Research Collection

Doctoral Thesis

Fragile Fences: Border Change and Territorial Conflict

Author(s): Schvitz, Guy

Publication Date: 2019-09

Permanent Link: https://doi.org/10.3929/ethz-b-000440786

Rights / License: In Copyright - Non-Commercial Use Permitted

This page was generated automatically upon download from the ETH Zurich Research Collection. For more information please consult the Terms of use.

ETH Library DISS ETH NO 26341

Fragile Fences: Border Change and Territorial Conflict

A thesis submitted to attain the degree of DOCTOROF SCIENCESOF ETHZÜRICH (Dr. sc. ETH Zürich)

presented by Guy Schvitz M.A., ETH Zürich

born on 07.12.1986 Citizen of the Netherlands

accepted on the recommendation of Prof. Dr. Lars-Erik Cederman, ETH Zürich Prof. Dr. Nils B. Weidmann, University of Konstanz Prof. Dr. David B. Carter, Washington University in St. Louis 2019

Abstract

This dissertation contributes to an improved understanding of the causes and consequences of border change by examining the following two questions: Why have some borders remained more stable than others? And how does border change affect the risk of conflict between and within states? Building on a recent strand of the literature that treats borders as institutions, I develop a set of arguments that help explain when and where border changes occur and how these changes affect the risk of conflict. To test these arguments, I rely on a new geocoded dataset that maps state borders from 1886 to the present. The first part of the dissertation deals with the formation and change of international borders. I argue that most borders are the outcome of interstate bargaining, in which states compete over territory but aim to reach a stable settlement, as failing to do so would entail considerable uncertainty and a risk of incessant conflict. To find agreements, states commonly rely on established focal principles, such as the use of natural features, historical precedents or car- tographic referents as the basis for new borders. While each of these principles facilitates agreement, not all of them result in borders that are equally well- defined. Straight borders derived from cartographic lines are less recognizable on the ground than borders that follow clear-cut natural features of historical precedents. If the location of borders remains less clear to people on the ground, it is less likely that borders effectively coordinate interstate relations, which reduces the opportunity costs of redrawing them. In an analysis of border dura- bility since 1886, I find that straight border segments have been less durable than irregular ones, while borders that follow prominent natural features or historical precedents have been more likely to persist. In the second part, I examine the persistence of border instability in a first global analysis. Borders have lasting effects on individual behavior and iden- tities, which makes it increasingly difficult to redraw them, the longer they have remained in place. However, this also suggests that border instability may persist for long periods of time. In historically unstable regions, existing borders are less firmly institutionalized, while previous borders may effectively compete with current ones as the commonly accepted divide, which increases the risk of further instability. To test these arguments, I rely on geocoded data on interstate territorial disputes, and introduce a new dataset that maps the territories claimed by separatist movements since 1946. I find that regions with unstable borders prior to World War II have remained more prone to inter- state territorial disputes, domestic disputes and border change in the period thereafter. The third part adopts a broader regional perspective and examines the spillover effects of border change. Borders have been kept in place in part by international rules and norms aimed at reducing instability and uncertainty in international politics. However, these rules can and have been violated, and are not consistently enforced across time and space. Potential revisionist states base their choice to initiate a dispute in part on the observed behavior of other, proximate states. Successful border changes nearby demonstrate the weakness of international constraints, suggesting favorable conditions for revisionism. In a dyadic analysis of territorial changes and disputes since 1816, I find that territorial changes between one pair of states increase the risk of new disputes between other, neighboring states. This effect is largely driven by conflictual instances of territorial change, but is also mostly limited to the pre-World War II period. Zusammenfassung

