INTRODUCTION
Territorial autonomy is one of the most widely propagated measures to pacify multi-ethnic countries. Supporting this rationale, an influential literature finds that territorial autonomy reduces the risk of ethno-territorial rebellions (Bakke Reference Bakke2015; Cederman et al. Reference Cederman, Hug, Schädel and Wucherpfennig2015; Hechter Reference Hechter2000). However, recent observations suggest that autonomy generates tensions over control of the subnational government and the distribution of economic resources. Thereby, it may increase ethnic violence at the subnational level. Discussing Indonesia’s Aceh, Barter (Reference Barter2018, 299) concludes that autonomy has shifted “the locus of conflict from the national level to a new, regional scale.” Green (Reference Green2008, 428) argues that decentralization in Uganda has replaced “national” with “local-level conflict.” Kendhammer (Reference Kendhammer2010, 50) finds that Nigerian federalism has exacerbated violence “at the local level.” Bhattacharyya, Suan Hausing, and Mukherjee (Reference Bhattacharyya, Hausing and Mukherjee2017, 150) argue that India’s federal system unleashes “bloody conflicts at the sub-state level.” Finally, Juon and Rohrbach (Reference Juon and Rohrbach2023, 441) find that Ethiopia’s ethnofederalism has fostered “more localized forms of conflict.” Does territorial autonomy, rather than reducing ethnic violence, redirect it to the subnational level?
Investigating whether these observations form part of a generalizable pattern is of utmost political relevance. Violence between ethnic groups at the subnational level is increasing around the world (Pettersson, Högbladh, and Öberg Reference Pettersson, Högbladh and Öberg2019). In some regions, it has even become the main threat to people’s livelihoods (Fjelde and Østby Reference Fjelde and Østby2014). If territorial autonomy increased violence at the subnational level, its pacifying credentials would be in severe doubt.
Unfortunately, there is only limited cross-national evidence on how territorial autonomy affects violence between ethnic groups at the subnational level. Both theoretically and empirically, quantitative research on territorial autonomy overwhelmingly focuses on rebellions against the national government. Conversely, with few exceptions, quantitative research on subnational ethnic violence has not investigated the role of territorial autonomy, although a majority of such violence is demonstrably related to territorial issues (von Uexkull and Pettersson Reference von Uexkull and Pettersson2018, 961).
Addressing these gaps, I theorize and systematically investigate the conditions whereby territorial autonomy redirects ethnic violence to the subnational level. I proceed from the observation that autonomy provides important benefits to ethnic groups that control the subnational government. These may profit from overrepresentation in the subnational administration, disproportionate influence over its policies, and privileged access to economic goods, including land, resource rents, and jobs. This generates tensions between them and other groups that remain excluded from subnational government.
I argue that subnational tensions are more likely to escalate, if ethnic representation in the central government is unequal. If representation favors those groups that are marginalized at the subnational level, they have incentives to make far-reaching demands and may employ violence to provoke government intervention in their support. At the same time, groups that control the subnational government will be reluctant to offer substantial concessions, as these would reduce their ability to resist potential future infringements on their autonomy. Conversely, if central government representation favors the same groups that control the subnational government, they will be especially uninhibited in monopolizing political power and economic resources within their regions. In turn, this generates especially combustible grievances among their disadvantaged peers and increases the risk of episodic mass-driven violence.
I systematically test these arguments for a comprehensive sample comprising all multi-ethnic countries between 1989 and 2019. Thereby, I provide generalizable evidence that avoids potential selection bias when studying only cases where autonomy was associated with subnational violence. Avoiding overaggregation to the national level, which would preclude studying subnational violence, my units of analysis are the ethnic group and dyad settling within each region. I provide new time-variant data on regional boundaries and fine-grained information on each region’s degree of autonomy. I connect this with georeferenced and ethnically attributed information on civil violence, directed against the national government, and communal violence, whereby ethnic groups engage in violence against each other at the subnational level.
I find that groups in highly autonomous regions are less likely to employ civil violence against the center, regardless of whether they control the regional government. Hence, the pacifying consequences of autonomy extend even further than previously acknowledged. However, territorial autonomy conversely increases the risk of communal violence between ethnic groups at the subnational level. In line with my argument, these risks are especially pronounced, if central government representation is unequal. This underlines warnings that territorial autonomy may shift violence to the subnational level. Yet, it suggests that these risks can be countered by broad-based inclusion in the central government. An instrumental variables analysis indicates that these findings are unlikely due to strategic considerations whereby autonomy is tailored with respect to anticipated violence. I further demonstrate that autonomy affects groupwise economic attainments, one-sided violence, grievances, and developments during subnational conflicts as implied by my argument.
LITERATURE REVIEW
A vast literature investigates the consequences of territorial autonomy for ethnic violence. Mostly focusing on national-level dynamics, studies indicate that autonomy increases ethnic minorities’ opportunities to revolt against the national government, but reduces violence by alleviating their grievances (Bakke Reference Bakke2015; Cederman et al. Reference Cederman, Hug, Schädel and Wucherpfennig2015; Hechter Reference Hechter2000). In contrast, the relationship of autonomy with ethnic violence at the subnational level has received considerably less consideration (Hillesund Reference Hillesund2019).
One reason for this focus on national-level violence is that most studies employ units of analysis that correspond to the national (Hechter Reference Hechter2000; McGarry and O’Leary Reference McGarry, O’Leary and Noel2005) or (national) minority level (Cederman et al. Reference Cederman, Hug, Schädel and Wucherpfennig2015). Yet, an increasing proportion of ethnic violence has its roots in subnational tensions and is contained to specific regions (Barter Reference Barter2018; Cunningham and Weidmann Reference Cunningham and Weidmann2010; Lacina Reference Lacina2017). Moreover, it often does not involve the national government as a primary contestant, but pits distinct ethnic subgroups within these regions against each other directly (Boone Reference Boone2014; Fjelde and Østby Reference Fjelde and Østby2014; Hillesund Reference Hillesund2019).
A growing body of research discusses subnational ethnic violence, but does not generally investigate the role of territorial autonomy. One literature discusses conflicts between autochthonous sons-of-the-soil and recent in-migrants. Fearon and Laitin (Reference Fearon and Laitin2011) show that these are rooted in tensions over land, which may escalate if the central government sides with the in-migrants. Another literature focuses on communal violence between ethnic groups more broadly and highlights unequal access to economic resources as a driving factor (Fjelde and Østby Reference Fjelde and Østby2014). However, as noted in both literatures, autonomy critically shapes the distribution of these resources, including land, in the first place (Boone Reference Boone2014; Reference Boone2017; Cunningham and Weidmann Reference Cunningham and Weidmann2010; Fjelde and Østby Reference Fjelde and Østby2014). This underlines the importance of investigating the connection between territorial autonomy and subnational ethnic violence.
Some studies investigate the determinants of national and subnational ethnic violence side-by-side. Raleigh (Reference Raleigh2014) finds that areas inhabited by nationally excluded groups experience more civil war-related violence, whereas areas with politically irrelevant groups see more communal violence. Hillesund (Reference Hillesund2019) shows that political exclusion increases violence against the national government, whereas economic marginalization generates localized inter-group violence. These studies illuminate how ethnic inequalities shape the choice between national and subnational ethnic violence. Moreover, they highlight the importance of incorporating territorial autonomy into this logic (Hillesund Reference Hillesund2019, 546).
