252). They both refer to proximity between rulers and citizens (Daloz 2009). They attend to the questions as to who representatives are, how they become representatives, and what outcomes they bring about (Arnesen and Peters 2018). In addition, the two concepts overlap. In fact, descriptive representation is associated with input legitimacy (see Snagovsky et al. 2020). In the same token, normative procedures to designate representatives (formalistic representation) relate to input legitimacy (see Schmidt 2013), while the real-life policy outcomes for the people (substantive representation) rather speak to output legitimacy (Arnesen and Peters 2018).

A theoretical conjecture emerging from the literature is that descriptive representation has a communicative role (Stauffer, 2021). It broadcasts the political system’s intentions and willingness to attend to represented groups’ needs and perspectives, and gives a clue as to who is (not) entitled to participating in politics (Mansbridge 1999). Descriptively representative state institutions therefore increase trust in and legitimacy of state institutions, rulers, and processes (Stauffer 2021; Mansbridge 1999). Moreover, descriptive representation, particularly in divided societies, most likely shapes substantive representation (Ejdemyr, Kramon, and Robinson 2018; Theisen, Strand, and Østby 2020). In these societies, clientelism and patronage mediate service provision, as rulers tend to channel scarce resources to their co-ethnics (Burgess et al. 2015; Kramon and Posner 2016). As such, perceptions and experiences of output legitimacy and substantive representation would differ across identity groups depending on whether a group is represented in ruling institutions or not.

Following these lines of reasoning, political settlements have significant influence on perceptions of representation (McAllister 2005; Schmitt and Wessels 2005), and hence on legitimacy. In particular, political systems based on the proportional representation (PR) model arguably lead to greater feelings of political representation (Blais, Singh, and Dumitrescu 2014). The underlying mechanisms rest on two factors. First, in PR systems, because citizens have a range of political options to choose from (Lijphart 1994), the great majority of citizens will recognize themselves in at least one of the parties in state institutions. The second mechanism hinges on the idea that the diversity of ruling elites implies a multiplicity of policy preferences, thus increasing the possibility to have one’s preferences reflected in policy. Accordingly, we would expect Burundians to report superior perceptions of political representation and input legitimacy because the country has opted for consociational power sharing, a political system premised on proportional representation. However, this type of institutional design is also considered detrimental to good governance. It reinforces neopatrimonialism and ethno-clientelism (Kendhammer 2015; Piacentini 2019). Power sharing, particularly in the executive, is arguably vulnerable to corruption (Bogaards 2024; Haass and Ottmann 2017; Kendhammer 2015), because political elites in such system have a short time horizon (Haass and Ottmann 2017). This is because postconflict power-sharing institutions are generally provisional, having a “sunset clause” (McCulloch 2017). Because of the vulnerability to corruption, few resources may remain for nonexcludable services. Hence, we would expect Burundi in this regard to score lower than Rwanda on substantive representation and output legitimacy. Inversely, the Rwandan political system, which is not premised on proportional representation, would expectedly score lower on perceived political representation and input legitimacy.

Political settlements and postwar socio-economic development in Burundi and Rwanda

In both Burundi and Rwanda, while ethnicity is not the sole important identity marker, it remains the most politically and historically salient identity. Through indirect rule, the Belgian colonial power favored the Tutsi minority at the expense of the Hutu majority, estimated to count 85 percent of the population, according to outdated colonial surveys (Uvin 1999, 253). In doing so, ethnic identities became institutionalized and rigidified, also by the issuing of ethnic identity cards. When they became separate and independent states, the two countries followed different political paths. In Burundi, the Tutsi minority remained in power after independence and depoliticized ethnicity as a legitimizing strategy, while in Rwanda, Hutu elites took over and politicized ethnicity by institutionalizing ethnic quotas in state institutions (Uvin 1999). In both countries, the decades following independence were marked by regular outbreaks of ethnic violence, which exacerbated ethnic tensions and fueled the number of refugees in the diaspora. ¹ The 1990s would turn out to be the most dramatic for both countries. It started with a disruption of the political landscape, when the conjunction of internal and external factors resulted in the introduction of multi-party politics (Makinda 1996).