Die vorliegende Arbeit trägt zu einem besseren Verständnis der Ursachen und Folgen von Grenzveränderungen bei, indem sie die folgenden beiden Fragen untersucht: Weshalb sind einige Grenzen stabiler geblieben als andere? Und wie wirken sich Grenzveränderungen auf das Risiko von Konflikten zwischen und innerhalb von Staaten aus? Ausgehend von neueren Studien, die Grenzen als Institutionen betrachten, entwickle ich eine Theorie die hilft zu erklären, wann und wo Grenzveränderungen auftreten und wie sich diese Veränderungen auf das Konfliktrisiko auswirken. Zur Prüfung meiner Hypothesen verlasse ich mich unter anderem auf einen neuen geokodierten Datensatz, welcher die Staatsgrenzen von 1886 bis heute abbildet. Der erste Teil der Arbeit befasst sich mit der Entstehung und Veränderung von Staatsgrenzen. Ich gehe von der Annahme aus, dass die meisten Grenzen das Ergebnis zwischenstaatlicher Verhandlungen sind, bei denen Staaten um Territorium konkurrieren. Gleichzeitig streben sie aber eine stabile Lösung an, da das Fehlen einer festgelegten Grenze erhebliche Unsicherheit mit sich bringen würde. Um eine Lösung zu erzielen, stützen sich Staaten in der Regel auf allgemeine Prinzipien als Grundlagen für neue Grenzen, wie die Verwen- dung von natürlichen Merkmalen, historischen Grenzen oder Längen- oder Breitengrade. Während jedes dieser Prinzipien eine Einigung erleichtert, führen sie nicht immer zu eindeutig festgelegten Grenzen. Gerade Grenzlinien, welche entlang bestimmter Längen- oder Breitengrade festgelegt wurden, sind vor Ort oft weniger klar erkennbar als Grenzen, welche Merkmalen wie Flüssen, Wasserscheiden oder historischen Grenzverläufen folgen. Wenn der genaue Grenzverlauf für die lokale Bevölkerung unklar bleibt, steigt die Gefahr von Grenzstreitigkeiten - und damit der Anreiz für Grenzveränderungen. Eine Analyse der Stabilität von Grenzen seit 1886 bestätigt, dass gerade Grenzen häufiger neu gezogen wurden als ungerade Grenzen, während Grenzen, welche natürlichen Merkmalen folgen stabiler geblieben sind als andere. Im zweiten Teil der Arbeit untersuche ich die Nachwirkungen von Gren- zveränderungen. Die Stabilität von Grenzen folgt einer pfadabhängigen Logik, mitunter weil Grenzen über die Zeit das Verhalten und die Indentität der Bevölkerung nachhaltig prägen. In der Folge wird es zunehmend schwieriger, bestehende Grenzen neu zu ziehen. Dies bedeutet aber auch, dass historische Grenzveränderungen eine Region nachhaltig destabilisieren können. In his- torisch instabilen Regionen sind bestehende Grenzen weniger stark verankert, während frühere Grenzen zum Gegenstand neuer Gebietsstreitigkeiten werden können. In solchen Regionen gibt es demnach eine erhöhte Gefahr für weitere Konflikte und Gebietsveränderungen. Zur Prüfung dieser Hypothesen stütze ich mich auf geokodierten Daten zu Gebietsstreitigkeiten zwischen und inner- halb von Staaten. Letzterer Datensatz wurde eigens für diese Analyse erstellt. Meine Analysen zeigen, dass Regionen, die vor dem Zweiten Weltkrieg häufige Grenzveränderungen erlebt haben, in der Nachkriegszeit deutlich anfälliger für Territorialkonflikte und Grenzveränderungen geblieben sind. Der dritte Teil der Arbeit befasst sich mit möglichen Ansteckungseffek- ten von Grenzveränderungen. Bestehende Grenzen verdanken ihre Stabilität unter anderem internationalen Regeln und Normen, die darauf abzielen, in- ternationale Instabilität zu verringern. Diese Regeln wurden aber in einigen Fällen verletzt, und werden bis heute nicht immer und überall konsequent durchgesetzt. Bei einzelnen Grenzveränderungen besteht die Gefahr einer Präzedenzwirkung, da sie weitere Staaten dazu ermutigen könnte, eine Neu- ziehung von Grenzen anzustreben. Eine dyadische Analyse von Gebietsverän- derungen und Territorialkonflikten seit 1816 zeigt, dass Gebietsveränderungen zwischen einem Staatenpaar das Risiko von neuen Gebietsstreitigkeiten zwis- chen anderen, benachbarten Staaten erhöhen. Dieser Effekt ist hauptsächlich auf konfliktträchtige Gebietsveränderungen zurückzufühern, ist aber auch grösstenteils auf die Zeit vor dem Zweiten Weltkrieg beschränkt. Acknowledgments

My time as a graduate student has flown by and has overall been an incredible experience. I am extremely grateful to everyone who has helped and supported me along the way. Most of all, I am indebted to Lars-Erik Cederman, who gave me the opportunity to write this dissertation and has provided so much advice, encouragement and guidance over the years. Nils Weidmann has been an excellent co-supervisor, and has offered vital feedback and comments at various stages of this project. David Carter was kind enough to mentor me during my research stay at Washington University in St. Louis, where I was able to develop large parts of this dissertation, and benefited from his expertise on borders and international relations. I am also grateful to Beth Simmons for inviting me to her interdisciplinary workshop on borders at the University of Pennsylvania, which has been a tremendous source of inspiration. My research stay abroad would not have been possible without the support of the Swiss National Science Foundation, which I gratefully acknowledge.

I am also happy to have worked with such great colleagues since my arrival at ETH. Seraina Rüegger has been a wonderful office-mate and collaborator with a great sense of humor and a never-ending supply of snacks. In the office next door, Luc Girardin has helped me tackle many technical issues and has helped me improve my computer skills. Across the hallway, Carl Mueller-Crepon and Yannick Pengl have contributed to my work with countless comments and suggestions, and have taught me many tricks of the trade. Micha Germann and Andreas Schädel generously shared the fruits of their data collection efforts with me, which has facilitated my own coding work. I would also like to thank my current and former colleagues Fran Villamil, Theresa Leimpek, Mirjam Hirzel, Manuel Vogt, Nils-Christian Bormann, Philipp Hunziker, Nora Keller, Kunaal Sharma, Govinda Clayton, Enzo Nussio, Allard Duursma, Owen Frazer, Va- lerie Sticher and Daniel Finnbogason for creating such a nice work environment.

Writing this dissertation would not have been possible without the support of my loved ones. In particular, I would like to thank my parents, my extended family and my friends for supporting me over the years, and for reminding me what is important in life. But most importantly, I want to thank Katherine Woolbright for always being there and for brightening up my days. I am so happy to spend my time with you and cannot wait to see what comes next.

Contents

List of Figures III

List of TablesV

1 Introduction1 1.1 Main Arguments ...... 3 1.2 Plan of the dissertation ...... 5

2 Borders and conflict in the literature7 2.1 Introduction ...... 7 2.2 Borders in political geography and beyond ...... 8 2.3 Borders and conflict in international relations ...... 13 2.4 Summary and contributions ...... 19

3 Main arguments 23 3.1 Introduction ...... 23 3.2 Definitions ...... 25 3.3 Borders in historical context ...... 26 3.4 Border drawing as a bargaining process ...... 30 3.5 The path dependence of international borders ...... 37 3.6 Rules, norms and precedents ...... 44 3.7 Summary and outlook ...... 46

4 Mapping the international system: The CShapes 2.0 dataset 47 4.1 Introduction ...... 47 4.2 Coding procedure ...... 47 4.3 Applications of the data ...... 54 4.4 Comparison to existing datasets ...... 60 4.5 Conclusion ...... 61

