Why Escalate?: Symbolic Territory and Strategy Choice in Conflicts Over Self-Determination

Abstract The quest for self-determination on the basis of shared identities motivates many subnational conflicts worldwide, but our understanding of the linkages between identity, contested land, and strategies is incomplete. What role do symbolic territorial attachments play for strategy choice in self-determination demands? I argue that groups tend to escalate their demands if they can exploit mobilization advantages of symbolic territory, creating identity-based territorial focal points for mobilization. I present new data on strategy choice in self-determination demands worldwide between 2005 and 2015. Using multinomial logit estimations, I find that symbolic territory contributes to limited degrees of escalation, and discourages armed conflict.


Introduction
The literature on subnational conflicts has developed compelling arguments to account for extensive violence, such as civil war or large-scale terrorist attacks, ranging from inequalities and state capacity to territorial concentration. 1 However, the characteristics defining the salience of the core conflict issue, the contested territory, is not fully recognized. Territorial characteristics are critical for mobilization and nation-building processes, which relate collective identities and national identification to territorial features. 2 While existing literature has addressed the linkage between identity and territory in the context of international and substate conflicts, 3 it has mostly focused on a single dimension of symbolic attachments, which does not reflect the complexity of intangible territorial value.
I contend that a more nuanced treatment of symbolic territorial value sheds new light on the strategy choice of self-determination groups, addressing the question of how symbolic territorial value affects the choice of conflict strategies by self-determination groups. I adopt the notion that groups choose the least costly strategy in seeking self-rule, which is often conventional politics, and resort to nonconventional strategies, such as protest and violence, when conventional political channels are not an option. 4 Which strategy they employ, I argue, hinges on territorial characteristics. I state that groups resort to nonconventional strategies when they value their land for symbolic reasons, which provides groups with mobilization advantages stemming from the identity relevance of symbolic territory. I thus expect that increasing degrees of escalation benefit from the mobilization advantage of symbolic territory.
The paper speaks to two larger debates. The first debate is concerned with the territorial sources of subnational conflict. While the relevance of strategic terrain and individual resources in civil wars has received significant attention, 5 the salience of prevalent symbolic attachments to the contested territory is much less understood. Measures such as length of residence or historical loss of autonomy 6 capture only part of the complexity of symbolic territorial attachments, disregarding at least two ways in which symbolic features matter: First, focusing on individual dimensions of symbolic value, such as homeland status or past autonomy, underestimates the complexity of symbolic attachments, by identifying a limited set of potential sources of symbolic territory. Second, these approaches theorize indirect links between conflict and territory, which do not fully reflect how territory directly affects mobilization. However, sources of emotional attachments are more diverse, and recent research demonstrates that the challenges to quantification can be overcome. 7 The second debate I speak to addresses strategic choices in subnational conflicts. Existing work shows that armed resistance is more likely with higher degrees of exclusion from state power. 8 Regionally concentrated groups benefit from mobilization advantages, higher capability and legitimacy, and are thus more likely to violently demand self-rule. 9 Rugged terrain provides safe havens for weak rebels, 10 while violent demands are more likely in high capacity environments and larger states, with more marginalized and peripheral groups. 11 Concessions encourage future separatist challenges, 12 while rebels use diversionary violence to secure territorial gains. 13 The literature on nonviolent subnational resistance campaigns assesses drivers of nonviolence and civil war, 14 linkages between state repression and nonviolent resistance, 15 diffusion effects from neighboring countries, 16 or organizational pressures fostering the adoption of terrorist tactics. 17 What remains unclear is how intangible, symbolic territorial features relate to conflict strategies.
I systematically investigate the role of symbolic territory in subnational self-determination groups' choice of conflict strategies. I present new data on the strategies of substate groups worldwide demanding self-determination from national governments between 2005 and 2015. Estimating multinomial logit models, I evaluate my expectation regarding the choice of nonconventional strategies conditional on symbolic territory. Distinguishing between limited escalation and armed conflict shows that while groups with symbolic attachments are more likely to use limited escalation, the absence thereof is associated with armed conflict. These findings are most robust for limited degrees of escalation. However, strategic decisions are conditional on the domestic and structural context. Larger, economically highly productive groups enjoy greater bargaining power vis-a-vis the government and tend to refrain from nonconventional strategies. These results are not conditional on material or strategic sources of territorial value, such as resource availability or terrain ruggedness. The findings are illustrated with the Kurdish demand for self-rule from Iraq. Taken together, my results highlight the relevance of disaggregating conflict strategies by showing variation in strategy choice in relation to symbolic territorial features.
Territory, self-determination, and conflict strategies I argue that self-determination groups are more likely to pursue their demand with nonconventional means if they value their land for symbolic reasons. I develop this argument by, first, elaborating on the relevance of symbolic territorial attachments in self-determination demands, and, second, linking these attachments to strategy choice through a mobilization argument.
I define self-determination groups as "a group of people that shares a collective identity and believes it has a legitimate right to self-rule." 18 This includes demands related to the exercise of authority in specific issue areas, such as language or taxation, and claims referring to the structure of governmental decision-making, such as regional parliaments.