In Rwanda, the establishment of new political parties in the early 1990s exposed the vulnerability of the ruling regime (Andersen 2000). Around the same time, the Rwandan Patriotic Front (RPF), mainly composed of descendants of Tutsi refugees living in Uganda, attacked. The democratization wave and external military threat forced the Rwandan regime to the negotiating table, but ultimately violence resumed and the political elites played the ethnic card. Considered accomplices to the RPF, all Tutsi inside Rwanda were branded enemies of the state through intensive government propaganda. A genocide against the Tutsi followed the downing of the presidential plane on April 6, 1994. In barely three months, an estimated 600,000 Tutsi were killed, as well as many Hutu perceived as “moderates.” Many more Hutu died in the years afterwards, either in reprisal killings, from disease in refugee camps, or from dire wartime conditions (Verpoorten 2020). The genocide and civil war ended with a victorious RPF, which put in place a unity government joint with opposition parties to the defeated regime. Two Hutu individuals led the first postgenocide government respectively as president and prime minister. In 2000, however, Hutu president Bizimungu was forced to resign, and Paul Kagame, a Tutsi born and raised in exile, officially rose to power. In May 2003, Rwanda adopted a new constitution. Three months later, Kagame was elected president. From then on, Rwanda has seen a clear and “silent” reversal of ethnic political power relations. Tutsi have increased their descriptive political representation from below 10 percent to roughly 60 percent of ministerial positions (De Roeck et al. 2016). To deal with ethnicity, postgenocide Rwandan elites conceal ethnicity and promote the idea of the union of all Rwandans (Longman and Rutagengwa 2004). The political settlement branded as “consensual politics” (Beswick 2011; Stroh 2010) rests on two main tenets: discouraging (and even criminalizing) appeal to ethnicity in politics and ensuring that no single political party can claim more than 60 percent positions in government. This strategy arguably serves to fake inclusiveness and “impose a one party’s ideology” (Reyntjens 2006, 1107). Political parties and individuals that challenge the RPF narrative and political hegemony are not allowed to compete and participate in state institutions (Niesen 2013) or have (been) “co-opted, discredited, exiled, or withdrawn” (Beswick 2011, 497).

In Burundi, the postindependence one-party system dominated by the Tutsi minority also came under pressure in the 1990s. It responded by a “government of national unity” that organized democratic elections in 1993, in which Melchior Ndadaye was elected as the first Hutu president. The electoral campaign had however revived ethnic demons (Reyntjens 1993), which played against the former single Tutsi-dominated Uprona party. The Tutsi minority, who made up the majority of security institutions, felt great apprehension over the unprecedented ethnic shift of power. Soon after the election, in October 1993, Tutsi army officers staged a coup, in which the president and his close collaborators, including his constitutional successors, were killed. The consequences in the short run were massive reprisal killings of Tutsi civilians and, in the longer run, the gradual emergence of Hutu rebellions. Over the years, the Hutu-dominated National Council for the Defence of Democracy/Forces for the Defence of Democracy (CNDD-FDD) became the most important rebel group in Burundi. It ultimately signed a ceasefire in 2003 after several years of peace talks. The politico-military process led to the adoption of a new constitution in 2005 where quotas assured the sharing of power at different levels (De Roeck et al. 2016; Vandeginste 2014). A 60/40 percent division among Hutu and Tutsi was introduced for ministerial positions and at the level of the National Assembly. CNDD-FDD registered as a political party and won the 2005 elections with Hutu Pierre Nkurunziza elected as president. This made Burundi one of the few success stories of transition from war to democracy (Lemarchand 2007). The initial optimism soon vanished, however. The 2010 elections, boycotted by most of the opposition parties, were marked by important security incidents and massive human rights violations. An even more important blow to the transition from bullets to ballots was the 2015 political crisis following Nkurunziza’s decision to run for a third unconstitutional mandate. This decision sparked violence and refugee streams. Despite the ethnic power-sharing adoption and implementation, the CNDD-FDD became hegemonic. Almost all the Tutsi representatives and Hutu in institutions are CNDD-FDD members (Ndayiragije and Raffoul 2024).

In sum, the 1990s in Rwanda implied a clear although “silent” reversal of ethnic political power relations and an increase of descriptive political representation by Tutsi from below 10 percent to roughly 60 percent of ministerial positions. In Burundi, the 1990s also meant a reversal of descriptive political representation across ethnicities, but one engrained in the constitution and here it was the Hutu majority who gained descriptive political representation, raising their representation among executive positions from around 30 percent to 60 percent (De Roeck et al. 2016).

