1. Introduction
When the devastating civil war in Sri Lanka finally came to an end in 2009, the Sri Lankan government had been fighting the Tamil Tigers (LTTE) for over 25 years, over the course of which the latter became the main public face of Tamil opposition. Yet, the LTTE was by no means the only organization claiming to represent the Tamil ethnic group in Sri Lanka. For instance, the Eelam People's Revolutionary Liberation Front (EPRLF) and the Ceylon Workers’ Congress (CWC) similarly advocated an agenda of Tamil self-determination, but in contrast to the LTTE, they favored regional autonomy over outright secession (Pattanaik, Reference Pattanaik2014). These intra-group divisions have persisted to the present day, with some Tamil organizations making demands for increased access to central state power, while others continue to advocate for greater autonomy for the group. Similar internal disagreements have characterized most ethno-political movements, from the Civil Rights movement in the United States to the antagonism between the Zanzibar Organization and the Civic United Front in Zanzibar and the Kurdish mobilization in northern Iraq.Footnote 1
What explains this intra-group divergence in organizational demands with respect to a group's territorial status? An extensive body of scholarship documents the far-reaching consequences of internal divisions in (ethno-)political opposition movements. These range from the risk of conflict escalation (Cunningham et al., Reference Cunningham, Bakke and Seymour2012) and violence against civilians (Wood and Kathman, Reference Wood and Kathman2015) to conflict resolution and the likelihood of government concessions (Cunningham, Reference Cunningham2011), organizational longevity (Phillips, Reference Phillips2015), the achievement of movement goals (Krause, Reference Krause2014) as well as movements’ ability to enact broader social and political change if and once they gain power (Beissinger, Reference Beissinger2022, 359–416). In short, internal divisions within opposition movements multiply the front lines of conflict and, thus, decisively affect both the dynamics and outcomes of intra-state conflict. Yet, we know little about why such internal divisions appear in the first place, and why some ethnic movements are more divided than others.
Our paper addresses this gap by developing and empirically testing three pre-registered hypotheses on the structural sources of disagreements in ethnic group demands over the group's territorial status. We focus on territorial demands, specifically, because territorial conflicts make up the majority of ethnic civil conflicts in the post-World War II period and self-determination movements continue to challenge state governments globally (Butt, Reference Butt2017). Moreover, territorial conflicts are particularly critical as they affect international state borders, thus fomenting conflict in broader regions and potentially even disrupting the balance of the international system.
Much of the long-standing literature on territorial conflicts has been concerned with analyzing the initial emergence of self-determination movements, the ensuing state responses, and the escalation to violence (see e.g. Cunningham, Reference Cunningham2014; Griffiths, Reference Griffiths2015; Butt, Reference Butt2017). While this literature theorizes the conflict-fueling impact of fragmentation and intra-movement power struggles (e.g. Krause, Reference Krause2014; Cunningham et al., Reference Cunningham, Bakke and Seymour2012), it leaves mostly unexplained how fragmentation comes about in the first place. Those studies that do explore the sources of fragmentation tend to focus on organizational proliferation and splits only (Asal et al., Reference Asal, Brown and Dalton2012; Seymour et al., Reference Seymour, Bakke and Cunningham2016). Yet, arguments from the ethnic politics (Horowitz, Reference Horowitz1985, 359–416), social movement (Haines, Reference Haines1995), and international conflict literatures (Diehl, Reference Diehl1992) all point to the relevance of the actual issues and demands raised by actors—above and beyond organizational constellations. While organizational fragmentation can affect the nature and diversity of movement demands (Tokdemir et al., Reference Tokdemir, Sedashov, Ogutcu-Fu, Leon, Berkowitz and Akcinaroglu2021; Vogt et al., Reference Vogt, Gleditsch and Cederman2021), diversity of actors and diversity of demands often do not overlap neatly. Moreover, organizational fragmentation and splits can themselves be the result of preference heterogeneity (Seymour et al., Reference Seymour, Bakke and Cunningham2016).
Another strand of the literature, mainly anchored in bargaining theory, treats opposition demands as a function of available opportunities and actors’ mobilization capacity, which determine bargaining leverage (e.g. Jenne et al., Reference Jenne, Saideman and Lowe2007). From this perspective, demands themselves are little more than signaling devices that reflect the possible maximum concessions actors can obtain from the government (or other rivals). Yet, as emphasized by both constructivist research (Finnemore and Sikkink, Reference Finnemore and Sikkink2001) and social movement scholars (McAdam, Reference McAdam1999), political opportunities and mobilization capacity are influenced by actors’ subjective perceptions. Thus, different interpretations of the same reality often lead to “genuine” differences in both preferences and demands among political actors facing, in principle, the same opportunities and leveraging the same capacity.Footnote 2 Moreover, actors’ preferences do not solely reflect their motivations, but also their opportunities. Therefore, within-movement variation in actors' objective opportunities may equally result in “genuine” disagreements in preferences and demands. In short, we assume that, beyond strategic signaling, observed variation in territorial demands across actors within the same ethnic movement is at least partly due to underlying discrepancies in these actors’ “normative” preferences.