Some pioneering studies discuss the consequences of territorial autonomy for subnational ethnic violence. Bhavnani and Lacina (Reference Bhavnani and Lacina2018, 8) argue that autonomy exacerbates tensions over scarce land, but avert their escalation by enabling local nativist policies. Boone (Reference Boone2014, 73) shows that neocustomary land tenure regimes, often connected to autonomous ethnic homelands, insulate sons-of-the-soil conflicts at the subnational level (cf. Boone Reference Boone2017). Cunningham and Weidmann (Reference Cunningham and Weidmann2010) show that ethnic heterogeneity within regions is a predictor of overall ethnic violence. Lacina (Reference Lacina2017) investigates the conditions under which minorities in India’s states employ violence to attain their own state. Together, these studies underline the need to investigate how territorial autonomy affects subnational ethnic violence more broadly, beyond sons-of-the-soil conflicts. Moreover, they highlight the need for further disaggregation, such as tracing violence to distinct subgroups within specific regions.
Finally, numerous case studies show that territorial autonomy critically drives subnational violence in diverse settings. They indicate that autonomy may empower specific ethnic subgroups, who disproportionately benefit from economic resources connected to it. In turn, this generates grievances among their disadvantaged peers (Barter Reference Barter2018). Studies indicate that such tensions have generated subnational violence in Ethiopia (Juon and Rohrbach Reference Juon and Rohrbach2023), India (Adeney Reference Adeney2007; Bhattacharyya, Suan Hausing, and Mukherjee Reference Bhattacharyya, Hausing and Mukherjee2017), Kenya (Elfversson Reference Elfversson2019), Nigeria (Kendhammer Reference Kendhammer2010), Uganda (Green Reference Green2008), and Indonesia (Barter Reference Barter2018). These findings rest on persuasive evidence. Yet, they might be prone to similar difficulties which face the wider literature on autonomy: the focus on prominent cases where violence did, in fact, occur (Cederman et al. Reference Cederman, Hug, Schädel and Wucherpfennig2015). This underscores the importance of also considering evidence from cases where subnational ethnic conflicts may have been attenuated by territorial autonomy.
In sum, expanding the substantive research focus from national to subnational violence, disaggregating units of analysis, and remedying potential selection bias will facilitate our ability to explain whether, and under which circumstances, territorial autonomy entails a trade-off between ethnic violence at the national and subnational levels. In this study, I aim to take that next step.
THEORETICAL ARGUMENT
Concepts
I start by introducing the main concepts and actors in my framework. By territorial autonomy, I refer to institutional arrangements that transfer self-rule to spatially delimited administrative units, such as federal states, provinces, or regions. I focus on each unit’s degree of autonomy at a given moment in time.Footnote 1 Following existing research (Bakke Reference Bakke2015; Hooghe et al. Reference Hooghe, Marks, Schakel, Osterkatz, Niedzwiecki and Shair-Rosenfield2016; McGarry and O’Leary Reference McGarry, O’Leary and Noel2005), I conceive of this as co-varying with a unit’s policy, fiscal, and political competencies. This conception encompasses federations, such as Nigeria and India, and decentralized states, such as Uganda and Kenya. Moreover, it also includes arrangements limited to parts of the state territory, such as Aceh in Indonesia. All these arrangements have been linked to assertions that territorial autonomy generates subnational ethnic violence in the case studies referenced above.
Building on the widely used UCDP typology (UCDP 2021), I distinguish between two types of violence. First, by civil violence, I refer to violence in state-based conflicts between the national government and ethnically based rebels. This includes conflicts over national government control—for example, the Tutsi-led 1990 uprising in Rwanda. It also includes self-determination conflicts that involve the national government—for example, Bosnia’s secessionist civil wars between 1992 and 1995. Conversely, it does not include conflicts that exclusively revolve around subnational government control or regional boundaries (cf. von Uexkull and Pettersson Reference von Uexkull and Pettersson2018)—for example, recurring Somali-Oromo violence in Ethiopia.
Second, by communal violence, I refer to violence in conflicts between two ethnically defined non-state actors, which do not involve the central government directly and are typically waged at the subnational level (cf. Fjelde and Østby [Reference Fjelde and Østby2014, 743]; Hillesund [Reference Hillesund2019, 531–2]). This encompasses both violence between informally organized ethnic groups, such as between Hausa-Fulani and indigenous groups in Nigeria’s Plateau state (Hillesund Reference Hillesund2019, 544), and between ethnically based organizations, such as between Sunni and Kurdish militias in Syria. Conversely, my definition excludes violence between non-state actors that lack an ethnic basis, such as Mexico’s cartel violence. Moreover, it also excludes violence whereby subnational government authorities or armed organizations target unarmed civilians in a one-sided way.
Actors
Territorial autonomy establishes subnational government tiers which are subject to similar ethnic inclusion/exclusion dynamics as the national government (Cederman, Gleditsch, and Buhaug Reference Cederman, Gleditsch and Buhaug2013). Following Barter (Reference Barter2018, 301), I distinguish between two types of ethnic subgroups, defined by their ethnicity and influence in their administrative unit. First, by second-order majorities, I refer to ethnic groups that are politically dominant in an administrative unit. This often corresponds to regional majorities—for example, the Sindhi in Pakistan’s Sindh province or French speakers in Canada’s Québec province. In some cases, it refers to groups that are institutionally designated dominant in a unit—for example, the Bodo in India’s Bodoland.
Second, by second-order minorities, I refer to all other ethnic groups in an administrative unit. This includes members of groups that do not control any unit—for example, India’s Gorkha and Ethiopia’s Sidama before 2020. It also includes members of groups that control an administrative unit of their own, but find themselves outside its borders. This often refers to small ethnic exclaves in regional boundary areas—for example, Ethiopian Amhara settling in localities adjoining Amhara region.
I ideal-typically assume that both second-order majorities and minorities are split into elites who are in a position to influence political outcomes and into a larger proportion of masses who lack direct political influence.
I focus on three interactions between these actors (Figure 1). First, within each ethnic group, elites compete for group leadership and mass support. Second, at the subnational level, groups seek access to subnational government representation and the resources tied to it. Third, subnational elites interact with the central government, by bidding for national representation and support.
Figure 1. Actors and Interactions
Given the cost of violence, I expect elites to prefer to strike peaceful bargains that reflect their mutually observed capabilities—for example, as regards the distribution of subnational government offices and local economic goods. However, where such bargains are unavailable or themselves risky (Walter Reference Walter2009), they may consider violent escalatory strategies.
Civil and communal violence represent two such escalatory strategies. Both enable subnational elites to escalate bargaining for political influence or economic resources (Cederman, Gleditsch, and Buhaug Reference Cederman, Gleditsch and Buhaug2013; Hillesund Reference Hillesund2019). Yet, besides their inherent costs, their feasibility is circumscribed by elites’ ability to conjure mass support for them. Civil violence, which is directed at the central government, becomes more feasible where group members assign blame for horizontal inequalities to it (Cederman, Gleditsch, and Buhaug Reference Cederman, Gleditsch and Buhaug2013). In contrast, communal violence becomes more feasible where group members assign blame over horizontal inequalities to other groups (Hillesund Reference Hillesund2019). Moreover, in most contexts, civil violence faces a higher capability threshold than communal violence (Fjelde and Østby Reference Fjelde and Østby2014).
Territorial Autonomy and Subnational Tensions
I proceed from the observation that, depending on the competencies tied to it, control over subnational government translates into numerous political and economic benefits. These include representation in subnational government, influence over its economic, educational, and cultural policies, and access to land, resource rents, jobs, cash transfers, and communal goods (Fjelde and Østby Reference Fjelde and Østby2014). In turn, the distribution of these benefits has important consequences for the magnitude and direction of groupwise grievances.