Next to very different political turns, postwar Rwanda and Burundi traveled different socioeconomic paths. Rwanda has recorded socioeconomic progress, with an average annual growth rate of 7.4 percent in the period 2000–22, compared to only 2.4 percent in Burundi. As a result, while both countries had similar GDP per capita levels in the early 1990s, the numbers have now diverged, between 2,237 per capita GDP in Rwanda and 714 in Burundi (PPP, constant 2017 international $). Rwanda also made impressive progress in education and health outcomes to the extent that the country became a top achiever of the UN Millennium Development Goals (McKay and Verpoorten 2016). Some argue that the Rwandan regime is “forced” to gain output legitimacy because, controlled by a minority Tutsi elite, it has a deficit of input legitimacy (Chemouni 2016). Hence, to gain both domestic and international legitimacy, the Tutsi minority elite not only emphasizes that it succeeded in ending the genocide but also plays off its socioeconomic performances. One could also argue that the RPF—having come to power with an outright victory and thus monopoly over power—had the leeway to implement its social and economic policies without hindrance, a considerable advantage that cannot be expected in contexts of “chaotic” grand coalition governments (Chemouni 2018). ²

While the RPF regime was needing and able to deliver on matters relating to substantive representation, the same cannot be said for Burundi (Chemouni 2016). Burundi’s postwar economy has stalled. In terms of human development, the country fell short of meeting the Millennium Development Goals (Ndoricimpa 2014). Chemouni (2016) has contended that CNDD-FDD leaders in Burundi came to power with an important reservoir of input legitimacy as a grassroot movement largely supported by the Hutu population. This exonerated the regime from the obligation to deliver on output legitimacy. Over time, the context of state underperformance has arguably shaped the Burundian citizens’ expectations towards their government. Burundians ended up lowering their expectations and evaluating the government based on input legitimacy considerations rather than output considerations (Stel and Ndayiragije 2014).

Data and methods

Data collection

Whereas life history researchers usually collect a few stories, we have stories of 302 individuals in Burundi and 412 in Rwanda. The respondents were part of a sample stratified geographically across rural communities. Seven communities were chosen in Rwanda and six in Burundi (see Figure 1). The choice was guided by the principle of maximum variation, aiming at a large variance in conflict episodes and postconflict experiences across locations. To select individuals, we listed all households in each researched community with the names of household heads and their ethnic markers. Subsequently, we selected households through a random sampling scheme, stratified by ethnic subcategories. Then, we interviewed the household head or another adult member. ³

Figure 1. The location of research sites: (a) Burundi and (b) Rwanda. Note: The locations correspond to small administrative sectors. The names and delimitations correspond to the provinces prior to the administrative reforms of the 2000s, which were still the reference points for interviewees at the time of the data collection.

Source: Authors’ compilation in ArcGIS.

Since Rwanda does not allow reference to ethnicity (contrary to the Burundian situation at the time of fieldwork), the stratification across ethnic subcategories was based on alternative markers that underlie these categories and that are commonly used by Rwandans. The subcategories related to Tutsi include “genocide survivor” and “returnee” ⁴ and those related to Hutu include “not accused in gacaca,” ⁵ “accused in gacaca,” and “liberated prisoner” (see also Ingelaere and Verpoorten 2020). The use of proxy terms to capture individual ethnicity is an effective tried-and-tested strategy to circumvent challenges to research Hutu and Tutsi experiences and perceptions in postgenocide Rwanda (see, for example, Blouin and Mukand 2019; Ingelaere and Paviotti 2023). ⁶ Table 1 provides an overview of the sample by ethnic group. It shows that the share of Tutsi in our sample is close to 30 percent in Rwanda as well as Burundi, about twice their estimated population share, which is the intended consequence of the stratified sampling. ⁷

Table 1. Sample observations by ethnicity, and across interview rounds.

Note: Attrition indicates the share of Round 1 respondents that could not be traced in Round 2. Most of Round 2 interviews with Rwandan respondents took place in 2011, but 38 were conducted in 2015 with respondents who could not be interviewed in 2011, mainly because they were in prison or in re-education camps.

To allow for a quantitative analysis, the stories were structured by a ranking exercise. The respondents systematically commented on political representation and ranked it on a scale of −5 to +5 for every year in their life history. Figure 2 shows the “ladder” used for the ranking exercise. The respondents situated themselves on the ladder, through time, starting with the year of the interview, by answering the following question: “Currently, on what step [on the ladder] do you situate your experience of political representation?” Subsequently a move back in time was made to the year of marriage or the first year of adult life (if single), repeating the question for that point. The same question was then asked with reference to the past, asking a rating for every year.