Self-determination and territory
I contend that territory is highly salient in demands for self-rule for two reasons: First, territory is conceptually central to self-determination conflicts, due to the strong territorial component of both the demands for and the realization of almost all types of selfdetermination, ranging from autonomy rights to outright independence. Sovereignty and political power are executed over and involve the control of a distinct area of land and its population. 19 Second, territory matters for self-determination conflicts due to the mobilization advantages stemming from its relevance for the development of a robust and cohesive group identity. Territory is often related to specific events in the history of the group, perceived as homeland, and considered to be a critical component ensuring the survival of the group. 20 Much of the existing literature on self-determination demands has focused on ethnic features, but mobilization is often explicitly territorial and regional. 21 In mobilization processes, territory and the connections with the population settling it is constructed as a natural way to structure space. Direct and personal attachments to specific places are reinforced through "history, memory and myth." 22 The meaning of regional identity to individuals affects how territorial identities matter for political behavior. 23 How symbolic constructions of the region's past are translated into mobilization, however, is not fully understood. 24 The literature on international territorial conflict classifies territorial features along two general dimensions into, first, intrinsic value capturing tangible and concrete characteristics of the territory, such as strategic relevance, and, second, relational value including intangible and symbolic elements, such as identity ties. 25 In conflicts, tangible issues are easier to compromise, while intangible issues are more likely to be zerosum. 26 Subnational conflict literature has adapted this distinction, 27 but two issues require further attention. First, one strand of work focuses on a single aspect of symbolic territorial value, such as sacred spaces, 28 or autonomy loss. 29 This implies that groups value their land only along this single dimension, suggesting that other dimensions of symbolic attachments are not relevant or salient. While being highly valuable for uncovering the underlying mechanisms, this approach focuses our understanding of symbolic land on one dimension of a complex, multidimensional phenomenon. A second strand of literature jointly considers multiple empirical proxies for symbolic attachments, theorizing indirect links between territory and conflict. 30 While offering elegantly quantifiable concepts, these approaches cannot account for the translation of territorial features into conflict processes. To take an example, extended residence in a specific territory does not automatically translate into stronger attachments to the land, but might rather be associated with assimilation processes weakening symbolic links to the space. An additional step is required to convincingly connect the length of residence in a territory to its symbolic relevance.
My concept of symbolic value responds to these challenges. It refers to intangible territorial characteristics, such as historical autonomy or religious relevance, which are critical for identity formation and nationalist mobilization. Through merging natural and territorial features with existing identity narratives, 31 nationalism links the "material requirements of life and the emotional requirements of belonging" through symbolic territory. 32 By supplying both tangible and intangible sources of existence, access to territory ensures the security of the community. Many groups highly identify with "their" territory, so that "individual members have come to see the occupation or control of a territory as inseparable from their existence as nationals." 33 Territorial value is thus a culturally and politically diverse geopolitical discourse that generates frames for understanding the politics surrounding it. 34 I expand on existing work, which either focuses on individual components of symbolic value or theorizes indirect links, by assessing the relevance of various features of symbolic territorial value. I emphasize that territorial symbols have an objective foundation in the history and self-definition of the group, which link contested land and collective identity. As Smith 35 puts it, "nationalists cannot, and do not, create nations ex nihilo." References to symbolic territory always require a historical or otherwise empirical basis to gain traction in mobilization processes. Territoriality is a purposeful spatial strategy, "the attempt by an individual or group to influence, affect, or control objects, people, and relationships by delimiting and asserting control over a geographic area." 36 Recent research on international territorial conflict supports this notion. 37 Manekin, Grossman, and Mitts, 38 for instance, argue that territorial features are usefully conceptualized as political preferences rather than territorial characteristics. I acknowledge the relevance of value construction processes and stress that symbolic territorial attachments are at once factual and posses constructed meaning.
I argue that symbolic value directly links collective identities to the contested territory. Other sources of territorial value, such as material or strategic characteristics, are highly relevant in substate conflicts, but create indirect links between land and group. To take an example, externalities from resource extraction can generate grievances which in turn foster collective mobilization for self-rule. 39 Separatist mobilization in the oil-rich Niger delta is an example. Resources might also map onto existing identity and symbolic cleavages, and thus further strengthen self-determination demands. The discovery of the North Sea oil fields in the 1970s, for instance, gave leverage to the existing demand for Scottish autonomy. The limited ability of the state to provide public goods to remote territories might cause grievances toward central rule, while rugged terrain enables insurgencies by providing safe havens. 40 However, a significant share of the effect of ruggedness on civil war is mediated by political exclusion. 41 States emphasize sovereignty over the group territory based on strategic considerations of maintaining territorial integrity, where territorial control helps to ensure physical survival. 42 Deterring potential secessionists in multinational states avoids precedent setting but might lead to indivisible conflicts. 43 Group preferences deviate from state priorities: Controlling the settled land ensures the survival of language, customs, and religion, upholding collective identity. Ensuring territorial sovereignty over symbolic sites, such as historical battlefields or sacred places, provides continuity between past and present, and prevents devaluation of sites.