5 Defining the outlines: Border drawing and border durability 63 5.1 Introduction ...... 63

I CONTENTS

5.2 Explaining border durability and change ...... 65 5.3 Research design and data ...... 69 5.4 Results ...... 75 5.5 Discussion and conclusion ...... 82

6 Vicious cycles: The persistence of border instability 85 6.1 Introduction ...... 85 6.2 Border instability and its consequences ...... 86 6.3 Research design and data ...... 89 6.4 Results ...... 97 6.5 Conclusion ...... 105

7 Border change, spillover and new territorial disputes 107 7.1 Introduction ...... 107 7.2 Dangerous precedents ...... 109 7.3 The spillover effects of border change ...... 112 7.4 Research design and data ...... 114 7.5 Results ...... 126 7.6 Conclusion ...... 132

8 Conclusion 135 8.1 Summary and main contributions ...... 135 8.2 Limitations ...... 137 8.3 Directions for future research ...... 139 8.4 Policy implications ...... 140

Appendix 143 A.1 Descriptive statistics ...... 145 A.2 Robustness ...... 149 B.1 Descriptive statistics ...... 151 B.2 Robustness ...... 153 B.3 Creating voronoi cells ...... 158 B.4 Data: GeoSDM ...... 161 C.1 Descriptive statistics ...... 165 C.2 Results not shown in main text ...... 167 C.3 Robustness ...... 171 C.4 Data: Coding conflictual vs. cooperative changes ...... 172

References 175

II List of Figures

3.1 Territorial disputes and border drawing as a cycle...... 31 3.2 Focal points in border drawing ...... 35

4.1 Preview of the CShapes 2.0 dataset in 1886 ...... 48 4.2 States and dependencies over time ...... 51 4.3 Map of Europe in 1936 used to code country borders ...... 55 4.4 Preview of the data: Borders in Southeast Europe 1886-1946 . . 55 4.5 Territorial changes over time as recorded by CShapes 2.0 . . . . 57 4.6 Trends in state size ...... 57 4.7 Intersecting historical borders with ethnic settlement areas . . . 59 4.8 Estimated trends in ethnic fractionalization ...... 60 4.9 Comparing CShapes 0.6 and 2.0: Africa in 1946 ...... 61

5.1 From country polygons to border segments. Panels A and B: Borders before and after adjustments. C: Extracting boundary lines. D: Splitting lines into historical segments ...... 71 5.2 Border segment age in 2018 ...... 71 5.3 Calculating fractal dimension scores. A: Box counts across dif- ferent grid resolutions. B: Fractal dimensions of increasingly squiggly lines. C: Fractal dimensions scores for Mali in 2018 (100 km segments) ...... 74 5.4 Predicted survival curves ...... 78 5.5 Non-proportional hazards: Simulated first differences ...... 81

6.1 Maps of territorial disputes in South Asia since 1946 ...... 94 6.2 Equal-area Voronoi cells (50’000 km2)...... 94 6.3 Predicted probabilities of main outcome variables...... 102 6.4 Robustness: Varying cell sizes ...... 103

7.1 Two types of dyadic dependence ...... 118 7.2 Mapping dyadic dependence ...... 118 7.3 ICOW territorial disputes 1816-2001: Incidence and Onsets . . . 123

III LIST OF FIGURES

7.4 Conflictual and cooperative territorial changes over time . . . . 123 7.5 Territorial disputes by region ...... 124 7.6 Territorial dispute onset: differences in predicted probabilities . 131

A1 Correlation plot: Main explanatory variables ...... 146 A2 Average border straightness and alignment over time ...... 147 A3 Distribution of segment lengths across alternative datasets . . . 147 A4 Robustness: Varying segment lengths ...... 149 B1 Correlation plot: Main explanatory variables ...... 151 B2 Covariate balance after matching ...... 156 B3 Connectivity matrix used in spatial filtering ...... 157 B4 Voronoi Tessellation: Example ...... 158 B5 Voronoi Tessellation: Example (continued) ...... 159 B6 Cell sizes across alternative grid specifications ...... 160 B7 SDM: Dominant claims over time ...... 162 B8 GeoSDM: Preview of territorial claims ...... 164 C1 Territorial changes by region ...... 165 C2 Correlation plot: Main explanatory variables I ...... 166 C3 Correlation plot: Main explanatory variables II ...... 166

IV List of Tables

5.1 Cox PH models: Border durability ...... 76 5.2 Stratified Cox Models: Border durability ...... 80

6.1 Past border instability and interstate territorial disputes . . . . . 98 6.2 Past border instability and domestic territorial disputes . . . . . 99 6.3 Past border instability and border changes ...... 100

7.1 Logit models: Territorial dispute onset, 1816-2001 ...... 128 7.2 Logit models: Territorial dispute onset, 1816-2001 ...... 129

A1 Summary statistics: Chapter5...... 145 A2 Summary statistics: Chapter5 (standardized variables) . . . . . 145 A3 Grambsch-Thernau proportitionality test I ...... 148 A4 Grambsch-Thernau proportitionality test II ...... 148 A5 OLS models. Dependent variable: Border change ...... 149 B1 Summary statistics: Chapter 6 ...... 152 B2 Main results, alternative measures of border instability . . . . . 154 B3 Main results, regional sub-sample: Africa ...... 155 B4 Main results, regional sub-sample: World except Europe . . . . 155 B5 Main results, controlling for conflict persistence ...... 155 B6 Main results after matching ...... 156 B7 Main results after spatial filtering ...... 157 C1 Summary statistics: Chapter 7 ...... 165 C2 Logit models: Territorial dispute onset, 1816-1945 ...... 167 C3 Logit models: Territorial dispute onset, 1816-1945 ...... 168 C4 Logit models: Territorial dispute onset, 1946-2001 ...... 169 C5 Logit models: Territorial dispute onset, 1946-2001 ...... 170 C6 OLS models: Territorial dispute onset, 1816-2001 ...... 171 C7 OLS models: Territorial dispute onset, 1816-2001 ...... 171 C8 Coding conflictual vs. cooperative territorial changes ...... 173