The existing empirical evidence on the far-reaching consequences of movement divisions, cited above, suggests that disagreement as such should matter, independent of the specific type of demands on which ethnic group representatives agree or disagree on. Such disagreement in terms of preferences has previously been analyzed more narrowly within self-determination movements only (e.g. Seymour et al., Reference Seymour, Bakke and Cunningham2016), thus excluding, ex-ante and by design, ethnic groups that do not raise such demands. However, the majority of politically mobilized ethnic groups—and, thus, of the universe of ethnic groups that could make self-determination demands—does not make such demands (Vogt et al., Reference Vogt, Gleditsch and Cederman2021, 1303). This implies that the absence of territorial claims is itself an integral part of the outcome to be analyzed. Therefore, our study assesses disagreement over group demands across ethnic organizations employing an integrated approach that considers the full spectrum of possible territorial policy preferences: from continued integration in a centralized state to regional autonomy (i.e., authority over a local administration vis-à-vis a national government) and outright secession (i.e., the withdrawal of a specific subnational territory from the existing state).
We focus on three within-group structural divisions to explain preference divergence within ethnic groups: heterogeneity in resources and economic activities, cultural divisions, and territorial fragmentation. We argue, first, that these structural factors shape both elites’ and ordinary group members’ expectations of the benefits of ethno-regional autonomy or secession compared to integration in a centralized state; and second, that within-group heterogeneity in these structural factors generates divergent preferences across different segments within the group, resulting in heterogeneous demands towards the government. For their part, state governments keen on undermining any centrifugal mobilization can actively attempt to exploit such within-group heterogeneity in structural conditions to their own advantage, fomenting divisions to counter (actual or expected) self-determination efforts.
Our pre-registered research design measures the degree of disagreement over territorial policies within “group-organization clusters,” which refer to one or more political organizations linked to the same ethnic group. For this purpose, we introduce an expanded version of the Ethnic Power Relations—Organizations (EPR-O) dataset (Vogt et al., Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015), which provides annual codings of the demands raised by various types of political organizations representing specific ethnic groups—ranging from Non-Governmental Organizations (NGOs) and political parties to armed organizations. We combine these data with existing, partly spatially-derived, indicators of groups’ resources and economic income streams, religious divisions, as well as geographic fragmentation. Our empirical results lend support to one of the three pre-registered hypotheses. We find that the presence of different income sources increases the degree of disagreement over self-determination across organizations representing a given ethnic group. Thus, our results shed new light on the structural sources of the movement-internal divisions that decisively affect not only key dynamics, but also the outcome of conflicts in multiethnic states.
2. Ethnic group divisions and intra-group disagreement over territorial status
In line with long-standing arguments in political sociology (Lipset and Rokkan, Reference Lipset and Rokkan1967), we assume a fundamental social and political cleavage between the state center and the periphery to be at the heart of territorial conflicts. However, going beyond this center-periphery dichotomy, peripheral populations, even within the same ethnic/ethno-regional group, are seldom homogeneous. They often hail from different cultural sub-groups, inhabit different geographic areas, represent different social strata and subscribe to diverging ideological convictions, engage in different economic activities, etc. They thus often have heterogeneous political interests and preferences, including those over the group's territorial status. While ethno-nationalist mobilization has become a defining feature of contemporary world politics, most politically mobilized ethnic groups do not make self-determination claims (Vogt et al., Reference Vogt, Gleditsch and Cederman2021, 1303). Moreover, where centrifugal demands do emerge, they seldom remain unchallenged, but are typically countered by opposing claims in favor of the status quo. Thus, we conceive of the full relevant spectrum of policy preferences on this territorial cleavage axis as ranging from no territorial claims (i.e., integration in the central state) on one end to secession on the other, with (continued or increased) regional autonomy within the existing state borders as a middle ground.Footnote 3
While self-determination demands are often framed in collective terms and refer to public goods, elites possess the necessary resources to publicly articulate and pursue them (Esteban and Ray, Reference Esteban and Ray2008). Yet, for specific group demands to become politically salient, elite interests also need to resonate to a certain extent with “ordinary” (i.e., non-elite) group members. More or less formal organizations usually serve as the vehicle of group mobilization as they encourage individual participation through the distribution of “private” benefits and rewards and/or clientelist relationships (Hechter et al., Reference Hechter, Friedman and Appelbaum1982). Hence, we expect ethnic groups’ territorial policy preferences to be “politicized” by ethno-political organizations claiming to represent these groups. This focus on organizations also provides a reasonable threshold of social relevance of a given policy position, as even small fringe organizations can have profound political ramifications (e.g. Haines, Reference Haines1995). Even if organizations receive external support—for example, from the state government—they should constitute relevant representatives of ethnic group preferences as long as their claims and membership are rooted in the corresponding ethnic group.Footnote 4 Indeed, as we explain below, our argument expects state governments to actively attempt to exploit existing heterogeneity within ethnic groups to their own advantage, lobbying particular segments of the peripheral population and/or co-opting parts of their elite, in order to counter (actual or expected) self-determination efforts.