Symbolic territory and strategic choice I argue that groups valuing their land for symbolic reasons benefit from mobilization advantages, which make the employment of nonconventional strategies more likely. 44 Subnational conflict literature traces violent escalation to interactions of motivations and opportunity structures. 45 Civil war is more likely where groups are excluded from power, have higher mobilization capacity, and look back on a history of conflict. 46 Indiscriminate violence against initial protests and the availability of safe havens foster armed resistance of politically excluded groups. 47 State repression, however, increases protest particularly in contexts with civil war history. 48 In territorial civil war, rebels do not necessarily focus their attacks on their claimed homeland, but might engage in diversionary violence against civilians in order to distract government forces. 49 Self-determination groups rationally choose the strategy which is most likely to be successful at the lowest cost. 50 This includes opportunity costs from individual mobilization, casualties, resource investment, discontinuation of economic activity, or state repression. Groups have several strategic options at their disposal: Regular or conventional strategies involve political contestation in existing institutional channels, whereas irregular or nonconventional strategies operate outside of the institutionalized political process. I speak of conventional, or regular, strategies when "in a specific year, one or more organizations was actively making public demands for greater self-determination on behalf of the group," 51 including, for instance, election campaigns or lobbying activities. These are usually the least costly for two reasons: Mobilization relies on institutionalized channels of political action, and the state is less likely to impose costs of repression as the group adheres to existing rules. 52 If channels for political access are absent or not functional, the expected benefit of nonconventional strategies might outweigh their higher costs.
What role do symbolic attachments play in mobilization for self-determination? Symbolic value links identities to the features of the contested land. I contend that symbolic territory is a source of individual and collective identity, increasing the commitment of supporters and the perceived legitimacy of the claim. In nationalist mobilization, symbolic territorial features constitute and reflect collective identities: "[S]pecific places become synonymous with the society's rootedness there. As such, histories, memories and myths are effective means of binding people together and of binding them to specific territories." 53 Symbolic territory, I maintain, strengthens group identity and mobilization capacity, fostering conditions critical for collective action. 54 Focal points derived from territorial characteristics serve as clearly identifiable principles defining group territory, 55 which, in turn, allows for social monitoring and norm enforcement toward group members and leaders. References to territorial features thus strengthen group cohesion by increasing individual and collective identification with the land and ease mobilization by helping to overcome collective action problems.
Symbolic attachments create territorial belonging and generate greater legitimacy of demands. Where groups can make a credible argument that territory is the basis of their collective identity, essential to their history, their ideological and physical survival, selfgovernment claims gain more traction. Promoting territorial control based on symbolic value is thus a rational strategy from the perspective of the group. 56 I argue that symbolic territorial features are salient in self-determination demands along several dimensions: Sacred sites central to the collective belief system link religious identities to territory, potentially leading to indivisible and violent conflict. 57 Groups that were politically self-administered in the past or lost their autonomy have a motivation to organize and the capacity to do so, being in effect more likely to pursue separatist goals. 58 Attachments stemming from violations of landrights through forced resettlement or expulsion increase the perceived value of the land. The voluntary and self-funded return of Chechens years after their forced displacement to a territory already partly resettled illustrates this process 59 that might facilitate mobilization even for costly contestation strategies. Finally, cultural territorial identification addresses the close link between lifestyle and the territory that groups live off. Where identity and livelihood are threatened, groups might be more ready to escalate. Empirically, these four dimensions are often interlinked. I propose the following expectation to test their relevance for strategy choice: H1: Self-determination groups are more likely to employ nonconventional conflict strategies when they value their land for symbolic reasons compared to territory that is not symbolically valuable.
Even in a perfectly open and accessible system, however, conventional strategies might not be a realistic choice. Lack of resources for political organization or prior discursive commitment to a specific strategy might inhibit participation in institutionalized political processes. 60 Empirically, strategy choice does not therefore reflect a linear escalation process-starting out at party politics and culminating in civil war-but dynamically responds to opportunity structures. Moreover, choices are not isolated from each other over time and space, as diffusion effects might motivate the employment of specific strategies. 61