V

Chapter 1

Introduction

We live in a world populated by territorial states, whose sovereignty is delimited by precise linear borders. Although it is easy to take the current landscape of states for granted, recent events have again shown that the world’s territorial divisions remain far from settled. In 2014, Russia annexed Crimea from Ukraine and has since supported secessionist groups in the country’s east that aim to break up the country even further. The United Kingdom’s 2016 vote to leave the European Union has raised fears that the reinstatement of a "hard border" between the Republic of Ireland and Northern Ireland could reignite old tensions and has boosted demands for Irish reunification and Scottish independence. In 2017, Catalonia unilaterally declared its independence from Spain, which has met fierce resistance from the central government, raising concerns of a potential violent escalation. Most recently, India’s decision to revoke the autonomy of Jammu and Kashmir has once again triggered serious tensions with neighboring Pakistan in the decades-old dispute over the region. The cases of Ukraine, Britain, Catalonia and Kashmir are not just rare exceptions, but rather are part of a larger trend, as existing borders are increasingly called into question. In many parts of the world, states continue to dispute sovereignty over territory. Recent years have also seen a steady rise of separatist movements, many of which aim to carve out their own states. Such attempts to redraw the political map have often provoked considerable violence, as illustrated by the examples of Ukraine and Kashmir. These ongoing disputes follow significant geopolitical upheaval in recent decades, most notably the collapse of the Soviet Union and Yugoslavia. The world map has changed substantially in recent years and may well continue to change in the near future. This dissertation contributes to an improved understanding of the causes and consequences of border change by examining the following two questions: Why have some borders remained more stable than others? And how does border change affect the risk of conflict between and within states? Although

1 1. Introduction the question of border change is frequently debated in the literature, it has rarely been examined empirically. Most research in political science has focused on relationships between states or has examined politics within them, often taking existing states and their borders for granted. There have been some exceptions to this overall trend. Most prominently, a number of studies on state formation and state size have have offered arguments that are also relevant when it comes to explaining border change (Friedman, 1977; Tilly, 1990; Alesina & Spolaore, 2003; Lake, 2003; Abramson, 2017). For example, a study by Lake and O’Mahony(2004) finds that historical changes in regime types help account for temporal variation in state size. Although undoubtedly important, this literature has focused largely on aggregate trends and systemic developments, while saying less about what causes individual border changes. Based on existing research therefore, it is difficult to know when and where border changes are most likely to occur. There has been more progress in the literature when it comes to unpacking the consequences of border change. In particular, a number of recent studies have examined borders at a local level, treating them as institutions that coor- dinate interstate relations. As such, recent studies have shown that the way in which new borders are drawn affects the risk of interstate conflict, as well- defined borders reduce the potential for disagreement or uncertainty about the limits of jurisdiction (Carter & Goemans, 2011, 2014; Goemans & Schultz, 2017). Another recent study shows that historical border instability can persist as border precedents make it easier for state leaders to claim territory and dispute existing borders (Abramson & Carter, 2016). Despite these important contribu- tions, a number of limitations remain. For one, recent studies are mostly limited to specific world regions, which raises questions about the external validity of their findings.1 Another limitation is that existing studies have focused entirely on the interstate domain and have yet to examine the relationship between border change and domestic conflict. There are good reasons to assume that border changes can also affect the potential for separatism, and therefore this particular relationship warrants further study. Lastly, while recent studies have examined the "local" consequences of border change, others have argued that border changes may destabilize borders on a larger regional scale by setting precedents for further attempts to redraw the map (Zartman, 1966; Touval, 1972; Fearon, 2004). Despite the importance of this claim, it has rarely been tested empirically. This dissertation aims to address each of these shortcomings, as explained in more detail in the following sections.

1One exception are Carter and Goemans(2011, 2014) who use a global list of secessions and territorial transfers in the 20th century.

2 1. Introduction

1.1 Main Arguments

To answer my research questions, I develop a set of arguments that help explain how and where borders are drawn, what accounts for their stability over time and how border changes can create further instability down the road. In a general sense, my arguments build on the idea that borders are institutions. Borders are not just lines on the map that separate state territories from each other but also serve to constrain and coordinate the behavior of states and individuals and are part of a larger legal and normative framework that makes up the modern international order. If borders are well-defined and mutually recognized, they facilitate the peaceful coexistence of states, reduce uncertainty in domestic and international politics and are likely to produce a range of political and economic benefits to states on both sides (Simmons, 2005). The view of borders as institutions raises the question of how they are created in the first place. I argue that most borders are the outcome of interstate bargaining, in which parties compete over territory but have a common interest in finding a stable settlement as failing to do so would entail considerable uncertainty and a risk of incessant conflict. To overcome basic obstacles in allocating territory and establishing borders, states commonly rely on a set of established focal principles, which guide border negotiations and are used to identify a limited set of outcomes. These principles include the use of natural features, historical precedents or cartographic referents as the basis for new borders (Schelling, 1980). While each of these principles may help states to overcome basic bargaining obstacles, not all of them help to ensure that the new border is well-defined and recognizable on the ground. In turn, if new borders remain poorly defined and their location remains unclear to people on the ground, it is less likely that they effectively coordinate relations between states, which reduces the opportunity costs of challenging and redrawing borders. Based on this logic, I argue that the use of some focal principles may result in borders that are more durable than others. Once borders are established, their redrawal may have destabilizing conse- quences. To anticipate these consequences, it is important to first consider what accounts for the overall stability of existing borders. To this end, I rely on two main explanations. The first is path dependence. I argue that stable borders decisively shape domestic politics and have lasting effects on the behavior and identities of people on the ground. Border consolidation follows the logic of increasing returns: The more people adapt to existing borders and organize their lives around them, the more attractive it becomes for others to do the same (Abramson & Carter, 2016). Therefore, the longer existing borders remain