Based on these premises, we argue, first, that ethno-regional preferences for or against—as well as the degree of—self-determination are a function of the expected benefits of ethno-regional autonomy or secession compared to integration in a centralized state and, as a consequence, that within-group differences in self-determination demands result from diverging preferences across different segments within the same group. If different segments diverge in their expectations of the benefits of self-determination, they will disagree in their preferences, resulting in heterogeneous demands towards the governments advanced by different organizations that claim to represent a particular ethnic group. Second, we theorize that such preference divergence originates from three possible within-group structural divisions: (i) heterogeneity in economic activities that generate regional wealth, (ii) cultural divisions, and (iii) territorial fragmentation. These structural factors shape both elites’ and ordinary group members’ expectations of the benefits of self-determination, and divisions along them can be exploited by state governments keen on undermining any centrifugal mobilization.
Economic activities, often the result of climatic and geological conditions, shape the policy preferences of those who engage in them and, thus, constitute a key driver of political claim-making (Auty, Reference Auty1995). Where distinct economic sectors, related to specific assets, diverge in their policy preferences, distinct interest groups will emerge to pressure state authorities (see e.g. Frieden, Reference Frieden1992). This should also be true for ethnic group preferences and claim-making with respect to the group's territorial status. Different rents, different requirements in terms of labor and capital, and different patterns of production and taxation all affect how group members engaged in specific economic activities perceive the relative benefits of the group's possible relationships to the central government.
At the elite level, a crucial consideration should be whether the benefits of specific economic activities can be reaped more or less easily within or outside the existing state. Scholarship on natural resource wealth has long emphasized the difference between point-source and diffuse resources (Boschini et al., Reference Boschini, Pettersson and Roine2007). Point-source resources, such as oil, diamonds, and other minerals, are extracted from a narrow geographic base. They are also highly valuable and can relatively easily be stored, transported, and converted to cash (i.e., they are highly fungible). As a consequence, they possess a high degree of “appropriability” (Boschini et al., Reference Boschini, Pettersson and Roine2007), allowing (state) elites to more or less immediately reap their benefits in the form of rents. Hence, elites from ethnically distinct regions with such point-source resources have incentives to seek the establishment of an independent state (or at least demand increased control over these resources) in order to capture the resulting rents (Collier and Hoeffler, Reference Collier and Hoeffler2006). At the same time, ordinary group members are often adversely affected by the environmental and social externalities of this concentrated form of natural resource exploitation, resulting in grievances among broader segments of the population (Aspinall, Reference Aspinall2007). This should make it easier for interested ethno-regional elites to mobilize the group's rank and file on behalf of their claims. Moreover, the “appropriability” of these point-source resources also raises expectations on the part of ordinary group members that increased group autonomy will entail immediate economic benefits, thus further promoting the alignment of elite and grassroots interests in favor of autonomy and secession demands.
By contrast, other economic income sources in a group's territory, such as agricultural production—a typical example of a diffuse resource—are more likely to lead to the diverging preference in favor of continued central state integration. Given that the rents from agricultural production are more difficult to capture, elites’ immediate benefits from increased territorial autonomy are lower. What is more, increased separation from the central state might incur significant costs on elites engaged in agricultural production, in the form of reduced access to the physical and institutional infrastructure provided by the central state government, including established access to foreign and domestic markets and the use of key transport infrastructure, such as roads, airports, ports, etc., outside of the group's settlement area. To be sure, many of these infrastructural and institutional obstacles should equally apply in the case of point-source resource production, but the latter's high “appropriability” endows regional elites with greater means to build the necessary extraction and transport infrastructure (e.g., by attracting foreign investors keen to get a share of the rents) and access alternative markets.
In addition, in contrast to natural resource exploitation, which is often controlled by large foreign companies, agricultural production tends to involve a larger set of smaller players. This allows governments to enact differential taxes across producers, favoring or discriminating specific (ethno-)regions (Kasara, Reference Kasara2007). In other words, governments should possess increased leeway in dealing with ethno-regional elites engaged in agriculture, providing opportunities for negotiation and/or co-option, tying selected elite factions more closely to the central state and thus sowing disagreement in self-determination preferences where other elite factions do favor increased self-determination.Footnote 5 As a consequence, where ethno-regional groups rely on different types of income sources, disagreements over self-determination are more likely. We summarize this expectation in the following hypothesis:
Hypothesis 1 (Resource Rents): The presence of diverse income sources, such as point-source natural resources and agriculture, in a group's settlement area increases disagreement over self-determination demands across ethnic group organizations.