Illustration
The Kurdish self-determination movement in Northern Iraq exemplifies the mechanism. Scattered across Iraq, Turkey, Syria, and Iran, 62 Kurds constitute about 20% of the current Iraqi population. The rejection of the right to self-determination, initially granted in 1920 in the wake of the dissolution of the Ottoman Empire, determined much of the modern history of the Kurds. Their political status has been characterized by exclusion and repression, including Arabization strategies entailing ethnic cleansing, forced displacement and poisonous gas attacks, which have led to a series of uprisings. A regional administration was established in the early 1990s, 63 and enshrined in the 2005 constitution, eventually leading to the establishment of Kurdistan Autonomous Region. Kurds claim that many of the negotiated provisions of the 2005 constitution have been violated, and seek to strengthen their bargaining position toward Baghdad. A referendum in 2017 established strong support for an independent state.
The status of Kirkuk, located in the disputed territories wedged between Kurdistan region and the remainder of Iraq, illustrates the relevance of symbolic attachments. All of Kurdistan is considered indivisible homeland, and considered equally valuable. Kirkuk, specifically, has been described by senior politicians across party platforms as the "heart" or the "Jerusalem of Kurdistan," the control of which is considered nonnegotiable. While not carrying religious connotation in the narrow sense, these labels indicate the place of the city in public discourse, as well as the salience of territorial control: "Kirkuk has been, for at least half a century if not longer, the focal point of Kurdish nationalist aspirations in Iraq." 64 Arabization only increased the perceived value of the land: According to Human Rights Watch, "Kirkuk itself has become almost synonymous with the abusive Arabization campaign." 65 Legal reasoning, historical references as well as personal and collective sacrifices are invoked to establish territorial entitlement. For instance, former president Masud Barzani stated in an interview in 2004: "My father sacrificed himself and his revolution in 1974 for the sake of Kirkuk. If we should be forced to fight and lose everything we have accomplished we [still] would not bargain over Kirkuk's identity as the heart of Kurdistan." 66 The presence of oil fields-producing about 50% of the Kurdish oil revenue 67 -adds to the value of the territory for the Kurds as well as for both the Iraqi government and external actors. Arabization aimed at destroying the Kurdish majority in the North of Iraq, particularly in oil-rich cities such as Kirkuk, 68 facilitating Iraqi access to the revenue. The relevance of oil for Kurdish bargaining power thus critically depends on the relative strength of the government in Baghdad. 69 In sum, Kirkuk is highly valued not only as part of the indivisible homeland, but also for its historical relevance as a focal point in political repression and resistance against Iraqi domination.