3 1. Introduction in place, the more costly and difficult it is to replace them. The second expla- nation is more normative, and holds that borders have been kept in place in part by international rules and norms aimed at reducing instability. Over time, international rules have become increasingly restrictive, limiting the conditions under which border changes were seen as acceptable (Zacher, 2001; Goertz, Diehl, & Balas, 2016). According to this second explanation, state leaders have increasingly refrained from violating existing borders, as such actions became more costly on the international stage.

Following these explanations for border stability, the remaining part of my argument focuses on the consequences of border change. Based on the path- dependence argument, I argue that border instability can persist over long periods of time. In historically unstable regions, the future of current borders remains uncertain, which reduces their potential to generate joint benefits, and thereby lowers the opportunity costs of disputing and redrawing borders. In these contexts, previous borders may also continue to influence individual behavior and identities long after they have been removed and may therefore effectively compete with current borders as the commonly accepted divide. This makes border precedents a likely target for new revisionist and secessionist disputes. Territorial claims that follow precedents are easier to justify, and previous borders can easily be reinstated (Abramson & Carter, 2016). The existence of border precedents also enables leaders to retroactively construct nationalist claims about "lost territory" and historical injustices that serve as additional justifications and motivations for territorial change.

Based on the normative explanation, I argue that border changes also have the potential to destabilize borders elsewhere by encouraging other actors to challenge the status quo. While international rules generally impose strong constraints on territorial revisionism, these rules can be and have at times been violated. Moreover, the enforcement of international rules can vary across time and space, and is in many cases uncertain. The leaders of potential revisionist states base their behavior on what they assume they can get away with on the international stage. This assessment is based on recent interactions between other, proximate states that operate within the same regional context. As such, successful instances of border change between other states nearby can encourage other states to push for border adjustments, as they can be used to justify similar behavior and demonstrate the weakness of current constraints.

4 1. Introduction

1.2 Plan of the dissertation

This dissertation proceeds as follows: In Chapter2, I review the relevant litera- ture on borders and conflict in political geography and international relations and highlight the main contributions this dissertation aims to make. Chapter 3 develops the theoretical arguments to be tested in the subsequent chapters, which deal with the formation and change of international borders, the causes of border stability and the consequences of border change. To test my arguments, I make extensive use of Geographic Information Systems (GIS), which enable me to construct a range of spatial measures of border instability and examine their relationship with other spatial phenomena, such as territorial conflict. One major obstacle in the study of border change so far has been the lack of suitable data. To address this problem, Chapter4 introduces Cshapes 2.0, a new GIS dataset that maps international borders from 1886 to the present.2 The chapter provides an overview of the coding rules and data collection process and present a series of descriptive statistics and illustrative applications of the new dataset. In Chapter5, I rely on this new dataset to examine why some borders have remained more stable than others. To this end, I use information on the location and timing of border changes to trace the durability of border segments over time. I combine this information with indicators of border straightness, and their alignment with natural features and historical precedents. I generally find that straight borders have been less durable than irregular ones, while borders that follow rivers, watersheds and historical precedents have been more likely to remain in place than those that do not. Chapter6 examines the persistence of border instability within regions. Using grid cells as the unit of analysis, I develop several measures of border instability and examine their impact on three outcomes: territorial disputes between states, secessionist and irredentist disputes within states, and border change. As part of the analysis, I rely on GeoSDM, a new dataset that maps the areas claimed by separatist movements since 1946, based on the SDM dataset by Sambanis, Germann, and Schädel(2018). My general findings show that regions with unstable borders prior to World War II remained more prone to both types of territorial conflict and border change in the post-1946 period. A third analysis in Chapter7 examines the spillover effects of border change. I use dyadic data on territorial changes and interstate territorial disputes since 1816 and code a new variable that distinguishes conflictual from cooperative

2This dataset was created in collaboration with Luc Girardin, Seraina Rüegger, Lars-Erik Cederman, Nils Weidmann and Kristian Skrede Gleditsch, with the help of a team of research assistants.

5 1. Introduction exchanges of territory. I find that territorial changes between one pair of states increases the risk of new territorial disputes between other, neighboring states. This effect is largely driven by conflictual instances of territorial change but is mostly limited to the post-World War II period. Lastly, Chapter8 concludes the dissertation. I summarize my findings, highlight the main limitations of my work along with directions for future research, and discuss a number of policy implications.

6 Chapter 2

Borders and conflict in the literature

2.1 Introduction

Borders are central to our understanding of politics. They determine the size and shape of states, define the limits of state sovereignty, specify which rules apply on the ground, and identify the people that belong to each state, to name just a few of their main functions. Despite their obvious importance, borders have long received relatively little attention in political science. For the most part, political scientists have examined politics within states or have studied relationships between them, generally taking existing states for granted, and often treating their borders as a fixed quantity of little theoretical interest. A similar tendency exists in other disciplines. In their study on the size and shape of states, Alesina and Spolaore(2003, p. 2) criticize their fellow economists for having "taken the size of countries as ’exogenous’ - that is, not to be explained." The authors point out that borders, like all other institutions that shape human behavior, are the outcome of politics, which makes it crucial to understand the formation and change of international borders. There are a number of plausible explanations for the relative absence of border studies in political science. One is that international borders remained relatively stable following World War II, which made questions of border change seem less important than other, more immediate concerns, such as nuclear con- flict. Another is that in the late 20th century, a prevailing view held that borders were becoming increasingly obsolete as a result of globalization, technology and regional integration (e.g., Ohmae, 1991; O’Brien, 1992). Clearly, recent developments have shown that such hopes were misplaced. Most importantly, the collapse of the Soviet Union and Yugoslavia have reminded scholars that existing states and their borders cannot be taken for granted (Cederman, 1997). Moreover, as many borders remain disputed, we cannot rule out further in-