Beyond economic incentives, intra-group cultural divisions—for example, along religious or linguistic lines—likely lead to diverging self-determination preferences. Again, such cultural divisions are often quite sticky, as cultural change is typically a slow-moving process, and thus not easily affected by contemporary political interventions (Bormann et al., Reference Bormann, Cederman and Vogt2017). In particular, we focus on intra-group religious divisions and argue that sectarian differences within ethno-regional groups increase disagreement over self-determination demands (Sarigil and Karakoc, Reference Sarigil and Karakoc2016).
Existing research highlights both the prevalence of religious discrimination and its important role in spurring group mobilization (Basedau et al., Reference Basedau, Pfeiffer and Vüllers2016). Importantly, different religious segments of an ethnic group may face different levels of discrimination (or even persecution) because of their religious denomination. While elites from discriminated religious segments should have distinct incentives to seek more independence from the central government, their calls for self-determination may also fall on particularly fertile ground among the rank-and-file members of the group who are exposed to state-imposed religious discrimination. By contrast, elites from more favored religious segments should prefer a closer relationship with the central state government, especially if their denomination aligns with the one of the ethnic group in power. Indeed, for instance, in Iran, state-favored Shiite Kurds exhibit lower levels of recruitment into Kurdish self-determination groups than Sunni Kurds (Tezcür and Asadzade, Reference Tezcür and Asadzade2019). All else equal, such intra-group variation in state treatment is more likely, the more religiously divided an ethnic/ethno-regional group is:
Hypothesis 2 (Religion): Sectarian differences within an ethnic group increase disagreement over self-determination demands across ethnic group organizations.
Finally, geographic conditions crucially shape center-periphery relationships and territorial conflicts. Indeed, previous studies indicate that the degree of ethnic groups’ territorial concentration significantly affects political mobilization (Weidmann, Reference Weidmann2009). While many ethnic groups inhabit a relatively clearly demarcated, cohesive territory, others are less geographically concentrated, with different parts of the group inhabiting distinct areas or one part physically disconnected from the rest of the group. Such enclaves often feature distinct, potentially competing elite factions with divergent preferences for or against self-determination as a function of their proximity to, and connections with, the state center.
For instance, in the case of a geographic and/or demographic asymmetry between a group's main settlement area and a smaller enclave, elites from the smaller, more isolated area might have less to gain from secession, and secession (or even autonomy) for the main group territory will incur significant costs on elites outside of the seceding (or autonomous) territory, especially if this results in a loss of influence at the national level in the original state. Again, state governments can exploit such territorial fragmentation by selectively co-opting specific regional elite factions and, thus, dividing ethno-regional mobilization. Furthermore, territorial fragmentation not only affects group elites, but also ordinary group members. New borders can impede trading (especially in the absence of a neighboring ethnic kin state), disrupt other economic activities and even force some parts of the population to move, leading to divergent preferences for or against self-determination.Footnote 6 This leads to our third hypothesis:
Hypothesis 3 (Geography): Territorial fragmentation of ethnic groups increases disagreement over self-determination demands across ethnic group organizations.
3. Data and methods
This section explains the measurement of our outcome of interest: disagreement in self-determination demands among ethnic group organizations, as well as our choice of model specification. We introduce the independent variables for each hypothesis separately in the respective results section later.
3.1 Measuring intra-group disagreement over territorial status
As explained above, we focus on the policy positions propagated by ethno-political organizations to determine within-group disagreement over the group's territorial status. Our main data source is the Ethnic Power Relations—Organizations (EPR-O) dataset. EPR-O identifies political organizations representing specific ethnic groups from the Ethnic Power Relations (EPR) dataset (Cederman et al., Reference Cederman, Wimmer and Min2010; Vogt et al., Reference Vogt, Bormann, Rüegger, Cederman, Hunziker and Girardin2015) and records their stated political goals (e.g., secession, governmental power, religious rights) and organizational activities (e.g., use of violence or electoral participation), based on primary and secondary sources. Our study relies on a substantially expanded version of the original EPR-O dataset (Vogt et al., Reference Vogt, Gleditsch and Cederman2021), which included a global sample of 40 countries covering the years from 1946 to 2013. Our expanded version includes 50 additional countries in the period from 2000 until 2019 while also extending the temporal coverage of the previously included countries to 2019. Overall, the expanded EPR-O dataset contains information on 1154 organizations linked to 345 ethnic groups in 90 countries. It is available at https://doi.org/10.7802/2767.
EPR-O codes organizations’ autonomy and secession demands across their lifetime on a yearly basis. Whereas secessionist demands refer to “the ethnic group's secession from the home country and/or the unification with an ethnic/national group living in another state” (Gremler et al., Reference Gremler, Weidmann and Vogt2020, 16), regional autonomy demands are coded if a given organization advocated for “a special political arrangement where the region gains authority with respect to the national government” (ibid.).