By virtue of their dominant status in an administrative unit, second-order majorities are best poised to access these benefits. Moreover, territorial autonomy protects them from future discrimination by the central government (Hartzell and Hoddie Reference Hartzell and Hoddie2008). Thereby, similar to existing scholarship (Bakke Reference Bakke2015; Cederman et al. Reference Cederman, Hug, Schädel and Wucherpfennig2015; Hechter Reference Hechter2000; McGarry and O’Leary Reference McGarry, O’Leary and Noel2005), I expect territorial autonomy to reduce second-order majority members’ grievances against the national government.
In many contexts, second-order majority elites will have incentives to monopolize control over the subnational administration and its economic resources, to the disadvantage of second-order minorities. Their influence is often tied to the ability to improve their group’s political and economic status (Bhavnani and Lacina Reference Bhavnani and Lacina2018; Fjelde and Østby Reference Fjelde and Østby2014). Hence, second-order majority elites may deliberately exclude second-order minorities’ from subnational representation or economic resources, such as land and state jobs (Cunningham and Weidmann Reference Cunningham and Weidmann2010). In extreme cases, they may even try to remove second-order minorities from their unit altogether (Barter Reference Barter2018; Bhattacharyya, Suan Hausing, and Mukherjee Reference Bhattacharyya, Hausing and Mukherjee2017).
Consequently, territorial autonomy is less likely to improve the political and economic status of second-order minorities, who might even become more marginalized (Cunningham and Weidmann Reference Cunningham and Weidmann2010; Fjelde and Østby Reference Fjelde and Østby2014). By accentuating ethnic inequality at the subnational level, territorial autonomy makes it more likely that second-order minorities hold grievances against the second-order majority, which controls the subnational administration. Conversely, they are less likely to hold grievances against the central government, whose actions become less consequential (Fjelde and Østby Reference Fjelde and Østby2014; Hillesund Reference Hillesund2019). Hence, territorial autonomy redirects the grievances of second-order minority members away from the central government toward second-order majorities.
By shaping magnitude and direction of groupwise grievances, territorial autonomy redirects the locus of ethnic tensions from the national to the subnational level. For second-order majority elites, privileging their members in the distribution of political posts and economic resources is a more cost-effective strategy to improve their group’s status and secure influence than risky challenges against the central government. Conversely, for second-order minority elites, rallying against the second-order majority’s privileges becomes more attractive, due to the grievances directed against it. Moreover, this also promises them more easily attainable gains, such as subnational government representation or boundary changes.
These arguments can be illustrated with numerous examples. In India’s federal states, autonomy defused grievances against the central government and largely averted civil conflict (Bakke Reference Bakke2015). However, in many states, it fostered subnational majoritarianism which generated discontent among marginalized groups (Bhattacharyya, Suan Hausing, and Mukherjee Reference Bhattacharyya, Hausing and Mukherjee2017; Bhavnani and Lacina Reference Bhavnani and Lacina2018). In Indonesia’s Aceh, autonomy successfully ended its long-lasting secessionist war. However, Acehnese militants completely monopolized control over the autonomous province, which generated grievances among Aceh’s highland communities (Barter Reference Barter2018). In Nigeria, the proliferation of ethnic states deflated previously widespread secessionist movements. However, it generated discontent among minorities that were not recognized as indigenous and excluded from local economic opportunities (Hillesund Reference Hillesund2019; Kendhammer Reference Kendhammer2010). In Ethiopia, ethnofederalism reduced civil violence by appeasing each region’s titular groups. Yet, these often monopolized local power, which fueled agitation by non-titular groups for local power-sharing or regional boundary changes (Juon and Rohrbach Reference Juon and Rohrbach2023; Van der Beken Reference Van der Beken2015).
Uneven Central Government Inclusion and the Escalation of Subnational Tensions
So far, my argument has centered on territorial autonomy in isolation. However, there are good grounds to suspect that whether or not subnational tensions escalate depends on the third actor in my framework, the central government.
The central government can prevent the escalation of tensions between second-order majorities and minorities in two main ways. First, it can encourage conciliatory behavior by second-order majorities. Thereby, it can preempt the formation of grievances among second-order minorities and address subnational tensions at their root. Most importantly, it may encourage local power-sharing pacts (Bunte and Vinson Reference Bunte and Vinson2016; Elfversson Reference Elfversson2019). Second, where communal relations are already tense, it may take active measures to prevent them from escalating. For instance, it may mediate and impose compromises between second-order majorities and minorities (Elfversson Reference Elfversson2015; Wilkinson Reference Wilkinson2004).
However, the central government’s ability to take these deescalatory steps is constrained by groupwise representation within it. It is well positioned, if it includes representatives from both second-order majorities and minorities. In such constellations, it has leverage over both sides (cf. Svensson Reference Svensson2009). In turn, it can rely on preexisting elite networks (cf. Bakke Reference Bakke2015) to promote compromises that avoid costly violence. Conversely, the situation is less amenable to deescalation, if second-order majorities and minorities are unevenly included within the central government. Such constellations make it more likely that the central government disproportionately depends on the support of the included group (Elfversson Reference Elfversson2015; Fearon and Laitin Reference Fearon and Laitin2011; Lacina Reference Lacina2017).
Included Second-Order Minority/Excluded Second-Order Majority
A first type of uneven inclusion is asymmetric: the second-order majority controls the subnational government, but is excluded from the central government; conversely, second-order minorities are excluded from subnational government, yet form part of the central government.
In such asymmetric constellations, second-order minority elites may reasonably expect the central government—which includes their own representatives—to support settlements favorable to them (Elfversson Reference Elfversson2015; Lacina Reference Lacina2017; Wilkinson Reference Wilkinson2004). As a result, they have incentives to make especially far-reaching demands, such as prominent subnational representation or a significant share of economic resources. Moreover, they can also more credibly threaten to escalate their demands, should second-order majority elites remain unwilling to offer concessions. Similar to in-migrants from nationally dominant groups in sons-of-the-soil conflicts (Boone Reference Boone2017; Fearon and Laitin Reference Fearon and Laitin2011), they may expect government intervention on their side, should violence erupt.
In this constellation, efforts to resolve tensions between second-order majorities and minorities face obstacles in the form of information failures and commitment problems. These are similar to those complicating national-level bargaining between the government and ethnic minorities in the shadow of kin-state intervention (Cetinyan Reference Cetinyan2002). First, especially to nationally excluded second-order majority elites, it is unclear at which point the central government, often myopic to subnational struggles (Lacina Reference Lacina2017), will intervene. Hence, risk-averse second-order majority elites may react with concessions to second-order minorities’ far-reaching demands (Lacina Reference Lacina2017). However, they might also speculate that they can repress second-order minorities before violence reaches the threshold that triggers central government intervention. This is particularly likely under high degrees of autonomy, which enable second-order majority elites to coordinate repressive action against second-order minorities and allow them to finance, inculcate, arm, and rally their membership for this purpose (Cunningham and Weidmann Reference Cunningham and Weidmann2010; Hartzell and Hoddie Reference Hartzell and Hoddie2008).