7 2. Borders and conflict in the literature stances of territorial fragmentation and reconfiguration. Partly in response to these developments, borders have increasingly been studied as an important subject in its own right. Most notably, a recent strand of the literature that treats borders as institutions has offered important new insights into the emergence and settlement of territorial disputes between states, and has the potential to improve our understanding of international relations more generally. This more recent body of research is also part of a much larger border literature that extends across disciplinary boundaries. Border studies have had a long tradition in political geography. In particular during the early 20th century, geographers have extensively debated how and where borders are drawn and what makes for effective boundaries. Although most of this literature is descriptive and large parts of it are seen as outdated, it has provided much of the conceptual foundations of the current literature, and has offered a number of important insights that remain relevant until today. Following these early contributions, a more recent literature that emerged in the 1990s has taken a more decidedly interdisciplinary approach to the study of borders. While this literature has paid less attention to border drawing and border change per se, it has provided key insights on other related issues, such as the relationship between borders, territory and national identities. In this literature review, I discuss key studies on borders and conflict in international relations in relation to the broader border literature in political geography and related disciplines. I begin with a brief overview of classical border studies and more recent interdisciplinary approaches, before discussing research on borders and conflict in international relations, while highlighting the main contributions and shortcomings of each strand of the literature. In the final section, I discuss four specific gaps in the recent border literature that I aim to address in the remaining chapters of this dissertation.

2.2 Borders in political geography and beyond

Classical border studies dealt with how and where borders are drawn, focusing primarily on classifying borders according to various criteria and considering the political implications of different types of borders. This literature emerged in the late 19th and the early 20th century, a time in which many new borders were drawn across Africa, the Middle East, and Asia, while the map of Europe were frequently redrawn as a result of war and the decline of empires. In this context, scholars sought to understand how states can gain and defend territory, but some were also motivated by a search for more durable territorial divides as a means to promote peace (Broek, 1943; Minghi, 1963). Important contributions

8 2. Borders and conflict in the literature came from scholars that did not only study borders, but actively shaped them as well, as they often served as colonial administrators, government advisors or were part of boundary commissions.1 Aside from efforts to describe and classify borders, scholars also debated the historical origins of borders as well as the evolution of borders and their local societal impact. An early debate revolved around the notion of "natural" boundaries, which were often contrasted with "artificial" ones. Friedrich Ratzel, commonly seen as the founder of political geography, developed an influential theory of the state as an organism that is in constant competition with other states over territory, for which he coined the term Lebensraum. A key part of this struggle was the search for effective, defensible boundaries. In this regard, Ratzel(1897) distinguished between various types of natural borders defined by coastlines, mountains and rivers, and man-made borders defined by walls or latitude lines, generally noting that borders established along natural barriers had important defensive advantages. Subsequent studies drew similar comparisons between natural and artificial boundaries, usually treating the former as superior to the latter (Minghi, 1963). Most prominently, Curzon(1907) argued that borders that followed natural divides had clear defensive advantages, while also noting that such borders were easier to recognize on the ground, and are therefore more likely to be accepted by states on both sides. Although the distinction between natural and artificial borders may seem intuitive, it was soon abandoned by most scholars, who argued that it was overly simplistic and misleading. This is worth noting, as recent studies in other disciplines have again focused on the differences between natural and artificial borders (e.g. Englebert, Tarango, & Carter, 2002; Alesina, Easterly, & Matuszeski, 2011), which geographers tend to view as problematic (Fall, 2010). Obviously, all borders are human creations and therefore artificial. Some critics equated the idea of natural borders with environmental determinism, as it implied that political divides were pre-defined by nature (Broek, 1943).2 Some also pointed to the problematic history and the subjective use of the term. Historically, states had invoked natural borders to justify their territorial claims on the grounds of natural law (Broek, 1943; Pounds, 1951; Rankin & Schofield, 2004). In other words: "A natural frontier is simply that particular natural feature beyond a state’s present boundary to which it’s leaders would like to

1For example, Curzon and Holdich were part of several boundary commissions in Asia and Latin America. 2This was indeed the view of philosophers such as Rousseau, Herder and Fichte, who argued that cultural differences between nations were largely the result of natural barriers between populations, which defined the "true" borders of nation-states (Pounds, 1954; Fall, 2010).