Research design
My empirical assessment of the macro-level patterns of strategy choice in self-determination demands is based on the All-Minorities at Risk (A-MAR) list of subnational groups. 70 72 At least one faction of these groups demands self-rule, requesting autonomy or independence from the state. Much of the group-level data is territorially specific, and I geocode groups' settlement pattern based on the GeoEPR 73 which I extend for the groups that are not included. I elaborate on the coding procedure in supplemental appendix section 3. Throughout the paper, I make the pragmatic assumption that groups make demands to the land they inhabit. A list of all groups included, descriptive statistics, variable descriptions, and further details on the sources are provided in the supplemental appendix sections 1-3.

Conflict strategies
I present new time-variant data on the choice of conflict strategy in demands for selfrule for the period between 2005 and 2015, distinguishing between conventional politics and nonconventional strategies that target the state or state-related individuals and institutions conceived of as representing the state. While, in line with Cunningham, 74 I conceive of conventional strategies as the preferred strategy, as it is often associated with the lowest cost, I contend that the choices are not linear. Groups would usually not start out using conventional strategies, and, when unsuccessful, work their way up to armed conflict. I expect that the choice of different strategic options responds to varying opportunity structures, based on, for instance, institutional conditions, variation in symbolic attachments, or internal group composition.

Measurement
Conventional strategies limit contestation to political action within existing channels of political decision making, such as party politics or lobbying. Many groups in Western democracies employ conventional strategies, for instance, Quebecois in Canada or Scots in the UK. Conventional politics is, however, not a phenomenon which is limited to advanced Western democracies, but also common for instance with indigenous groups in Latin America or Kurds in Iraq, which to some extent operate within less responsive democratic decision making structures.
Groups employing largely nonconventional strategies choose contestation strategies outside of the established channels with different degrees of escalation. I distinguish two types of irregular politics: limited escalation and armed conflict. Limited escalation covers both fatal and non-fatal violence that does not systematically escalate, and ranges from violent protests and property destruction to limited rioting, sabotage and individual acts of terrorism. It includes cases such as violent protests by some indigenous groups, minor terrorist acts by Corsicans in France, or civil resistance by Saharawis in Western Sahara. Armed conflict involves at least 25 battle-related deaths in one calendar year, following the UCDP definition, which requires significantly different resources compared to limited escalation, such as weapons and personnel. Chechens in Russia and Assamese in India are among the groups engaged in armed nonconventional strategies in the period of analysis.
The rationale of the coding is to identify the strategy which is mostly used by actors and organizations representing the group for each year within the period of analysis. However, in some years, individual groups employ regular politics and irregular strategies to a largely balanced extent. I thus include an additional category for cases of mixed conflict strategies.

Sources
To a large extent, I base my coding on news resources available from the LexisNexis database (see supplemental appendix section 3). Not all demands for self-rule are reflected in news reporting in every year, but all group claims included in the analysis are active in the respective year. Out of a potential n ¼ 1,408, this leaves 1,103 observations of group-years referenced in news reports. Missing values signify either of two empirical patterns: Either the demand is indeed not active in the respective year, or there is no newswire reporting. Particularly in non-democracies, claims that do not escalate to the level of armed conflict do not raise the same degree of media attention and thus potentially suffer from underreporting bias. Active demands might not get reported due to restrictions to press freedom. For groups engaged in armed conflict in a given year, I rely on the UCDP Armed Conflict Database. 75 Empirical patterns I map the average choice of conflict strategies over all groups by country in Figure 1. The map shows that the incidence of demands for greater sovereignty is a global phenomenon. There is a tendency, however, toward irregular strategies in non-democracies, which are characterized by limited access to political decision-making processes. In contrast, in advanced democracies, self-determination groups on average employ conventional politics in the period under investigation.

Symbolic value
Symbolic value refers to the intangible features of the land that define its relevance in the perspective of the people settling the territory. Based on dimensions proposed in the nationalism literature, 76 my measure links the religious, politico-historical, as well as culture-and lifestyle-related aspects of the territory to the inhabiting group. Providing an objective measure of the subjective relevance of territory, I respond to two gaps in the existing literature: First, I do justice to critical sources of territorial identification by providing a multidimensional measure of territorial attachments that is not specific to distinct cultural norms but applicable across cultural traditions. Thus, my concept goes beyond existing approaches by addressing a variety of dimensions on which territory matters for substate contestants. Second, I offer a generalizable specification that allows for global comparative analysis over time, operationalizing direct links between the land and its relevance in political conflicts. My measure thus allows for testing the relation between territory and conflict without recourse to more easily quantifiable concepts that can only proxy emotional attachment.

Measurement
I collect information on the following four dimensions: First, cultural value is coded when territorial characteristics are critical for the cultural identity of the group, including nomadic or traditional lifestyles, and the presence of UNESCO World Heritage sites, such as the Kasubi Royal Tombs of the Baganda in Uganda. Second, religious value captures significant sacred sites, ranging from the Osun-Osogbo Sacred Grove of the Yoruba in Nigeria to the Harmandir Sahib, the Golden Temple of Punjabi Sikhs. Third, political value accounts for the historical relevance of an area, such as a historical kingdom or past territorial autonomy. This applies for instance to the Cabindans in Angola, who gained territorial autonomy with the Treaty of Simulambuco in 1885. Finally, landrights claims include cases where groups are dispossessed from their land or forcefully displaced, such as the Chechens who have been forcibly relocated in the Soviet Union in the 1940s. In order to counter potential endogeneity concerns, at least two of four dimensions are present in each case for the symbolic value indicator to be coded 1. I assign symbolic value to a case only if the characteristic appeared chronologically prior to the onset of the demand for self-rule. Examples illustrating the coding procedure are provided in section 4 in the supplemental appendix.
The intention of this study is to shed light on how symbolic value is related to strategy choice of self-determination groups, assessing the explanatory power of a measure that makes allowances for the limits of existing work. What I am interested in is, thus, the presence or absence of symbolic attachments of the group to the territory they claim. Measuring variation in territorial salience on each of the four dimensions might shed more light on the timing of strategy choices. Given the multidimensionality and complexity of the measure, however, the collection of time-varying data in the large-n is challenging. Moreover, many cases are not very well documented. My measure is thus limited to establishing an absolute level of symbolic salience of the land. I assess the relevance of the four dimensions of symbolic value individually as well as the simultaneity of the features as a measure of value intensity in the robustness checks. Considering challenges in data collection, assessing variation of this complex measure of symbolic attachments is limited to individual cases.