Second, nationally excluded second-order majority elites may be unwilling to offer substantial concessions, even if they assess the risks of violent escalation and central government intervention as high. Owing to their exclusion at the national level, territorial autonomy constitutes a particularly valuable protection of their group’s future political and economic status—and, by extension, of their own political influence (Cederman et al. Reference Cederman, Hug, Schädel and Wucherpfennig2015; Cunningham and Weidmann Reference Cunningham and Weidmann2010; Hartzell and Hoddie Reference Hartzell and Hoddie2008). Concessions that reduce their control over subnational government—and empower rival groups associated with the center—weaken their strength to resist potential future encroachments on their attained autonomy. For instance, sharing control over the subnational executive makes it harder for them to coordinate and fend off attempts at re-centralization in the future. Hence, even in the face of likely intervention, they may, at least initially, be reluctant to offer concessions. Instead, they might try to proactively create facts on the ground, for example, by using subnational coercive organs, such as regional militias or police forces, to displace second-order minorities (cf. Bhattacharyya, Suan Hausing, and Mukherjee Reference Bhattacharyya, Hausing and Mukherjee2017).
The potential of this asymmetric constellation to attract central government intervention in favor of second-order minorities—and generate incentives for far-ranging demands that are prone to escalation—can be illustrated with the example of Ethiopia’s Benishangul-Gumuz region. In the early 1990s, the Gumuz controlled Benishangul-Gumuz region, but were excluded from the central government (cf. Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). Conversely, Amhara were marginalized within Benishangul-Gumuz, yet held prominent posts in Ethiopia’s central government. Tensions escalated into communal violence in 1992 and 1993, following Amhara agitation for far-ranging political rights within the region. Owing to unwillingness by Gumuz leaders to accommodate these demands, the central government intervened, sanctioned the creation of a special autonomous area for the Amhara in Pawe, and unilaterally decreed Amhara representation in Benishangul-Gumuz’ government (Juon and Rohrbach Reference Juon and Rohrbach2023; Kefale Reference Kefale2013).
A second example illustrates how this asymmetric constellation may make second-order majority elites less amenable to concessions and tempt them to use their subnational government control to proactively repress second-order minorities, even amid repeated central government intervention. In Nigeria’s Plateau state, government posts and economic opportunities were monopolized by groups who were recognized as indigenous, but excluded from the federal government. Plateau also hosted a disenfranchised Hausa-Fulani minority whose representatives were overrepresented in the federal government (Hillesund Reference Hillesund2019). Violence between these groups attracted the federal government’s intervention numerous times. For example, it created a majority-Hausa-Fulani district in Jos (Bunte and Vinson Reference Bunte and Vinson2016) and temporarily suspended Plateau’s autonomy (Ostien Reference Ostien, Chesworth and Kogelmann2009; UCDP 2021). However, instead of making indigenous elites more conciliatory, such moves “reinforced suspicions… of a conspiracy of northern Muslims” to sabotage the autonomy arrangement and claim parts of Plateau’s territory “for themselves” (Ostien Reference Ostien, Chesworth and Kogelmann2009, 17). Rather than offering concessions, they used their control of Plateau state’s government and influence over its economic and education policies to accelerate indigenization programs, which sparked renewed communal violence (UCDP 2021).
Included Second-Order Majority/Excluded Second-Order Minority
A second type of uneven inclusion is the inverse situation whereby the second-order majority is included in both the subnational and national governments, whereas second-order minorities are excluded at both levels.
In such symmetric constellations, it is second-order majorities who can conceivably expect central government support (Lacina Reference Lacina2017). This means they will be less inclined to offer substantial concessions to second-order minorities. Conversely, their clear disadvantage should dissuade second-order minority elites from making far-reaching demands and from purposefully initiating violence (Lacina Reference Lacina2017). However, uninhibited subnational majoritarianism that heavily disadvantages second-order minority members also generates particularly combustible mass grievances. These might escalate into episodic violence in several ways. First, aggrieved second-order minority members may engage in localized violence against second-order majorities, for example, to access vital economic resources such as land (cf. Hillesund Reference Hillesund2019). Second, initially peaceful mobilization by second-order minorities may escalate, for example, in clashes with counter-mobilized second-order majority members (Horowitz Reference Horowitz2001; Wilkinson Reference Wilkinson2004). Third, one-sided violence by second-order majority elites may force second-order minority members to take up arms in self-defense (Bhattacharyya, Suan Hausing, and Mukherjee Reference Bhattacharyya, Hausing and Mukherjee2017; Elfversson Reference Elfversson2019; Horowitz Reference Horowitz2001).
In this constellation, the central government might use its leverage over second-order majority elites to force them into concessions (Svensson Reference Svensson2009). However, the central government may be reluctant to do so, especially if the second-order majority makes up an significant part of its support base and if violence remains at a low intensity (Elfversson Reference Elfversson2015). Moreover, as communal violence in this constellation is driven by mass grievances, second-order minority elites may be unable to convince their membership to end the violence. Owing to their exclusion at the national level, aggrieved second-order minority members may not trust the central government’s ability to mediate impartially nor its willingness to uphold local peace agreements (Brosché Reference Brosché2014; Elfversson Reference Elfversson2019).
The intransigence of second-order minority elites in this second constellation—and its potential to spark episodic communal violence—can be illustrated with Indonesia’s Aceh province. There, tensions intensified following the inclusion of ethnic Acehnese in Indonesia’s central government in 2004, whereas Aceh’s highland minorities remained excluded at both levels (cf. Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). Counting on central government indifference, Acehnese elites increasingly discriminated against Aceh’s highland communities. This led to a mass exodus (UCDP 2021), but also generated “protest, ethnic tensions, violence and even provincial separatism” (Barter Reference Barter2018, 303). Acehnese militants cracked down on dissent violently and with relative impunity. In 2009, they killed protesters demanding the creation of a new province for Aceh’s highland communities. In 2013, another crackdown sparked clashes between Acehnese and ethnic highlanders that claimed dozens of casualties (Barter Reference Barter2018).
Another example is Kibaale district in Uganda. In Kibaale, the Banyoro were politically dominant; moreover, they also formed part of the central government coalition (cf. Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). In contrast, the Bakiga were not only politically subordinate in Kibaale, but excluded from the national government as well. Following increased in-migration by Bakiga settlers, tensions increased. Fearing a loss of power, Banyoro militants engaged in a campaign of violence and hate speech before the 2002 local elections. Following the unprecedented election of a Bakiga, Banyoro extremists responded with indiscriminate violence. Rather than reprimanding them, the central government intervened and reinstated an ethnic Banyoro to power (Green Reference Green2008). This exacerbated Bakiga grievances and sparked renewed episodic clashes.
Hypotheses
I distill two hypotheses from the above arguments. These capture the hypothesized trade-off induced by territorial autonomy, along with its moderation by uneven central government inclusion for communal violence:
Hypothesis 1. The higher an administrative unit’s degree of autonomy, the lower the probability that second-order majorities and second-order minorities will engage in civil violence.
Hypothesis 2. The higher an administrative unit’s degree of autonomy, the higher the probability that second-order majorities and second-order minorities will engage in communal violence against one another, if they are unevenly included in central government.
DATA AND RESEARCH DESIGN
I now test these hypotheses in a quantitative analysis of all multi-ethnic states between 1989 and 2019.Footnote 2 I identify all 139 states with an ethnic minority population share of at least 5% according to the Ethnic Power Relations dataset (Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). Thus, I also include countries where ethnic violence never erupted in my main analyses. This serves to avoid potentially biased findings arising from selecting on the dependent variable (Cederman et al. Reference Cederman, Hug, Schädel and Wucherpfennig2015).