9 2. Borders and conflict in the literature expand." Over time, the notion of natural borders was further muddied by state-led attempts to define them based on language and culture (Pounds, 1954). This has also been a tendency within parts of the literature that defined borders as natural if they matched ethnic or racial divides (Curzon, 1907; Alesina et al., 2011), leading to accusations of essentialism (Fall, 2010). Perhaps most importantly, geographers soon agreed that borders based on natural features were not inherently better than artificial ones (Minghi, 1963). Some authors questioned the effectiveness of natural borders as defensive barriers in the modern age (Brigham, 1919; Fischer, 1949). Others argued that aside from their defensive purposes, borders also serve to promote the peaceful coexistence of states, which required them to be "obvious" and "indisputable" above all (Lyde, 1915; Brigham, 1919). Rather than classifying borders based on physical geography alone, others have introduced political typologies, distinguishing between borders that serve as points of contact and those that serve as points of separation, or classifying borders based on their duration or level of conflict (Muir, 1975; Prescott & Triggs, 2008). Most prominently, Hartshorne(1936) categorized borders based on their impact on the cultural landscape: Antecedent borders were drawn before the surrounding areas were inhabited, and societies were expected to adjust to them as they populated the area. Subsequent borders were drawn in already populated places, but followed existing cultural divides. Superimposed borders were also established in populated areas, but were drawn without regard for cultural differences. Lastly, relict borders have ceased to exist as international borders, but continue to influence the cultural landscape. According to this view, any set of borders is likely to have a profound influence on social and economic patterns in the surrounding area, and are therefore expected to become firmly established in the cultural landscape over time (Hartshorne, 1936; Fischer, 1949). In addition to its efforts to describe and classify borders, the early litera- ture has provided much of the conceptual groundwork for subsequent border studies. Starting with the work of Ratzel(1897), most scholars have defined borders as lines that exist within frontiers, which were defined as larger zones on the periphery of states. This distinction has also informed a macro-historical argument on the origins of borders, which describes a gradual transition from a pre-modern system of states separated by frontier zones to a modern system of states separated by linear borders (e.g., Pounds, 1951; Kristof, 1959; Prescott, 1978). Moreover, studies on the evolution of borders have made another key contribution by identifying three stages of border drawing, which range from the allocation of territory, to the delimitation of borders on the map, and their

10 2. Borders and conflict in the literature demarcation on the ground (Jones, 1945). Most research on borders was conducted in the first half of the 20th century, a period during which international borders underwent substantial change. In the period following World War II, the overall interest in borders declined. As one explanation for this trend, Minghi(1963, p. 408) notes that the literature "has tended to concentrate the research effort within times of change, while interest in boundaries during more ’normal’ times has not been so great." Another reason why research on borders was largely abandoned was due to its problematic history. For example, Ratzel’s notion of Lebensraum eventually became a core component of Nazi ideology, and similar writings by other geographers were often used as justifications for colonial expansionism (Smith, 1980; Prescott, 1978; Newman, 2002). Aside from its ideological baggage, classical border studies have been criti- cized as overly descriptive and non-theoretical (Muir, 1975; Newman & Paasi, 1998). For the most part, this seems accurate. As these studies were concerned with practical questions of border drawing, few studies have developed or tested general arguments on the political and social impact of borders. Some notable exceptions include case studies on disputed borders, border change and border demarcation issues, which have provided a number of fascinating insights (Brigham, 1919; Hartshorne, 1933, 1937; Moodie, 1943). In particular, some early studies offer detailed accounts of how new borders have disrupted established patterns of trade, how minorities on the "wrong" side of the border adapt to their new environment and how borders shape the everyday lives and identities of people on the ground, which are all highly relevant for the study of border change today. In general however, it is fair to say that the early literature has been overly concerned with describing and classifying borders at the expense of developing and testing more general theories. After a period of relative silence, a large new border literature emerged in the 1990s. The renewed interest in borders was again a result of major political developments, such as the collapse of the Soviet Union and the resurgence of ethno-nationalism, but was also driven by an increasing concern with border control in response to migration, transnational crime and terrorism (Newman & Paasi, 1998; Andreas, 2003; Van Houtum, 2005; Newman, 2006; Agnew, 2008). This second wave of border studies has adopted a decidedly interdisciplinary approach, examining borders from a diverse set of perspectives such as ge- ography, political science, anthropology, history, sociology and international law (Newman, 2006). Within the recent literature, a large number of studies have explored questions of borders and identity, and the relationship between borders and other social boundaries (e.g., Sahlins, 1989; Paasi, 1996; White,

11 2. Borders and conflict in the literature

2000). Other studies have instead focused on the question of border control and border politics, examining increasing attempts by states to police and regulate cross-border flows of goods and people (Andreas, 2003; Coleman, 2004; Gavrilis, 2008; Nicol, 2005; Carter & Poast, 2017; Simmons, 2019) Surprisingly, the question of border change itself has received less attention in the recent literature.3 Again, this may be partly explained by the relative stability of borders in recent decades, and by a growing political salience of border security, which has generally been viewed as a more important issue (Simmons, 2019). At the same time, the literature has generally moved away from the traditional focus on the location of borders, and instead have paid more attention to their changing role in today’s world, and on the broader social and political context of borders, often referred to as "borderlands". Within much of the recent literature, questions of border drawing and border change also remain associated with the classical border literature and its problematic history. For example, Newman(2006, p. 176) has pointed out that: "To use the terminologies of demarcation and delimitation today is to be branded, at best, a traditionalist or, at worst, a determinist." Moreover, large parts of the recent literature have taken a sharp constructivist turn, with scholars often asserting that each border is unique and should be studied in its own particular context, and that generalizations have little value (Jones, 1945; Prescott, 1978; Van Houtum, 2005; Prescott & Triggs, 2008; Paasi, 2016). Although constructivist and context-specific approaches certainly have their own value, this should not rule out the search for broader empirical regularities. There are good reasons to believe we can make useful generalizations about borders and their political and social impact. After all, borders shape the overall environment in which states and individuals interact, and we can therefore assume that differences in the nature of borders and changes in their location will affect these interactions in systematic ways, as classical border studies have already suggested. Although many of the ideas and arguments discussed in the early literature should indeed be seen as outdated and problematic, I argue that some of its insights remain relevant until today. In addition, although questions of border control and border management are undoubtedly important, recent history has also shown that borders still can and do change, while the causes and consequences of such changes remain poorly understood. Therefore, there is a clear need for more systematic research on when and where border change occurs, and what effects it may have.

3Research on territorial conflict in international relations constitutes an important exception, which I will return to below.