Sources
The coding of the four dimensions is based on a wealth of sources, in particular the MAR Risk Assessments 77 and the World Directory of Minorities and Indigenous Peoples. 78 Moreover, I rely on reports by human rights organizations and NGOs such as Amnesty International, Human Rights Watch, or the International Crisis Group, and on various case study articles from academic journals, academic working papers, as well as news reports. I include information from advocacy groups or blog posts only after cross checking with additional reliable sources. Furthermore, the UNCESO World Heritage website, encyclopedic sources, and country studies by the Library of Congress provide additional information. I counter potential reporting biases by crosschecking this range of sources.

Empirical patterns
The bar plot in Figure 2 summarizes the distribution of conflict strategies across symbolic territorial value. Many observations, i.e., group-state-years, are engaged in conventional politics, 55% of which have symbolic territorial attachments. This is not surprising: Regular strategies are attractive due to low relative costs, and, given the territorial character of self-determination demands, symbolic attachments are particularly salient in these kinds of claims. While limited escalation is more common where symbolic features are salient, armed conflict is more frequent in the absence of territorial attachments. In addition, conflict strategies are more commonly mixed for cases where symbolic value matters. This pattern provides a first glimpse at the variation across types of irregular strategies, and highlights the relevance of disaggregating conflict strategies in conjunction with territorial value.

Control variables
Structural features and group characteristics matter for the choice of conflict strategies. I include five control variables in the analysis, and describe the variables and expected mechanisms below.
Groups that constitute a larger share of the state population are harder to ignore in the domestic political arena, as overriding their interests may be politically risky for elected incumbents. They are also better able to exert pressure on the government by political organization: Nonviolent campaigns involve larger numbers of participants and feature lower participation thresholds than violent strategies. 79 Thus, the Ethiopian Oromo or Belgian Flemings are more difficult to ignore than small communities such as Russian Yakut or Indian Naga. I include the relative size of the group population as a percentage of the national population figure, expecting that larger groups are more likely to mobilize for conventional politics. Absolute group size is mainly based on reports by human rights organizations, newspaper articles, the Ethnologue database. 80 For the relative value, I use national population figures provided by the World Bank, 81 and subsequently log transform the term.
The economic productivity of a region in relation to the remainder of the state determines the group's bargaining power. They more easily assert their relevance in decision making, while governments are less willing to part with an economically highly productive region, which provides a reliable source of income. Many governments are thus willing to grant autonomy to relatively rich regions, 82 such as Spain to the Catalans, or the UK to the Scots. I expect economically powerful groups to be less likely to engage in nonconventional strategies. I operationalize the economic situation as the differential of logged regional economic development according to Nordhaus, 83 taken from PRIO-Grid, 84 and the logged national GDP per capita. 85 The degree of groups' political exclusion describes their ability to participate in decision making processes at the center and reflects their bargaining power vis-a-vis the state. Groups that participate in government have limited incentives to mobilize for self-determination. Political exclusion, on the other hand, may reflect grievances toward the government as well as incentives to mobilize that increase the chances of violence. I hypothesize that excluded groups are more likely to engage in armed conflict. I use the EPR measure of exclusion from access to power at the state level, which is coded 1 where the group is excluded and 0 otherwise. 86 As it requires some time for grievances from political exclusion to become politically manifest, I create a one year lag.
Finally, the variable democracy reflects the accessibility and accountability of governments. Democratic states are in general more responsive to concerns and grievances by their population. As groups can rely on functioning institutional channels, they are more likely to engage in conventional political strategies. 87 Authoritarian regimes, in contrast, often effectively block this route. I account for the host state's regime type by including a democracy variable, which I take from the Polity IV project. 88 The variable is coded 1 if the state has a value of five or larger on the Polity 2 scale, and lagged by 1 year.

Analysis
I employ multinomial logistic regression to estimate the relation between strategy choice, the dependent variable, and symbolic value as the main explanatory variable, setting conventional strategies, which are the least cost-intensive, as the baseline category. Conceptually, the strategic options do not represent ordered categories but are associated with different strategic opportunities and environments. Multinomial logit, therefore, is an appropriate modeling choice. 89 I cluster standard errors on the group level to control for interdependence across observations, and account for potential temporal dependence with a 1 year time lag where relevant.
I hypothesize that groups are more likely to employ nonconventional strategies when they value their land for symbolic reasons. To test this expectation, I distinguish between limited escalation, armed conflict, and mixed strategies to assess different incentive structures of self-determination groups. Compared to party politics, lobbying, or other forms of conventional strategies, irregular action and particularly violence involve a high degree of organizational investment, which is often not feasible for separatist groups. Groups, therefore, do not escalate step by step, but employ strategies depending on symbolic attachments and structural conditions.