Figure 2. Ladder, on which respondents indicated their level of perceived political representation.

Source: From authors’ interview guide.

The ranking exercise was developed in a pilot phase that included 50 full life story interviews with 30 Hutu and 20 Tutsi respondents, each lasting between 7 and 14 hours, spread over several sessions. These interviews were conducted through open-ended questions touching on almost every aspect of the interviewee’s life. The subsequent structuring and focus on a limited number of salient themes (including political representation ⁸) reduced the interview time to 1.5–3.5 hours. However, also in their semi-structured form, the interviews gave the respondent room for “telling” their life history. In the pilot phase, we also explored how to best enquire about “political representation.” This enquiry revealed a shared understanding of “guhagararirwa” in Rwanda and “guserukirwa” in Burundi as “political representation,” which literally mean to act in one’s name, but take on a broader meaning when also taking into account their figurative meaning.

For the years since 2000, respondents explained in their own words what the reasons of change were in case they changed the ranking of political representation from one life history year to the other. In total, this resulted in 553 “narratives of change” related to political representation in Burundi and 729 in Rwanda. To analyze the narratives, we coded them systematically, which is explained next.

Coding

Through discourse analysis, we coded the narratives based on four different dimensions of political representation, inspired by Pitkin’s (1967) theorization: formalistic, descriptive, symbolic, and substantive representation. Table 2 presents our coding guide and an example of each dimension. If, in a single statement, we found more than one dimension, we used more than one code but started with the one that seemed most salient. To ensure that we had a common understanding of the narratives, the first two authors on this paper both coded all narratives—based on the understanding of the concept of political representation by different authors such as Daloz (2017), Mansbridge (1999), and Pitkin (1967). They then compared their coding, and discussed the reasoning behind each other’s coding to reach a consensus.

Table 2. Coding guide for narratives related to changes in perceived political representation.

Source: Authors’ synthesis based on Daloz (2017), Mansbridge (1999), and Pitkin (1967).

Biases

There are three potential threats to the validity of our data: recall bias where recalling past events or perceptions from memory results in bias (e.g. remembering positive experiences more vividly than negative ones, or vice versa); attrition bias caused by the non-random dropout of respondents over time; and desirability bias when participants tend to respond in a manner thought to be acceptable or favorable.

To get a sense of the magnitude of the recall bias, we exploit the fact that respondents were visited twice. The first wave of interviews took place in 2007 in Rwanda and 2008 in Burundi; the second wave took place in 2011 and 2015, respectively. In the second round of interviews, the respondents described their life history starting from 2000, so we have an overlapping period of life history years and PPR rankings across the two survey rounds, namely the years between 2000 and 2007/08. This allows us to assess recall bias. For the overlapping period 2000–08 for Burundian respondents, we find an average difference of 0.14 (on an 11-unit scale) across the two survey rounds, while the average gap is 0.11 for the overlapping period 2000–07 in Rwanda. These relatively small discrepancies reduce concerns of large recall bias. In the figures we present in the main text, we rely on Round 1 for the overlapping period. Relying instead on Round 2 does not alter our main conclusions (figures available on request).

When implementing consecutive data rounds, typically a proportion of respondents cannot be traced. When these “dropouts” systematically differ in terms of key characteristics of interest—PPR in our case—results may be biased. Across the two data rounds, 12.5 percent of the Rwandan respondents and 16.8 percent of the Burundian respondents dropped out. To investigate attrition bias, we compare the PPR levels reported in the first round across the dropouts and the traced respondents. We find that the reported PPR levels by these two subsamples are not too far apart, 0.48 units on average for Burundi for the period 1985–2008 and 0.22 units on average for Rwanda for the period 1985–2007. In our baseline results, we rely on the subsample of respondents that could be traced over time. Using the full unbalanced panel data gives very similar results and the same qualitative conclusions (figures available on request).

Finally, the third concern is social or political desirability, the latter of which can be very important in a repressive authoritarian regime. While we have no way of formally checking this, our data collection method likely limits the risk of socially or politically desirable answers. First, all interviews were taken in a private setting, without onlookers present. Second, the reporting is embedded in the respondent’s life history, thus imposing a “consistency constraint.” That is, the reported PPR levels need to be compatible with their related narratives of change and other events in the life history. While not completely ruling out biases, the use of a calendar approach through which event history data are collected has proven to be more reliable than standard survey approaches (Belli, Shay, and Stafford 2001), also in the context of data collection following traumatic events (Barber et al. 2016).