From these variables, we construct aggregate measures of disagreement over autonomy and secession demands by ethnic organizations linked to the same ethnic group (i.e., within a “group-organization cluster”).Footnote 7 In doing so, we assume that organizational demands are rank-ordered and can range from minimal to extensive with regard to their scope: Organizations can make no demands at all, demand regional autonomy, and/or demand outright secession from their state. For each organization we then record their maximum demand. For example, an organization demanding both autonomy and secession is treated equally to one demanding secession only. This leaves us with three categories of organizations: those whose maximum demand is status quo/none (0), autonomy (1), and secession (2).
With this categorization in mind, we calculate two measures of disagreement, which we illustrate using the example in Figure 1:
1. Disagreement (0/1): A dichotomous dummy variable indicating whether all organizations in a cluster agree (0) in their autonomy/secession demands, i.e., if all belong to the same category as defined above or whether there is disagreement (1). For instance, in Figure 1, Ethnic Group A would be coded with a value of 1 (disagreement), because its organizations make different demands on behalf of the group. In contrast, the organizations linked to Ethnic Group B both demand secession, resulting in a value of 0 (agreement) for this group.
2. Disagreement share: A continuous indicator of disagreement, defined as 1 subtracted by the maximum proportion of the organizations in each category over the three categories. For example, in Figure 1 the organizations linked to Group A all have different demands. The “share” of disagreement is thus 1 − 0.33 = 0.77.
Figure 1. Illustration of disagreement variables. Two ethnic groups (A and B) are represented by four ethnic organizations (A–D), each of which makes a particular demand. Note that according to the EPR-O coding rules, an organization can represent more than one group (such as Organization C in the figure).
Note that the second, continuous indicator can only take on a a limited set of values for clusters with a low number of organizations. For instance, for a group-organization cluster with two organizations the indicator will either be 0, 0.5, or 1. Still, these values are comparable across clusters. Appendix A provides summary statistics for our data.
3.2 Control variables
We introduce several control variables to that could confound the relationship between disagreement and our independent variables. From EPR-O, we compute the number of organizations linked to a specific ethnic group, as organizational fragmentation might affect movement demands (Tokdemir et al., Reference Tokdemir, Sedashov, Ogutcu-Fu, Leon, Berkowitz and Akcinaroglu2021; Vogt et al., Reference Vogt, Gleditsch and Cederman2021). We also control for the existence of ethnic organizations linked to more than one group (multiethnic organizations). Given their multiethnic nature, such organizations will rarely demand secession for one group while their existence might be more likely in the case of cross-group religious links (which in turn might be associated with intra-group religious fragmentation). Next, the presence of ethnic kin groups could introduce fragmentation as the latter can provide outside financing and will often espouse more radical policies (Seymour et al., Reference Seymour, Bakke and Cunningham2016). Consequently, we control for whether the ethnic group has one or more ethnic kin groups, according to the EPR dataset.
Furthermore, we include the group's population share in the country, as recorded in EPR. As groups become larger and possibly the majority in a country, secession demands could become less likely. At the same time, larger groups should be more internally heterogeneous. Existing regional autonomy arrangements likely shape organizations’ claims while the presence of natural resources as well as religious and territorial fragmentation can influence whether such autonomy has been granted in the first place. The same should apply to the incidence of violent conflict in the group region. We thus control for both variables in our analysis.
Finally, in terms of ethnic group inequality, we account for ethnic groups’ power status as recorded by EPR, as well as nightlight emissions in the group's settlement area as a measure of economic wealth (National Geophysical Data Center, 2014) calculated via PRIO-Grid cells. Given that the latter is not available for years prior to 1992, we only use it in the cross sectional analysis.
3.3 Estimation
We examine the determinants of intra-group disagreement drawing on our two alternative measures and using cross-sectional as well as time-series models. We opt for cross-sectional models in our main analysis, despite the fact that our preregistration specifies multilevel models as the main models. The reasons for this deviation from the preregistration are explained in Appendix W. Nevertheless, we report the results of the multilevel models in the appendix and briefly discuss their implications.
In the cross-sectional models, we focus on the period from 2000 to 2019 because codings for the newly included countries in EPR-O are limited to these years. We aggregate our explanatory variables using their mean values (in the case of continuous variables) and their mode (for nominal variables) during this period. All of these models include country fixed-effects and their standard errors are clustered on the country level. The time-series models leverage the full temporal range of our data. When using the binary disagreement measure as the outcome variable, we estimate linear probability models with two-way fixed effects. We refrain from estimating logistic regression in these analyses as this would lead to the exclusion of observations without variation in the dependent variable (i.e., group-organization clusters that always disagree or agree in their demands).Footnote 8 In the case of the continuous disagreement measure, we use linear two-way fixed effect models. Standard errors are again clustered by country in all these models.