Results
I report the results of multinomial logistic regression estimations with 95 and 99% confidence intervals in Figure 3 (see supplemental appendix Table A3, base model). The hypothesis predicts that self-determination groups are more likely to use nonconventional strategies when they attach symbolic value to their land. This prediction is confirmed for limited escalation, including irregular collective action that does not systematically escalate to large scale violence. For armed conflict, however, I do not find a statistically significant effect: Groups with symbolic attachments to their land are not significantly more likely to engage in armed conflict compared to conventional strategies. My expectation thus holds for limited forms of escalation, but not for armed conflict. It is thus crucial to distinguish between nonconventional conflict strategies of varying intensities to tease out the relevance of territorial attachments.
With regard to context conditions, I find that in regions that are better off economically in comparison to the remainder of the state, groups are less likely to choose irregular strategies. These claimants have higher bargaining power vis-a-vis the government, making conventional politics a more attractive option, and little interest in undermining their economic productivity. This holds for mixed conventional and nonconventional strategies as well. Moreover, in line with the extant literature, 90 groups operating in democracies have a lower likelihood to escalate their demand. This finding is only statistically significant for limited escalation, and does not translate into more severe forms of civil conflict. Figure 4 plots predictive margins of strategy choice for symbolic territorial value at the 99% level. Comparing the predictions for the choice of limited and armed escalation, respectively, vis-a-vis regular politics, I find that it is not so much the presence of territorial attachments, but rather the absence of symbolic value as operationalized here that makes a difference between limited escalation and armed conflict. While the predicted value of armed conflict is slightly higher than for limited escalation when it comes to symbolic value, self-determination groups without explicit attachments to their land have a much higher probability to systematically escalate their demand. At the same time, however, the probability to employ irregular strategies to a limited extent is lower for groups without symbolic attachments compared to those communities that assign value to their territory. Supplemental appendix Figure A1 shows that self-determination groups have a higher probability of being involved in armed conflict when they are politically excluded and internally divided. In sum, I thus find support for the hypothesis, but the relation between symbolic value and the choice of irregular conflict strategies is more complex than theory predicts.

Model robustness
Alternative sources of territorial value Symbolic attachments are only one way in which territory matters in conflict processes, and there is a rich literature assessing the tangible and intangible sources of territorial value in international and subnational contests. 91 A potential concern might be that symbolic attachments are constructed around alternative sources of territorial value, such as strategic or material issues, and not considering these sources might result in omitted variable bias. Substantively, as elaborated above, symbolic value is not plausibly constructed from scratch, and the proposed link between symbolic land and strategy choice is a direct one, whereas the effects of material and strategic features are mediated.
I use geospatial data on physical terrain features and resource deposits, which I connect to the groups' settlement areas, to substantiate this claim. Distance to the border measures strategic value, capital distance operationalizes remoteness, and ruggedness accounts for accessibility. Oil and diamonds measure resource availability. None of the variables is correlated with strategy choice or symbolic value to a significant extent (see supplemental appendix).
I add each of the variables individually to the base model 92 (supplemental appendix section 6.4) and confirm the core findings, most importantly the positive relation between symbolic value and limited escalation. Only the border distance measure reaches statistical significance, decreasing the likelihood of the small category of mixed strategies. All the other measures do not reach statistical significance, suggesting that endogeneity might be less of a concern. Further, I report models excluding cases with high correlation between symbolic and material or strategic value (supplemental appendix Figure A8). Notable changes only arise for the model excluding cases where material and symbolic value correlate. In this model, the symbolic value coefficient for limited escalation reaches a p-value of 0.019.
Alternative model specifications I further assess the validity of the findings by specifying alternative models (supplemental appendix section 6.1 and 6.2), including an extended set of covariates to account for other sources of heterogeneity. I find that more internally divided groups are more likely to escalate, which might be due to bargaining failures and defection of veto players. 93 In line with existing literature, groups excluded from political power in the preceding year are more likely to be involved in armed conflict. 94 Moreover, I estimate alternative specifications of symbolic territory, by, first, disaggregating the measure into its four components, and, second, modeling the simultaneity of symbolic features to proxy value intensity. Both specifications show that it is not a single component of symbolic value, but the interplay of multiple facets of emotional attachments that matter for strategy choice, and particularly limited escalation.