Findings

In this section, we start by discussing how respondents’ PPR evolved, leading up to and during identity-based mass political violence. It will reveal to what extent PPR is responsive to major macro-political events. Then we turn to the post-war recovery pattern and analyze the narratives that provide reasons for postwar changes in PPR.

PPR leading up to and during political violence

Figures 3a and 3b show PPR as reported by Burundian and Rwandan respondents throughout their life history years, starting from the year 1985. In the period 1985–92, Tutsi in Burundi clearly felt better represented than Hutu. This is unsurprising, because Tutsi officers monopolized power during this period, and the ethnic violence experienced in the past was not addressed but silenced through an imposed policy of national unity. In practice, ethnic sentiments and distinctions sharpened. Indeed, Lemarchand (2009, 129) remarks that it is not because one abolishes ethnic references that these identities cease to have meaning and force in daily life. The ethnic divide in rankings aligns with Lemarchand’s point.

Figure 3. Rankings of perceived political representation as reported in life histories. Notes: Based on the life story rankings of 302 traced Burundian respondents and 412 traced Rwandan respondents. Round 1 data series (solid line) is used for the overlapping period 2000–07/08. Round 2 data series (dotted line) starts from 2008/09 onwards. To yield representative results for the Hutu and Tutsi groups, we apply weights to the ethnic subgroups proportional to their population share. Source: Authors’ compilation in Stata.

In contrast to Tutsi in Burundi, Tutsi in Rwanda report lower PPR than Hutu in the 1980s. This is equally unsurprising given Rwanda’s Hutu majority rule at the time, and thus a (latent) awareness among Tutsi of not being represented. Hutu supremacy was institutionalized through a policy of ethnic quotas by which Tutsi were allocated 9 percent of government positions, with no real power. These PPR rankings thus empirically demonstrate that PPR is responsive to descriptive representation, and this both in the case of Burundi with its Tutsi minority rule and ethnic amnesia policy, and Rwanda with its Hutu majority rule and ethnic quota policy at the time.

Note the small dip in PPR ranking in Burundi in 1988, a year in which ethnic violence engulfed two northern communes (municipalities). The Tutsi inhabitants of the region were killed and their houses looted and destroyed. In response, the army and gendarmerie entered the region and killed Hutu civilians in an attempt to restore order, and in retaliation to the acts committed (Chrétien, Guichaoua, and Le Jeune 1988). In 1993, PPR in Burundi plunged to an all-time low. The political unrest that followed the president’s assassination and subsequent violence were unprecedented. It resulted in thousands displaced and others seeking refuge outside Burundi. As in Burundi, the onset of violence in Rwanda triggered a decline in PPR. PPR started declining in 1990, the year of the RPF invasion that marks the beginning of a civil war between the RPF and the Rwandan government. Both ethnic groups experienced extreme violence during this period. We find that PPR in Rwanda reaches an all-time low in 1994 for both ethnic groups, but more so for Tutsi who were targeted to be exterminated in the genocidal campaign.

PPR in the aftermath of political violence

After the 1993 low in Burundi, and the 1994 low in Rwanda, PPR starts a gradual recovery process. As Figures 3a and 3b show, the postwar recovery trajectories of PPR are characterized by nonlinearities, with periods of relatively fast recovery alternated by periods of relatively slow recovery or stagnation. This suggests the importance of macro-level events and policies in explaining changes in postconflict PPR. We illustrate this nonlinearity in Figure 4, where we focus on the recovery period and put changes in PPR instead of levels on the vertical axis. The grey bars indicate the average yearly change for Tutsi, while the black lines depict the average yearly change for Hutu. Apart from the recovery of PPR in the years immediately following its absolute low point, we note a number of recovery peaks, only some of which were experienced by both ethnic groups simultaneously. In particular, from 2000 onwards, the legacy of political violence gradually led to formal institutional changes, which had a positive effect on the self-reported changes in PPR in both countries, thereby providing empirical evidence that PPR is responsive to formalistic representation.

Figure 4. Average annual change in the rankings of perceived political representation. Notes: Based on the life story rankings of 302 traced Burundian respondents and 412 traced Rwandan respondents. Round 1 data series is used for the overlapping period 2000–07/08. Round 2 data series starts from 2008/09 onwards. To yield representative results for the Hutu and Tutsi groups, we apply weights to the ethnic subgroups proportional to their population share. Source: Authors’ compilation in Stata.