Unit of Analysis
My unit of analysis is the ethnic group (for civil violence) or dyad (for communal violence) in each administrative unit year. For this purpose, I assemble the new Significant Administrative Units Dataset (SAU), providing time-variant spatial information on administrative boundaries. I start off with the most recent polygons at the first-order administrative level, which I take from the Database of Global Administrative Areas (GADM).Footnote 3 I consult additional sources to identify changes in these boundaries over time. To obtain polygons for historical boundaries, I modify those provided by GADM and manually georeference historical maps. The result is a set of 6,152 geocoded polygons, each of which represents a specific unit in a given time period between 1988 and 2018.
I then assemble a list of ethnic groups who settle within each unit. I start off with the ethnic group list coded by the Ethnic Power Relations (EPR) dataset (Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). I then calculate the minimum distances between each unit and each group’s settlement area and identify all groups that settle within 50 km of each unit. I do so, rather than exclusively focusing on areas of overlap, for two reasons. First, EPR codes the area in which the majority of a group’s membership lives. This does not rule out that group members settle, and become involved in violence, in other parts of the country. For example, Ethiopia’s Amhara clashed with Oromo in Dire Dawa city numerous times, which is located outside their main settlement area (UCDP 2021). Second, pockets of group members often settle in boundary areas adjoining the units controlled by their kin. In line with my argument, such groups may mobilize in favor of regional boundary changes. For example, ethnic Somali in Oromia’s border regions demanded that these areas be transferred to the Somali region, which sparked recurring clashes with the Oromo (UCDP 2021).Footnote 4
Independent Variables
Unit-Level Degree of Autonomy
To assess my hypotheses, I require detailed information on each administrative unit’s degree of autonomy. To obtain such information, I assess, based on national constitutions and autonomy statutes embedded in them, each unit’s degree of autonomy. I code seven indicators, each normalized to a range from 0 to 1. Similar to the Regional Authority Index (RAI) (Hooghe et al. Reference Hooghe, Marks, Schakel, Osterkatz, Niedzwiecki and Shair-Rosenfield2016), these are grouped along three components (see Figure 2): policy autonomy (institutional depth/policy scope), fiscal autonomy (taxing competencies/borrowing competencies/financial guarantees), and political autonomy (independent executives/independent legislatures). I aggregate these indicators into an overall measure for each unit’s degree of territorial autonomy, ranging from 0 to 1 (see Figure 3 for a graphical overview of 2019).
Figure 2. Territorial Autonomy: Components and Indicators
Note: All indicators and components are normalized to a range from 0 to 1. See the Supplementary Material for measurement details.
Figure 3. Administrative Units, Territorial Autonomy, and Civil and Communal Violence, 2019
Second-Order Majorities and Minorities
My hypotheses require me to distinguish between second-order majorities and minorities within each administrative unit. I first determine whether an administrative unit is designated the ethnic homeland of a specific ethnic group. If this is the case, I identify said group as the second-order majority. For example, this applies to the Bodo in India’s Bodoland region, the Walloon in Belgium’s Wallonie region, and the Harari in Ethiopia’s Harari People’s state. In all other cases, I identify the group with the largest population share in each administrative unit as the second-order majority. I identify all other groups as second-order minorities.
Central Government Inclusion
To test Hypothesis 2, I require information on groupwise central government inclusion. I first construct a binary variable included based on EPR (Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015), which I employ as a control in my civil violence models. For my communal violence models, I construct two dyad-level equivalents of this variable: included/excluded (1 under uneven representation—if one dyad member is included, whereas the other is excluded—and 0 otherwise) and excluded/excluded (1 if both are excluded, and 0 otherwise).
Dependent Variables
To identify civil and communal violence events in each unit, I rely on the Uppsala Conflict Data Program Georeferenced Event Dataset (UCDP-GED) (Sundberg and Melander Reference Sundberg and Melander2013). This provides information on violence of either type that resulted in at least 25 battle-related deaths in at least 1 year between 1989 and 2019. To connect civil violence to each group, I use information on organizational claims and recruitment provided by the ACD2EPR dataset (Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). To connect communal violence to ethnic dyads, I attribute the participants on both sides of each non-state conflict (Pettersson, Högbladh, and Öberg Reference Pettersson, Högbladh and Öberg2019; Sundberg, Eck, and Kreutz Reference Sundberg, Eck and Kreutz2012) to each group by name (for informally organized conflicts) or by considering the ethnic composition of involved organizations (for formally organized conflicts). I then spatially intersect the location of conflict events with my units’ boundaries to obtain georeferenced dependent variables. These take the value 1 if a group/dyad was involved in at least one civil/communal violence event in a unit year and 0 otherwise.
Controls
I control for confounders that may simultaneously affect a unit’s degree of autonomy and risks of violence. In my civil violence models, I control for a group’s relative size, ranging from 0 to 1, at the national (Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015) and unit levels. In my communal violence models, I employ the dyadic equivalents of these measures, by controlling for their mean and absolute difference between them and for asymmetric situations where the second-order minority is the nationally most influential group in the given year (Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). I also control for a unit’s population as a proportion of the country’s total, and its geographic area (both logged).
I further control for factors that influence the incidence of violence in a unit: petroleum reserves (Lujala, Ketil Rod, and Thieme Reference Lujala, Rod and Thieme2007), variance in surface ruggedness (Shaver, Carter, and Shawa Reference Shaver, Carter and Shawa2019), and distance from the capital and national border (both logged). At the country level, I include standard controls for GDP per capita (logged), population (logged),Footnote 5 the level of democracy (normalized to a range between 0 and 1, based on Marshall, Gurr, and Jaggers [Reference Marshall, Gurr and Jaggers2019]), ethnic fractionalization (Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015), and national election years (Hyde and Marinov Reference Hyde and Marinov2012). To account for stable differences between world regions and non-linear time trends, I include region-Footnote 6 and year-fixed effects. Finally, I account for temporal dependence by including a cubic term measuring the number of years since last violence of the respective type, and spatial dependence by including spatial lags.Footnote 7
In this article’s Supplementary Material, I provide detailed information on each of these data contributions.
QUANTITATIVE ANALYSIS
My sample includes all 139 multi-ethnic countries, as identified above, between 1989 and 2019. It is delimited in three further ways. First, in my civil violence analyses, I exclude groups that EPR (Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015) judges as politically irrelevant or as holding the highest position in the national government.Footnote 8 These groups, by definition, never engage in civil violence in the EPR classification. For my communal violence analyses, I include all dyads, as groups that are nationally irrelevant or dominant may still engage in communal violence over subnational issues. Second, my sample includes all administrative units, including those where the nationally most powerful group constitutes the second-order majority. Third, as my theory focuses on the determinants of initial violence in a unit, rather than its spatial diffusion, I drop all group/dyad years that have seen violence of the respective type anywhere in the country in the previous year from my analysis. I show that my results are not driven by any of these decisions in Appendix 3.7.2 of the Supplementary Material.
In all analyses, I make sure my measurement of all independent variables is temporally prior to my measurement of the dependent variables. Hence, my measurement of administrative units and their degree of autonomy captures the situation on December 31 in year t, my controls capture the situation in year t, my measure for government inclusion refers to the situation on January 1 in year
$ t+1 $
, while my violence measures encompass any time point in year
$ t+1 $
.