4. Results
4.1 H1: Diverse income sources
Our first hypothesis expects diverse income sources, in particular point-source natural resources and agricultural production, to foster disagreement over the relationship with the state center across different elite factions. Our measurement relies on the ISAM-HYDE land use data from Meiyappan and Jain (Reference Meiyappan and Jain2012) and several sources that contain natural resource variables (gold: Balestri, Reference Balestri2012, gems: Lujala, Reference Lujala2009, diamonds: Gilmore et al., Reference Gilmore, Gleditsch, Lujala and Ketil Rod2005, petroleum: Lujala et al., Reference Lujala, Ketil Rod and Thieme2007). In terms of agricultural land use, the data from Meiyappan and Jain (Reference Meiyappan and Jain2012) is available in the EPR family of datasets. Based on the GeoEPR spatial dataset of group settlement regions (Wucherpfennig et al., Reference Wucherpfennig, Weidmann, Girardin, Cederman and Wimmer2011), it records the percentage of a group's settlement area covered by agricultural production. The variable is available for the years 1950, 1960, 1970, 1980, 1990, 2000, and 2010. We consider agricultural resource rents as present if at least 10 percent of the group's settlement area is used for agricultural production.Footnote 9 For natural resources, we again rely on the aggregated PRIO-cell data (Tollefsen et al., Reference Tollefsen, Strand and Buhaug2012) integrated in EPR. We use the sum of each resource type below, weighted across group territory as the basic unit for our operationalization. Natural resources are present if this index scores above 0, i.e. if any of the resources exist in the group's settlement area.
Next, we construct two variables to measure diverse income sources for each group-organization cluster:
• A dichotomous variable that takes the value of 1 if more than 10 percent (or other thresholds) of a group's land is used for agriculture and any natural resources (diamonds, gems, petroleum, gold) exist in the GeoEPR-settlement area of the group. If neither or only one of these conditions is met, the variable is coded as 0.
• A categorical variable indicating the number of different resources that exist on the group's territory (0: none, 1: a single resource; i.e., cash crops or natural resources, 2: both cash crops or natural resources).
Table 1 displays the cross-sectional models evaluating the relationship between different income sources and disagreement over self-determination demands (H1). Models 1 and 3 test the effect of a group territory featuring natural resources as well as agricultural rents, compared to groups with none or only one of the two income streams. Models 2 and 4 further distinguish between the latter two scenarios, including two separate independent variables for whether ethnic group territories feature neither income source and whether they feature only of the two income sources, respectively. Models 1 and 2 use the binary disagreement measure as the dependent variable, Models 3 and 4 the share of dissenting organizations (average between 2000–2019).
Table 1. Cross-sectional models testing H1
***p < 0.001; **p < 0.01; *p < 0.05; ⋅p < 0.1.
All models in Table 1 provide evidence in favor of our hypothesis. Models 1 and 3 predict a significantly higher value of disagreement and share of disagreement among organizations linked to the same ethnic group if both natural resources and agricultural revenue streams are present. Similarly, if neither relevant agricultural activity nor natural resources are observed in an ethnic group's territory, the disagreement decreases compared to ethnic groups with both income sources (Models 2 and 4). This also holds true if only one of those income sources is present rather than both.
Figure 2 plots predictions from Models 1 and 2 in Table 1. The results demonstrate a substantial increase in the share of disagreement within organization-group clusters if both natural resources and agricultural production are present. The plot on the left of Figure 2 displays an increase in the predicted value of the disagreement variable as organization-group clusters move to having both natural resources and agricultural.Footnote 10 The size of this effect is substantial: An increase from the minimum (0) to the maximum value (1) of our income streams variable predicts a 0.2 point increase in the value of disagreement. The effect plot on the right side shows a similar effect: If natural resources and agricultural production are present in a organization-group cluster (“both”), predicted values of disagreement are substantially higher than in constellations where one (“none)” or one income stream is present (“single).”
Figure 2. Marginal predictions of Table 1, Models 1 (left) and 2 (right), all other variables held at mean or mode.
Our robustness tests provide further evidence in favor of our expectations. Models with the linear two-way FE design (Appendix B, s. Section 3), logistic two-way FE regressions (Appendix C) as well as a linear two-way specification that exclude groups without variation in the dependent variable across time (Beck, Reference Beck2020, calls this “all-zero” groups) all indicate that diverse income streams are associated with higher levels of agreement. Finally, multilevel models with observations nested within years and countries (Appendix D) also support our conclusions.