A first case in point for Burundi is the year 2000, when the Arusha Peace and Reconciliation Agreement was signed, resulting in power-sharing and ceasefire agreement which lay the basis for the formal relinquishing of the power monopoly held by Tutsi in Burundi. Self-reported PPR spiked, more so for Hutu than for Tutsi. The overall 2005 peak for Hutu in Burundi marked the launching of a new constitution and elections; the latter resulting in a clear victory of the CNDD-FDD with 59 percent in legislative elections (Daley 2007). It also marked the reversal of PPR across ethnic groups in Burundi. As is clear from Figure 3a, for the first time in the country’s history, Burundian Hutus felt more politically represented than their Tutsi country mates.

In Rwanda, 2000 marked the end of the insurgency war in the northwest and the start of a reconciliation and unity narrative in public discourse (Ingelaere and Verpoorten 2020). The 2003 peak in Rwanda coincided with the introduction of a new constitution and the first national-level elections won by Paul Kagame, with 95 percent of votes (Kiwuwa 2005), and his RPF party winning 74 percent of seats in the parliament (Stroh 2010). These formal institutional changes coincided with an upsurge in PPR, including for Hutu whose descriptive representation decreased. Only in 2005, the year marking the start of the gacaca information rounds, do we find that Hutu experience a drop in PPR. These information rounds led to a new wave of accusations against Hutu, whereas Tutsi RPF soldiers who retaliated against Hutus were “off-limits for Gacaca” (Rettig 2008).

Now that we have discussed which macro-political events coincided with noticeable changes in PPR, we turn to the narratives that give the self-reported reasons behind these changes, thereby providing us more insights “from below.” Doing so will, for instance, help us to unravel the puzzle of the relatively large increases in PPR by Hutu in “Tutsi-ruled” Rwanda.

Self-reported reasons for changes in postwar PPR

As explained in the “Data and Methods” section above, the open-ended narratives were collected in the second data collection round from life history year 2000 onwards, and then coded according to Pitkin’s (1967) four types of political representation: formalistic, symbolic, descriptive, and substantive. Figure 5a compares the relative frequency of these four categories across Burundian and Rwandan respondents, aggregated for the decade 2001–11. ⁹ Figure 5b adds a disaggregation layer across ethnic groups. Figures 6a and 6b set out the frequency of the four categories across time for Burundi and Rwanda, respectively.

Figure 5. Relative frequencies of the coded reasons for change in perceived political representation. Notes: BDI, Burundian; RWA, Rwandan. Based on the life story rankings of 302 traced Burundian respondents and 412 traced Rwandan respondents. To allow for cross-country comparison, we provide relative frequencies (instead of absolute frequencies), with the coded narratives summing to 100 percent for each of the bars. In addition, and also for reasons of comparison, we have restricted the time period to 2001–11, thus dropping the life history years 2012–15 for Burundi.

Figure 6. Frequencies of the coded reasons for change in perceived political representation, across time. Notes: Based on the life story rankings of 302 traced Burundian respondents and 412 traced Rwandan respondents. To allow for cross-country comparison, we provide relative frequencies (instead of absolute frequencies), with the coded narratives summing to 100 percent across the 11 years. In addition, and also for reasons of comparison, we have restricted the time period to 2001–11, thus dropping the life history years 2012–15 for Burundi. Source: Authors’ compilation in Stata.

We find that narratives mentioning positive formalistic representation account for 16.3 percent of the narratives in Burundi and 13.8 percent in Rwanda. As can be seen in Figures 6a and 6b, for both countries, narratives mentioning positive formalistic representation peak in the election years 2005 and 2003, respectively. Examples of narratives that highlight formalistic representation as a reason for an increase in PPR include: “We elected the president of the republic and I hope he will start realizing the promises made to the population” (Rwanda, Hutu, liberated prisoner, man, aged 39 years, 2003; emphasis added also in all following quotes); “The authorities had accepted to share power with other politicians who were fighting” (Burundi, Hutu, never moved, man, aged 50 years, 2002).

Positive symbolic representation accounts for 14.6 percent of the narratives in Burundi and 11.7 percent in Rwanda. We again note a concentration in the election years 2003 and 2005, respectively, thus indicating an association between formalistic and symbolic representation. For example: “The authorities treat me as any other citizen. The current chief of sector even paid me a visit… He asked me a number of questions on the size of my family and the situation of my husband” (Rwanda, Tutsi, survivor, woman, aged 53 years, 2005). “I could knock on the door of an authority without being threatened” (Burundi, Hutu, never moved, man, aged 51, 2004).