To assess my hypotheses, I estimate Equation 1 and Equation 2 using logit models with standard errors clustered by country:
$$ \begin{array}{ll}\begin{array}{lll}{y}_{gut+1}=& {\hskip-0.15em \beta}_0+{\alpha}_1autonom{y}_{ut}+{\alpha}_2include{d}_{gt+1}+& \\ {}& {\hskip-0.1em \beta}_1{X}_{1gt}^{\prime }+{\beta}_2{X}_{2gut}^{\prime }+{\beta}_3{X}_{3ut}^{\prime }+{\beta}_4{X}_{4ct}^{\prime }+& \\ {}& {\hskip-0.15em \gamma}_r+{\delta}_t+{\epsilon}_{gut},\end{array}& \end{array} $$
$$ \begin{array}{ll}\begin{array}{lll}{y}_{dut+1}=& {\beta}_0+{\alpha}_1autonom{y}_{ut}+{\alpha}_2incl./excl{.}_{dt+1}+& \\ {}& {\alpha}_3autonom{y}_{ut}\cdot incl./excl{.}_{dt+1}+& \\ {}& {\beta}_1{X}_{1dt}^{\prime }+{\beta}_2{X}_{2dut}^{\prime }+{\beta}_3{X}_{3ut}^{\prime }+{\beta}_4{X}_{4ct}^{\prime }+& \\ {}& {\gamma}_r+{\delta}_t+{\epsilon}_{dut},\end{array}& \end{array} $$
where
$ {y}_{gut+1} $
and
$ {y}_{dut+1} $
denote the probability of civil/communal violence involving group g and dyad d in unit u at time
$ t+1 $
.
$ {X}_{1gt}^{\prime } $
,
$ {X}_{1dt}^{\prime } $
,
$ {X}_{2gut}^{\prime } $
, and
$ {X}_{2dut}^{\prime } $
are my group/dyad and group/dyad-unit-level controls,
$ {X}_{3ut}^{\prime } $
my unit-level controls, and
$ {X}_{4ct}^{\prime } $
my country-level controls, as introduced above, respectively.
$ {\gamma}_r $
and
$ {\delta}_t $
are fixed effects at the region and year levels. Finally,
$ {\epsilon}_{dut} $
and
$ {\epsilon}_{dut} $
denote the group/dyad-unit year error terms, respectively.
Table 1 presents four models containing my main results. Focusing on civil violence, models 1 and 2 are based on Equation 1. To test Hypothesis 1, these separately include only second-order majorities and second-order minorities, respectively. Turning to communal violence, models 3 and 4 are based on Equation 2, estimated for all second-order majority/minority dyads in each unit. Model 3 introduces my variables for territorial autonomy and included/excluded as separate terms, while model 4 interacts them to test conditional Hypothesis 2. Figure 4 visualizes the average marginal effects of territorial autonomy for the observed values in the sample.
Table 1. Territorial Autonomy and Civil/Communal Violence Incidence
Note: Country-clustered SEs in parentheses; constant, group-/dyad-, country-, unit-level controls, and cubic terms for group-/dyad-wise peace years included but not reported; maj. = second-order majority; min. = second-order minority. The dependent variable is a binary variable equal to one if there is at least one instance of civil/communal violence involving a group/dyad in a given unit. For full results, see Table A5 in Appendix 2 of the Supplementary Material. *
$ p<0.05 $
, **
$ p<0.01 $
, ***
$ p<0.001 $
.
Figure 4. Territorial Autonomy and Civil/Communal Violence Incidence of Second-Order Majorities (maj.) and Second-Order Minorities (min.)
Note: Partial effects and 95% confidence interval of territorial autonomy (based on models 1–4 in Table 1). Maj. = second-order majority; min. = second-order minority.
The results support my argument that territorial autonomy decreases the risk of civil violence not only for second-order majorities, but also for second-order minorities (Hypothesis 1). The marginal effect of territorial autonomy in models 1 and 2 indicates that second-order majorities in units with maximum levels of autonomy are 1.01% less likely to engage in civil violence in a given year (p = 0.02), while second-order minorities are 0.22% less likely to do so (p = 0.02). This corresponds to a sizeable reduction of civil violence risks in relative terms—around 69% in both cases, as a unit’s degree of autonomy increases from 0 to 1. Thereby, my findings echo previous research highlighting the pacifying effect of territorial autonomy. Moreover, they indicate that this effect extends to second-order minorities, whose grievances autonomy redirects toward the second-order majority.
Turning to communal violence, my results do not support assertions that territorial autonomy increases communal violence across the board. In model 3, the unconditional marginal effect of territorial autonomy, although positive, is statistically indistinguishable from zero (p = 0.25). However, consistent with my argument, my results indicate that territorial autonomy substantially increases these risks for unevenly included dyads (Hypothesis 2). In model 4, the marginal effect of territorial autonomy indicates that in units with maximum levels of autonomy, communal violence among included/excluded dyads becomes 0.33% more likely in a given year (p = 0.018). This corresponds to an almost eightfold increase of this risk in relative terms, as autonomy increases from 0 to 1.
In Appendix 3 of the Supplementary Material, I show that these findings are unlikely accounted for by alternative explanations and robust to alterations to the specification.
ACCOUNTING FOR ENDOGENEITY
In spite of these substantively meaningful and robust results, it is not yet possible to infer that territorial autonomy causally affects civil and communal violence risks as indicated. In particular, my results could be biased in two opposite ways (McGarry and O’Leary Reference McGarry, O’Leary and Noel2005). First, it might be that opportunistic governments appease threatening groups by providing them units of their own and awarding them higher degrees of autonomy. Second, the inverse situation applies where they attempt to preempt violence by strategically demobilizing troublesome groups, for instance, by withholding autonomy from them.
Previous research demonstrates that governments frequently provide autonomy to ward off civil violence by conflict-prone groups (Cederman et al. Reference Cederman, Hug, Schädel and Wucherpfennig2015). In line with the first type of endogeneity outlined above, this would lead me to underestimate the pacifying effect of autonomy for civil violence involving second-order majorities, while potentially overestimating it for second-order minorities. The situation is less clear as regards communal violence. Communal violence often remains confined to specific locations and is thereby less consequential for national government survival. However, it may nevertheless entail longer-term threats, for instance, by escalating into civil war (Fjelde and Østby Reference Fjelde and Østby2014). Where national governments anticipate such risks, they might be reluctant to provide autonomy to units that are most at risk of communal violence. In line with the second type of endogeneity discussed above, this would lead me to underestimate the conflict-inducing effect of territorial autonomy on communal violence.
To address this issue, I instrument for unit-level autonomy. Focusing on former European non-settler colonies in Asia and Africa, I build on two insights by Cederman et al. (Reference Cederman, Hug, Schädel and Wucherpfennig2015): First, minority groups of larger absolute size are more likely to obtain autonomy in the colonial period and inherit it after independence. Second, the strength of this relationship is conditional on the former colonizer’s identity. Specifically, it is attenuated in colonies formerly ruled by France, which employed a system of centralized, direct rule. In contrast, other colonizers often relied on indirect rule. Together, this creates systematic variation in autonomy in the postcolonial period that is conceivably unrelated to anticipated risks of violence.
My setup is mostly analogous to my main models. My approach requires four adjustments, however. First, administrative boundaries may themselves be endogenous to violence. I hence switch my unit of analysis to the ethnic group- and dyad-grid cell-year (Tollefsen, Strand, and Buhaug Reference Tollefsen, Strand and Buhaug2012). Second, I exclude grid cells in which the largest group is the politically most powerful group according to EPR, as my instrument only applies to areas inhabited by minorities. Third, I employ the grid cell equivalents of my unit-level variables. Fourth, in my communal violence models, instead of interacting territorial autonomy with included/excluded, I estimate separate specifications for all dyads and the subset of included/excluded dyads.