However, an alternative explanation for these results could be that we are simply capturing additive effects of individual income sources. More precisely, natural resources and agricultural production in themselves may contribute to disagreement in self-determination demands, and the presence of both concurrently then solely measures their additive effect. In Appendix E, we introduce several models that compare the absence of all income sources to each constellation (only natural resources, only agricultural production, both). In all models, only the coefficients for both income sources at the same time exhibit a significant, positive relationship with disagreement on self-determination demands. This suggests that it is not individual income sources that matter for disagreement or self-determination, but their co-occurrence. In additional robustness tests, we thus replace agricultural production with the presence of pasture land from Meiyappan and Jain (Reference Meiyappan and Jain2012) as a proxy for livestock cultivation, which constitutes another diffuse resource (Lambin and Geist, Reference Lambin and Geist2003) (Appendix F) and apply different agricultural production thresholds (Appendix R). Again, the results provide support for our theoretical expectations. Overall, we interpret these findings as strong evidence in favor of H1: If an organization-group cluster relies on agricultural production and extraction of natural resources at the same time, this is associated with more disagreement on self-determination demands.
4.2 H2: Religious sectarianism
Our second hypothesis expects within-group religious fragmentation to increase the likelihood of disagreement over territorial preferences. To test this expectation, we construct three variables from the EPR-ED dataset, which provides information on the linguistic, religious, and somatic differences within ethnic groups included in EPR:
• A dummy variable that measures whether several religious segments exist (1) or a group overwhelmingly adheres to a single religion (0).
• The number of religious segments recorded in ED (min. 0, max. 3).
• A Herfindahl–Hirschman index (Alesina et al., Reference Alesina, Devleeschauwer, Easterly, Kurlat and Wacziarg2003) of the fragmentation in the religious composition of the group, based on the population shares of each religious denomination within a group. Higher values indicate more fractionalization.
Table 2 displays the cross-sectional models testing our expectations that religious fractionalization leads to divergence in self-determination demands (H2). Models 1–3 use the dichotomous disagreement measure as the dependent variable, Models 4–6 use the share of disagreement. The results do not provide support for our hypothesis. Across all models in Table 2, coefficients do not reach standard levels of statistical significance. The results of alternative model specifications (Appendix G–I) even point in the opposite direction: Some of the two-way FE specifications in Appendix H as well as the multilevel models in Appendix I indicate that more religious segments could be associated with a lower probability of disagreement. However, these multilevel models fail to converge, which is why we prefer the simpler fixed effects and cross-sectional specifications above. Overall, our results do not clearly support our expectations formulated in H2.
Table 2. Cross-sectional models testing H2
***p < 0.001; **p < 0.01; *p < 0.05; ⋅p < 0.1.
In Appendix J we additionally consider how the religions of a given ethnic group relate to that of the ethnic group in power, which constitutes a more direct test of the effect of religious discrimination or preferential treatment of a specific group segment on within-group disagreements over demands. These models provide some support for our hypothesis. If at least one religious segment of a group matches the largest religious segment of the ruling group in a given year, within-group religious fragmentation is associated with higher values of disagreement. Yet, this effect disappears when including information on whether a religion is favored by the incumbent government or not from Brown and James (Reference Brown and James2022).
Finally, our theoretical argument could be expanded to other within-group cultural cleavages that could foment disagreement over self-determination demands. Specifically, in Appendix S we evaluate the impact of linguistic within-group divisions, again relying on the EPR-ED dataset. We find some evidence that a higher number of linguistic segments within a group increase the probability of disagreement.Footnote 11
4.3 H3: Territorial fractionalization
Finally, to evaluate the impact of groups’ geographic fragmentation on disagreements over territorial demands, we rely on the GeoEPR-dataset (Wucherpfennig et al., Reference Wucherpfennig, Weidmann, Girardin, Cederman and Wimmer2011), which provides information on settlement patterns for each politically relevant group in EPR. Specifically, we construct three measures:
1. A dichotomous indicator of whether the group is fragmented (i.e., has more than one settlement area).
2. A count variable of non-neighboring settlement areas (non-adjacent polygons) for each group.
3. A Herfindahl–Hirschman index of groups’ territorial fragmentation, based on the population shares of each territorial cluster (as used in Weidmann, Reference Weidmann2009). Higher values indicate more territorial fractionalization.
Table 3 presents the results of the cross-sectional linear models testing H3. Models 1-3 use the dichotomous disagreement measure as the dependent variable, Models 4–6 the share of disagreement. None of these models support our expectation. While the coefficients tend to point in the expected direction, they do not remain statistically significant. The same is true for the two-way time series cross-sectional FE linear specification in Appendix K as well as the logistic and linear two-way FE specifications without “all-zero” observations (Appendix L). Finally, the results of some of the multilevel models in Appendix M indicate a negative effect of territorial fragmentation on disagreement. However, again, these models fail to converge which means that their findings have to be treated with caution. On the whole, we find little evidence for a systematic relationship between territorial fragmentation and within-group disagreement over territorial demands.