Positive descriptive representation accounts for only 2.9 percent of narratives in Burundi and 1.9 percent in Rwanda. While we see a marked reversal in PPR across ethnic groups over time, explicit mention of ethnicity in respondents’ narratives of change is relatively rare. This could be because ethnicity is only latently playing a role in a respondent’s PPR, or because respondents prefer not to name it explicitly, and we therefore did not code it as such. It is noteworthy though that the low explicit mentioning of ethnicity occurs not only in Rwanda, with its ethnic amnesia policy, but also in Burundi, with its explicit ethnic quotas. In Burundi, this conforms to contemporary political developments whereby an increased intra-ethnic competition for power has tempered the inter-ethnicity polarization (see also Raffoul 2020). Here are examples of a few narratives that were explicit: “There was a mix of ethnicities in the administration. This is why there was a change” (Rwanda, Hutu, accused in gacaca, man, aged 48 years, 2006); “All ethnic groups were represented” (Burundi, Hutu, former prisoner, man, aged 45 years, 2006).

Positive substantive representation, or the “acting for” dimension, accounts for an almost equal share in both countries: 43.0 percent of the narratives in Burundi and 43.8 percent in Rwanda. Examples of narratives pointing at substantive representation include: “Local authorities distributed aid such as beans, maize flour, clothing, and hoes” (Burundi, Hutu, never moved, aged 71 years, 2002); “Improvements in the detention conditions. It was possible and easier to chat with our friends prisoners” (Rwanda, Hutu, accused, man, aged 65 years, 2002).

Figure 5b shows that positive substantive representation in Rwanda is relatively high for Rwandan Hutu. We find that 45.9 percent of PPR changes by Rwandan Hutu are explained in terms of positive substantive representation compared to 36.6 percent for Rwandan Tutsi. One potential explanation could be that the (Hutu) Tutsi community anticipated (worse) greater substantive representation from the government, given their (decreased) increased descriptive representation, and this did not happen. The pattern also corroborates the conjecture that the Rwandan Tutsi elite tries to win hearts and minds of the Hutu majority with popular policies, and that this pays off despite the concentration of power by a small Tutsi elite, and the regime’s heavy-handedness and elimination of political opponents (Chemouni 2016; Hintjens 2008). It is also possible that this pattern stems from government propaganda that succeeds in promoting a single narrative that highlights the regime’s (socioeconomic) success. However, propaganda cannot be the whole story, as it does not explain why mentioning of positive substantive representation is higher for Hutu than for Tutsi. It neither is in tune with the observation that a relatively large share of narratives in Rwanda point to negative substantive representation, and more so for Tutsi (24.6 percent) than for Hutu (17.2 percent). If the “acting for” dimension were aligned with the “standing for” dimension, we would expect the reverse pattern, with Rwandan Tutsi referring more to net positive substantive political representation than Rwandan Hutu.

Such an alignment is exactly what we see in Burundi. The Hutu majority that enjoyed a postwar improvement in descriptive political representation refers more to all four forms of positive representation than Burundian Tutsi. For instance, and as clear from Figure 5b, positive substantive representation accounts for 44 percent of narratives for Burundian Hutu, compared to only 32 percent for Burundian Tutsi. Thus, here, there is no compensation but rather an amplification across the different dimensions of political representation. Comparing our two cases, we thus find that the inter-ethnic gap in PPR is amplified in Burundi where descriptive and substantive representation benefit the same group, but attenuated in Rwanda. As a result, Figure 3a shows an inter-ethnic divergence over time in PPR for Burundi, while Figure 3b shows an inter-ethnic convergence in PPR for Rwanda.