To construct my instrument, I interact the absolute size of the largest group in a grid cell in the given year (logged)Footnote 9 with a dummy variable identifying former French colonies. Analogously to Cederman et al. (Reference Cederman, Hug, Schädel and Wucherpfennig2015), I expect grid cells with larger minority groups to inherit higher degrees of territorial autonomy in the postcolonial period. In contrast to other former colonies, this relationship should be considerably weaker in former French colonies. A key limitation of this instrument is that it is largely time invariant and hence reaches only moderate levels of instrument strength. However, as I show in Appendix 4 of the Supplementary Material, it remains a relevant predictor of territorial autonomy in the time period analyzed; moreover, my results remain robust to procedures that account for weak instruments.
Figure 5 visualizes the results of four bivariate probit models that instrument for grid cell-level autonomy, along with their uncorrected equivalents. My findings are in line with the expectation that areas at risk of civil violence will attain higher degrees of autonomy, while autonomy is more likely to be withheld from areas most at risk of communal violence. While the average marginal effect of territorial autonomy on civil violence remains statistically non-significant in models 6 and 8 (p = 0.437 and p = 0.075, respectively), it is more strongly negative than in the uncorrected corresponding models using the same limited sample (5 and 7). Moreover, the positive effect of territorial autonomy on communal violence between included and excluded groups remains positive, statistically significant, and becomes substantially stronger in model 12.
Figure 5. Instrumental Variable Approach: Territorial Autonomy and Civil/Communal Violence Incidence of Second-Order Majorities and Second-Order Minorities
Note: Sample includes former European non-settler colonies in Africa and Asia. Partial effects and 95% confidence interval of territorial autonomy (based on models 5–12 in Tables A6 and A7 in Appendix 4.1 of the Supplementary Material).
While far from ideal, these results increase confidence that territorial autonomy indeed reduces the risk of civil violence, while increasing the risk of communal violence for unevenly included second-order majority/minority dyads. In Appendix 4 of the Supplementary Material, I provide full model results, examine the underlying assumptions of this approach, and discuss potential violations of the exclusion restriction.
MECHANISMS
Additional Quantitative Analyses
In a final step, I substantiate my arguments by investigating the stepwise implications of my hypothesized mechanisms. I start by summarizing the results of quantitative analyses that examine additional outcomes of interest. In Appendix 5 of the Supplementary Material, I provide more details on these analyses along with full results.
First, I have argued that second-order majorities disproportionately benefit from territorial autonomy, for example, in terms of access to economic resources. To probe this argument, I examine the association of territorial autonomy with stable nightlight emissions in each group’s settlement area within each unit. Figure 6a shows that autonomy is indeed associated with higher nightlight emissions in areas inhabited by second-order majorities, but with lower emissions in areas inhabited by second-order minorities.
Figure 6. Territorial Autonomy, Nightlight Emissions, One-Sided Violence, and Grievances
Note: Partial effects and 95% confidence interval; based on model A161 in Table A8 (Appendix 5.1.2 of the Supplementary Material), models A162–A164 in Table A9 (Appendix 5.2.2 of the Supplementary Material), and models A165 and A166 in Table A11 (Appendix 5.3.4 of the Supplementary Material); maj. = second-order majority; min. = second-order minority.
Second, I have argued that territorial autonomy may create incentives for second-order majority elites to forcibly displace second-order minorities. In extreme cases, they may engage in violent ethnic cleansing, especially amid (symmetric) uneven inclusion in the central government. Using data on one-sided ethnic violence in Africa and Asia (Sundberg and Melander Reference Sundberg and Melander2013; Fjelde et al. Reference Fjelde, Hultman, Schubiger, Cederman, Hug and Sollenberg2021), I examine this implication. Figure 6b shows that territorial autonomy indeed substantially increases the risks of one-sided violence by second-order majorities against second-order minorities, especially if second-order majorities are included in central government, while second-order minorities are excluded.
Third, I have argued that territorial autonomy alleviates the grievances of second-order majorities, while redirecting second-order minorities’ grievances to the subnational level. To probe this argument, I make use of a large collection of ethnically attributed mass surveys, relying on survey items that proxy for grievances with individual perceptions of belonging to a group that is discriminated (Juon Reference Juon2023). Figure 6c indicates that territorial autonomy is associated with lower grievances by second-order majorities, as given by this measure. Conversely, this association is weaker and statistically non-significant for second-order minorities. While in line with the observable implications of my argument, one important limitation of this procedure is that I am unable to directly test whether territorial autonomy redirects second-order minority grievances toward the second-order majority specifically.
In addition, I also quantitatively examine how uneven government inclusion affects central government mediation attempts, interventions, negotiations, and agreements during ongoing communal conflicts in Africa. Overall, the results are in line with the mechanisms I postulated as driving violent escalation under the two different constellations of uneven central government inclusion. I report these analyses in Appendix 5.4 of the Supplementary Material.
Model-Testing Small-N Analysis
Finally, I conduct a model-testing small-N analysis to “gain contextually based evidence” that my model “actually ‘worked’ in the manner specified” by my theory (Lieberman Reference Lieberman2005, 442). For this purpose, I select three cases where model 4 indicates that territorial autonomy substantially increased the risks of communal violence between included/excluded dyads. In Appendix 5.5 of the Supplementary Material, I provide more details on the selection criteria; Figure A26 in the Supplementary Material situates the three selected cases in my sample.
Sindh-Mohajir in Sindh Province, Pakistan
My first case is situated in Sindh province in Pakistan, where ethnic Sindhi form the second-order majority. Provincial autonomy alleviated their grievances and deflated previous Sindhi secessionist movements. Sindh also harbors a large Mohajir minority. Mohajir grievances turned against the Sindhi in 1972, after the provincial government attempted to install Sindhi as the sole official language, which would have barred Mohajirs from the administration. Addressing these grievances, the Mohajir Qaumi Movement (MQM) demanded the separation of Mohajir-dominated Karachi from Sindh. This was vehemently opposed by Sindhi elites, who depended on Karachi for its urban job market. Tensions escalated following the breakdown of a short-lived coalition that had included the Mohajirs at the national level. Prime Minister Bhutto (herself a Sindhi) reneged on promises of concessions, which exacerbated Mohajir discontent and sparked Sindhi-Mohajir communal violence in 1990 (Adeney Reference Adeney2007; Zaidi Reference Zaidi1991). Violence subsided following military intervention and the renewed national-level inclusion of Mohajir representatives in the late 1990s (cf. Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). Eventually, this generated cooperative, non-violent relations between Sindhi and Mohajirs (Adeney Reference Adeney2007, 150).
Dinka-Nuer in Jonglei State, South Sudan
My second case are the Dinka and Nuer in South Sudan’s Jonglei state. In 2005, the creation of an autonomous South Sudanese region, which encompassed Jonglei, alleviated grievances against the central government and ended a long-lasting secessionist civil war. However, communal violence between Dinka and Nuer also became more pronounced in this period, claiming more than 200 lives in 2010 alone (UCDP 2021). This escalation followed the inclusion of the Dinka into Sudan’s central government, whereas the Nuer remained excluded. Following the creation of South Sudan as an independent, decentralized state in 2011, tensions in Jonglei increasingly revolved around the boundaries of counties and districts within Jonglei—which determined access to arable land and state jobs (Brosché Reference Brosché2014, 131). Tensions also intensified over the distribution of development resources by Jonglei’s government, which strongly favored Dinka-inhabited areas (Brosché Reference Brosché2014, 133). A factor contributing to the escalation was the uneven use of sanctions by the Dinka-dominated central government, which disproportionately targeted the Nuer (Brosché Reference Brosché2014). Violence escalated further, when allegations of a coup attempt led to the expulsion of Vice President Riek Machar’s Nuer faction from government in 2013 (Brosché Reference Brosché2014).