Selection effects
Groups' choices and constraints determine two observable outcomes: nonconventional strategies and armed conflict. Two stage selection models are well suited to account for this of nonrandom selection process. This requires the specification of an exclusion restriction, one or more variables that affect the selection process (Y 1 : nonconventional strategies) but not the final outcome (Y 2 : armed conflict). I include press freedom as exclusion restriction, arguing that groups operating in less open political environments have a harder time to engage in conventional politics. Armed conflict requires different resources independent of press freedom. I recode the Freedom House press freedom indicator into a categorical variable ("free," "partly free," "not free"), with higher values indicating greater press freedom. The Heckman probit model (supplemental appendix Table A5, model 3) confirms the earlier results, showing that symbolic value is statistically significant in the selection equation, and that the negative association with armed conflict holds in the outcome equation. The Wald test is not significant, suggesting that I cannot reject the null hypothesis of independent equations. This increases confidence in specifying two independent models (model 1 and 2). Consistent with earlier findings, predicted values from the outcome equation (supplemental Figure A7) show that conditional on the selection process, the probability of armed conflict is higher in the absence of symbolic value. For more details please see the supplemental appendix.

Discussion
How can we account for the empirical importance of the absence of symbolic value for armed conflict? The first possible explanation can be derived from the central argument regarding the salience of symbolic territory in easing collective mobilization. I maintain that both self-determination groups and their state opponents take the mobilization advantages of symbolic value into account: While groups benefit from its salience for identity creation, states anticipate that conflicts over intangible issues are more likely to escalate and more difficult to resolve. A potential explanation thus is that armed conflicts happen where symbolic value is not salient, and escalation motivated by territorial attachments is no feasible threat.
Alternatively, mobilization advantages strengthen the position of self-determination groups when negotiating with the state. We might observe a selection effect, where communities with strong symbolic attachments are better able to mobilize for violent collective action. As the government anticipates this advantage, accommodation is rational to avoid costly conflict. For the group, armed escalation is thus not necessary. Empirically, we should observe more concessions and less armed conflict when groups value their land for symbolic reasons.
The third potential explanation suggests that the international response might be instrumental to strategy choice. Coggins shows that separatists depend on international recognition when proclaiming independence. 95 They might thus be equally concerned about responses of the state and the international community. 96 Accordingly, the UN is more likely to get involved in self-determination disputes with a violent history. 97 International and domestic concerns might therefore jointly determine which strategies groups choose in order to maximize chances for concessions.

Conclusions
In this paper, I assess the relation between symbolic territorial value and the choice of conflict strategies in subnational demands for self-determination. Due to its direct relevance for the collective identity and the close linkages to territory as the core contested issue, symbolic value increases the mobilization capacity of the group. I hypothesize that groups are more likely to escalate when they harbor symbolic attachments to their territory. Using new data on the strategy choice of substate groups demanding self-rule between 2005 and 2015, I find that symbolic territorial value matters. It does so, however, in ways not fully anticipated by my theory: When symbolic value is salient, groups are more likely to employ nonconventional strategies, particularly limited escalation, compared to conventional politics. I consistently find that the absence of symbolic territory is conducive to armed violence.
Two additional issues require further attention. First, the analysis is based on a limited number of years. Demands for self-determination, however, often last long and movements sometimes undergo significant changes in organizational structure or opportunity structures. We might therefore observe variation in behavior that an analysis covering the complete history of each self-determination demand is better equipped to pick up.
The operationalization of symbolic territory is a second feature that merits further attention. I measure symbolic territorial value across four dimensions-political, cultural, religious, and landrights-issues-which I aggregate to a binary variable indicating that at least two of the four dimensions are salient in order to reduce the potential for measurement error. Variation in symbolic territorial attachments measures differences across cases, not within cases over time. The aim of this article is to establish the relation between this comprehensive concept of symbolic value and strategy choice. Many conditions relevant in subnational conflicts, such as the size of groups or geographic conditions do not vary much. We know, however, that the salience of symbolic attachments does vary. 98 Assessing the relevance of temporal variation in the salience of symbolic territory, likely in individual cases, remains for future work.
The analysis demonstrates the value of disaggregating conflict strategies in order to identify groups' incentive structures. This approach does not only allow for a qualitative distinction between different degrees of violence, but also makes it possible to trace the dynamics of escalation and de-escalation in more detail compared to existing work.

Notes on contributor
Friederike Luise Kelle is a postdoctoral researcher at University of Zurich. She has been a researcher at the WZB Berlin Social Science Center and a fellow at the Weatherhead Center for International Affairs at Harvard University. Her research focuses on separatism and sovereignty referendums.