Discussion

Our findings demonstrate who experiences political gains and losses after identity-based violence in Rwanda and Burundi. Compared to the period before mass categorical violence, and following the reversal in descriptive political representation, PPR reverses across ethnic groups. The Tutsi respondents in Rwanda and the Hutu respondents in Burundi report increasing PPR when the Tutsi-dominated RPF and Hutu-dominated CNDD-FDD, respectively, access power and bring about changes in the ruling elite. PPR is also responsive to formalistic representation, as is clear from its rise at times of formal peace deals, the adoption of new constitutions, and the holding of elections. Also, after the coming to power of CNDD-FDD, Burundian Hutu report stronger gains in all forms of political representation than their Tutsi counterparts. In contrast, despite a strong grip on power in Rwanda by a Tutsi elite, it is not the Rwandan Tutsi but the Rwandan Hutu who report more substantive political representation, leading to a convergence of PPR across ethnic groups in Rwanda over time. This conclusion, that substantive representation closes the ethnic divide, thereby overcoming the deficit in descriptive representation in Rwanda, is also supported by the pattern of PPR over time and the associated narratives. While the election of Kagame in 2003 is less translated into an increase in PPR compared with the election in Burundi, we see more PPR changes in Rwanda after Kagame’s official coming to power and the consolidation of RPF dominance in state institutions.

These trends in PPR can be explained by focusing on the sources of political legitimacy: what makes the ruling elites legitimate in the eyes of those represented by them. Overall, one can say that in Burundi, the rulers mainly relied on input legitimacy, capitalizing on the reversal of historical injustices, and the emphasis on the rural and popular origins of the rebel movement. In contrast, in Rwanda, these potential sources of legitimacy made the RPF more illegitimate from the viewpoint of the Hutu majority of the population. Rulers in Rwanda, therefore, had to bet on a policy of output legitimacy (Chemouni 2016). In this article, we are the first to provide empirical evidence that this bet paid off. This implies that postconflict policies that seek to improve state-citizens relations and increase state legitimacy should, as much as they may prioritize liberal peacebuilding interventions, be also cognizant of the need to strengthen state capacities to deliver on substantive policies.

Our findings also suggest that people’s perception of political representation is influenced by various subjective considerations, including expectations, feelings, and prior fortunes. These subjectivities are also central to the concept of relative deprivation (Gurr 1993). Specifically, a group that is—objectively speaking—disproportionately overrepresented (e.g. Tutsi in Rwanda) can report a lower level of political representation due to “a mismatch between the social, economic or political goods people expect or feel that they are rightfully entitled to on the one hand, and what they are capable of attaining and maintaining on the other” (Hillesund et al. 2018, 465, authors’ emphasis). Conversely, a group (e.g. Hutu in Rwanda) that would expect to be disenfranchised considering the history of genocide can report a higher level of representation when the state invests in nonexcludable goods. In the same way, a relatively underrepresented group compared to its demographic weight (Hutu in Burundi) can report higher PPR, because the reference is the period of exclusion they endured for years, whereas a group that is overrepresented compared to its demographic weight (Tutsi in Burundi) can report a lower perceived representation, mainly because it uses as reference the time it enjoyed a monopoly of political power. Therefore, in divided societies, it would be ill-advised to take absolute outcomes as the relevant reference point. Instead, the benchmark that matters for citizens as they assess their fortune is the situation of their rival groups and how it has changed over time compared to theirs (see also Juon 2023).

Finally, we recognize two limitations of our study. First, our sample is composed of rural citizens. This focus on rural populations is intentional. Both in Rwanda and Burundi, the share of the rural population is about 80 percent (Hans & Naso, 2023); and it is important to understand how the largely elite-driven postconflict institutional designs resonate with peripheral populations (Agarin & McCulloch 2020). Still, a caveat is in order. Our findings may not extrapolate to urban populations. Urban citizens generally hold more critical opinions against the ruling elites, and the asymmetry of information between urban and rural populations may predispose the two groups to have diverging views on their political representation (Dorward and Fox 2022). Second, although in the design and execution of data collection, care was taken to minimize social and political desirability bias, we cannot exclude such a bias. A puzzle that emerges for instance in this respect is the scarce explicit mentioning of ethnicity in the narratives, which stands in contrast to the ethnicity-specific reactivity of PPR to inter-ethnic changes in descriptive political representation. What is reassuring for the validity of our comparative analysis though, is that this puzzle emerges to a similar extent for both our comparative case studies, though for different reasons. In Rwanda, this can be associated with the outlawing of explicit reference to ethnicity, accompanied by the ubiquitous promotion of “Rwandicity,” that is, the unity among all Rwandans. In Burundi, the pattern is consistent with the post-Arusha political reality of intra-ethnic competition, which has made ethnicity just one among other fault lines (Van Acker 2015). Thus, while ethnicity in the two countries remains recognizable to citizens and rulers, it is ultimately the political and institutional environment that determines how the ethnic division is perceived by citizens.