Ilaje-Ijaw in Ondo State, Nigeria
My third case is situated in Nigeria, where successive federal constitutions saw a decline of Yoruba and Ijaw secessionism. However, they also generated subnational tensions (Kendhammer Reference Kendhammer2010). One example is Ondo state, where ethnic Ilaje dominate the state government, whereas the Ijaw are a politically subordinate minority. Tensions between the two groups increased following the discovery of oil in 1997 (Damson Reference Damson2015) and in anticipation of Nigeria’s new 1999 constitution, which reinstated full autonomy. In 1998, the Ilaje-dominated state government produced a new map of Ondo, which put the oil rich area in Ilaje territory and awarded its entailed revenue to them. This generated outrage among Ondo’s Ijaw and led to violence between the two groups (Folami Reference Folami2017). A contributing factor to the violence was that the Ilaje dominated not only Ondo state, but through their Yoruba kin also had representation in the central government, whereas the Ijaw were excluded (Damson Reference Damson2015). This exacerbated Ijaw grievances against the Ilaje and increased their distrust against central government attempts to mediate. For instance, there were Ijaw accusations that the Ilaje received outside support from Yoruba militias (Folami Reference Folami2017). To address this distrust, federal peace keepers arrived in September 1999 (Damson Reference Damson2015), whereby Ijaw and Ilaje soldiers were not allowed to partake. This trust-building measure intermittently stopped the violence, although it flared up again in 2003 (UCDP 2021). In 2007, Ijaw representatives joined their Yoruba peers in the national government (cf. Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015), following which no new communal violence incidents were recorded in Ondo (UCDP 2021).
Discussion
The three cases illustrate this study’s hypothesized mechanisms. In all cases, territorial autonomy alleviated grievances against the central government and reduced civil violence. However, disputes relating to territorial autonomy, and resources connected to it, generated subnational tensions. These tensions increased sharply following moves by second-order majorities to monopolize access to these resources: in 1972, Sindhi elites attempted to make their language Sindh’s only official language; following South Sudan’s independence in 2011, Dinka leaders monopolized economic resources for their own group in Jonglei; and, in 1998, Ondo’s state government sought to monopolize oil rents for the Ilaje.
In line with my conditional argument, these tensions were most prone to escalate into communal violence during uneven government inclusion. In such periods, included second-order minorities made far-reaching demands and engaged in violence to provoke central government intervention (Jonglei). Conversely, included second-order majorities became less conciliatory, which sparked violence driven by grievances (Sindh and Ondo). Communal violence subsided in two cases (Sindh from the late 1990s and Ondo from 2007), when the central government included representatives from both communal contestants.
However, the cases also highlight some limitations of my framework: First, besides the central government, other actors at the national level—such as Pakistan’s powerful, autonomous military—play an important role. Second, as for South Sudan’s Nuer, elite factions below the group level may not act in accordance with my simplified group-based model. Third, the central government may take alternative mitigation measures, as the Nigerian government did in 1999.
CONCLUSION
In this study, I developed a theory about the diverging consequences of territorial autonomy for ethnic violence at the national and subnational levels. I tested my expectations using new global data on territorial autonomy, regional boundaries, and ethnic demographics within each region. As my measures are based on de jure institutions, my findings highlight several important institutional recommendations.
Territorial autonomy remains one of the most widely propagated institutions to counter the risk of ethnically based civil wars (McGarry and O’Leary Reference McGarry, O’Leary and Noel2005). This rationale is supported and even extended by my results. I find that territorial autonomy reduces risks of rebellion not only among groups who control the regional government, but also among groups who remain excluded from it.
However, I find that territorial autonomy conversely generates tensions between ethnic groups at the subnational level, driven by unequal representation in regional government office and distribution of its entailed economic benefits. Based on a comprehensive cross-national sample, my findings demonstrate that whether such tensions escalate into communal violence is conditional on ethnic inclusiveness at the national level. Violent escalation is substantially more likely if central government representation is unequal. This complicates efforts to reach negotiated compromises and exacerbates mass grievances.
Where territorial autonomy is adopted as a peace-building tool, it hence needs to be combined with national-level power-sharing to address intra-regional tensions. This rationale is reflected in two of this article’s case studies: In Pakistan’s Sindh province, previously conflict-prone relations between Sindhi and Mohajirs improved following the latter’s renewed central government inclusion in 1999 (Adeney Reference Adeney2007; Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). In Nigeria’s Ondo state, violence between Ijaw and Ilaje subsided after both groups were simultaneously included in the federal government in 2007 (UCDP 2021; Vogt et al. Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015). Echoing these findings, Elfversson (Reference Elfversson2019) shows that, only following the creation of a multi-ethnic cabinet in 1999, was the Kenyan government able to end protracted communal violence between Marakwet and Pokot in Kerio Valley. Elfversson and Sjögren (Reference Elfversson and Sjögren2020) similarly show that Kenya’s 2010 national accord attenuated ethnic polarization and prevented renewed violence between Kikuyu and Kalenjin within Nakuru and Uasin Gishu counties.
A second important remedy highlighted in this article is subnational power-sharing, whereby groups commit to sharing power and economic resources in regional government (Van der Beken Reference Van der Beken2015). In some contexts, such arrangements have demonstrably helped defuse inter-ethnic tensions generated by territorial autonomy. Again focusing on Kenya, Elfversson and Sjögren (Reference Elfversson and Sjögren2020) show that local power-sharing reduced Kikuyu-Kalenjin polarization in Nakuru, by lowering the stakes of local elections. In a quantitative analysis of Nigeria, Bunte and Vinson (Reference Bunte and Vinson2016) provide systematic evidence that local power-sharing reduces inter-group violence, by generating elite incentives to appeal for cooperation and reducing perceptions that out-groups are threatening.
More research is needed on the conditions whereby territorial autonomy generates subnational tensions between ethnic groups and the circumstances under which these escalate into violence. Future research might seek to account more comprehensively for dynamic changes in the degree and form of autonomy (Lecours Reference Lecours2021) than has been possible in this study. Additionally, scholars should more systematically disentangle the mechanisms connecting territorial autonomy and subnational violence. This requires a more systematic and formal examination of the conditions under which non-violent bargaining over subnational government control and the distribution of economic resources fails, due to information and commitment problems. Such an endeavor would also profit from increased data collection efforts—for instance, information on which side initiated subnational ethnic violence (Hillesund Reference Hillesund2019). Finally, researchers might directly examine whether territorial autonomy generates grievances and initially non-violent mobilization among regionally marginalized groups that are directed against their more privileged regional peers.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055424000017.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/RKV5ZI.
ACKNOWLEDGMENTS
Earlier versions of this research were presented at the ECPR Joint Sessions (May 2021) and the Annual Conference of the European Political Science Association (June 2021). I am grateful to participants for their suggestions and feedback. Special thanks to Dennis Atzenhofer, Lars-Erik Cederman, Paola Galano Toro, Luc Girardin, Maria Murias Munoz, Yannick Pengl, Guy Schvitz, Roberto Valli, Yaron Weissberg, the editors, and four anonymous reviewers for valuable comments.
FUNDING STATEMENT
This research was generously supported by the ETH Zürich Postdoctoral Fellowship program (Award No. 20-1 FEL-27) and by the Swiss National Science Foundation (Grant No. 207820).
CONFLICT OF INTEREST
The author declares no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The author affirms that this research did not involve human subjects.
Open access