Table 3. Cross-sectional models testing H3
***p < 0.001; **p < 0.01; *p < 0.05; ⋅p < 0.1.
4.4 Additional robustness tests
In addition to the robustness tests for each hypothesis described above, the appendix considers a number of further potential issues concerning all hypotheses. First of all, while we operationalized the “integrationist” end point of the territorial demands spectrum with the absence of any territorial demands, in principle, Lipset and Rokkan's (1967) territorial cleavage axis could be interpreted as extending to explicit claims by (ethno-)political actors for control over the central state. Thus, in Appendix T, we re-estimate our models using an alternative disagreement indicator that includes organizational demands for group access to national government power, as recorded in our EPR-O dataset, as an additional category. Second, reversely, it could be argued that our measurement of disagreement is too fine-grained and, as a result, lumping together under a common term of “disagreement” what in reality are very different types of disagreement (e.g., equating disagreement over regional autonomy versus secession with disagreement over status quo versus secession). To address this concern, the models in Appendix V consider an additional alternative outcome variable that distinguishes only between two demand categories – centripetal demands (access to government power or no demands) and centrifugal demands (separatism or autonomy) – and computes disagreement among ethnic group organizations over these two basic categories.
Third, the inclusion criteria and sources used in EPR-O to identify relevant organizations entail a relatively high threshold for organizations’ inclusion in the dataset, thus eliminating socially irrelevant organizations (or organizations existing only “on paper”). Nevertheless, the political significance certainly varies across different organizations. Hence, in Appendix W, we computed another alternative disagreement measure that takes into account organizations’ relative relevance within a given group-organization cluster, by weighting the organizations representing the same ethnic group by their age as a proxy of institutionalization (Horowitz, Reference Horowitz2010). Finally, we re-estimated our models excluding different types of organizations from the analysis: (i) organizations with temporary ethnic claims (organizations whose group links change, e.g., by appealing to ethnic group issues during one electoral cycle but not in the next) (Appendix N); (ii) organizations linked to several ethnic groups, so-called “multiethnic organizations” (Appendix O); (iii) organizations for which EPR-O coders reported low confidence scores in their codings (e.g., due to lack of reliable sources), which might lead to an underreporting of autonomy and secession claims (Appendix P); and (iv) organizations with limited activities recorded in EPR-O (namely, organizations that have neither used violence or participated in elections), which might be NGO's that might not fit our theoretical framework (Appendix Q). Our results remain robust in all these additional analyses.
5. Conclusion
Ethnic movements continue to challenge state governments globally, with many ethnic conflicts revolving around the status of groups’ settlement territories. Most of these movements are internally divided, with different group representatives disagreeing not only over tactics and competing over within-movement hegemony, but also disagreeing over the very goals of the movement itself. These internal divisions, in turn, affect key dynamics, as well as the outcome, of conflict in multiethnic states—from the risk of violence and terrorism to government concessions and conflict resolution (Wood and Kathman, Reference Wood and Kathman2015; Seymour et al., Reference Seymour, Bakke and Cunningham2016).
We have argued that intra-group disagreements over self-determination demands are driven by differences among the elites and the ordinary population of a group in terms of the expected benefits of ethno-regional autonomy or secession compared to integration in a centralized state. Such divergences, in turn, are likely the result of three different structural sources: different resources and associated income streams, intra-group cultural divisions, and geographic fragmentation of the group's territory. Our analysis uses a new version of the EPR-Organizations dataset that extends the geographic and temporal coverage of the first version. The dataset is freely available and can be used to study many other research questions related to ethnic organizations.
In our analysis, we find support for the first of our hypotheses, which links disagreement on self-determination demands to the presence of different income streams. More precisely, if groups depend on agriculture but at the same rely on income from natural resources, the level of disagreement among the organizations increases. This finding sheds new light on the structural sources of the movement-internal divisions that decisively affect not only key dynamics, but also the outcomes of conflict in multiethnic states.
While our analysis does not explicitly study the effects of this disagreement on outcomes such as mobilization or conflict, our findings have important implications for this literature. If some of the predictors of intra-group heterogeneity (such as the presence of multiple income streams) are linked to conflict, our analysis suggests an additional pathway by which this can happen, which should be taken into account in future work. The study also serves as a call to arrive at a more encompassing view of (ethno-)political movements that incorporates a wide variety of demands and actors. Diverse types of political organizations linked to the same group often interact, compete, and communicate with each other. Thus, focusing exclusively on radical actors like self-determination organizations turns a blind eye to more moderate voices (e.g., of NGO's or parties) that can often hold great weight as well.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2024.33. To obtain replication material for this article, see https://doi.org/10.7910/DVN/CJISMY. The updated EPR-O data can be found at https://doi.org/10.7802/2767.
Acknowledgements
We acknowledge support from Lars-Erik Cederman and the ICR research group during the data collection for this project. The use of the GROWup tool at ETH Zurich (developed and supported by Luc Girardin) was instrumental in this effort. The authors acknowledge funding from the German Research Foundation (DFG) under Germany's Excellence Strategy –EXC-2035/1 – 390681379 as part of the project “Mobilizing Inequalities: From Grievances to Conflict.”
Conflict of interest
The authors declare that they have no competing interests.
Open access