12 2. Borders and conflict in the literature

2.3 Borders and conflict in international relations

Research on borders and conflict in international relations initially developed separately from border studies in other disciplines. However, studies soon began to integrate insights from political geography and related fields. For the most part, studies have focused on the link between borders and interstate con- flict, treating borders as a proxy for interaction opportunities between states, or as dividing lines in territorial competition. More recent studies have shifted the debate by treating borders as institutions that structure international relations and create joint benefits in doing so. In this section, I review each strand of the literature in chronological order, after which I discuss a number of important research gaps.

2.3.1 Contiguity and conflict

In international relations, borders have long been examined as a measure of interaction opportunities between states. Some of the earliest statistical analyses on conflict showed that states that share a common border are more likely to fight. Richardson(1960) was the first to make this observation, furthermore showing that states with a larger number of neighbors were more likely to participate in large-scale wars. Subsequent studies confirmed these general findings (Diehl, 1985; Bremer, 1992), and further explored the links between contiguity and alliance politics and conflict diffusion (Siverson & Starr, 1991; Starr, 2013). The link between contiguity and conflict made sense from a realist perspec- tive, which views international politics as a perpetual power struggle among states. As Boulding(1964) had argued, states can more effectively project power in areas closer to their home base, which suggested that states are more capable of fighting their neighbors than remote adversaries. Building on this view, Starr(1978) argued that shared borders increase motives and opportunities for conflict, but also enable cooperative interactions. Still, the main assumption was that as neighboring states interact more frequently, it is more likely they will have something to fight over (Vasquez, 1995). As many have noted, one major limitation of research on contiguity and conflict is that shared borders may facilitate interactions, but cannot explain why wars emerge (Vasquez, 1995; Goertz & Diehl, 1992). To address this, subsequent studies have paid closer attention to the issues that states have gone to war over.

13 2. Borders and conflict in the literature

2.3.2 Territorial disputes and war

A key finding in research on conflict issues and their escalation is that territorial disputes are the leading cause of interstate conflict. Not only are disagree- ments over borders and territorial sovereignty more likely to trigger armed confrontations, they also increase their risk of escalation and recurrence, and generally lead to more destructive and protracted forms of conflict (Holsti, 1991; Vasquez, 1993; Kocs, 1995; Hensel, 2000; Vasquez & Henehan, 2001). As most territorial disputes arise between neighboring states, territory provides a more compelling explanation for why neighbors fight than contiguity alone (Vasquez, 1995). Most of the literature has examined the escalation and settlement of disputes (Huth, 1996; Vasquez & Henehan, 2001; Tir, 2006; Schultz, 2014), but some studies have examined their origins as well (Huth, 1996; Abramson & Carter, 2016; Goemans & Schultz, 2017).4 The question why territorial disputes often escalate has informed an exten- sive theoretical debate. Some studies have argued that territorial conflict is at least in part the result of an inherent human drive to claim and defend territory. Under certain conditions, human territoriality can give rise to violent conflict between competing groups of people (Ardrey, 1969; Vasquez, 1993; Johnson & Toft, 2014). This idea of course raises the question under what conditions this holds true, as the leap from "human nature" to political outcomes is quite large. In general, basic human traits are not very helpful when it comes to explaining the substantial temporal and regional variation in territorial conflict. In contrast, a more common approach defines territoriality not as a biological instinct, but as a deliberate strategy used to exercise control over people, resources and areas, which has taken on many different forms over the course of history (Sack, 1986). In general, studies that examine the value of disputed territories have proven more insightful. A key distinction was made between the material and sym- bolic value of territory (e.g., Newman, 1999; Hensel & Mitchell, 2005). The perceived value of territory helps explain why states seek to control some areas but not others, but also has key implications for the escalation and settlement of disputes. Whereas states should in principle be able to efficiently divide up a territory whose value is defined in strictly material terms, such a solution becomes much more difficult if the territory holds significant symbolic value. In particular, some territories are seen as part of the national homeland, and con- stitute an essential part of a group’s identity (Toft, 2003; Goemans, 2006; Shelef, 2015).5 An often cited example is the dispute between Israelis and Palestinians

4I return to these studies in the next section. 5In other instances, a territory may become viewed as indivisible if leaders tie their reputa-

14 2. Borders and conflict in the literature over Jerusalem, or dispute over Kashmir between India and Pakistan. In such instances, the territory at stake is perceived as indivisible by one or both sides of a dispute, which can make it extremely difficult to find a negotiated settlement (Hassner, 2003; Toft, 2003; Goddard, 2006). Indivisibility remains one of the most commonly accepted explanations for why territorial disputes often escalate. However, it may not necessarily pose an insurmountable obstacle to peaceful settlement. As Goddard(2006) argues, even if an issue is perceived as indivisible, parties may still reach a settlement if one side imposes its will on the other. Parties may choose to accept to divide “indivisible” territory if the costs of conflict are too high, or if diplomatic pressure from third parties is sufficiently high to produce an agreement. In addition, the perceived indivisibility of territory may also be partly the result of bargaining strategies and domestic politics, and therefore may be malleable. As parties negotiate themselves into a corner, they may also be able to negotiate their way out of it. In particular, generational changes in politics and the formation of new domestic coalitions as developments that may help parties to resolve previously intractable disputes (Goddard, 2006). Other recent studies have examined how nationalist claims over "homeland" territory may evolve over time. As Shelef(2015, p. 4) notes, homeland claims are specific type of territoriality that assert "that a particular group of people (the ’nation’) ought to control a particular territory because that land is part of who the people are." Due to their great symbolic importance, it is easy to see how homeland territories come to be seen as indivisible. However, it is important to note that ethnic and national homelands are not simply pre-determined by history or demographics, but are in part subjectively defined as part of a political process. As such, the definition of homeland that groups are willing to fight for may change in response to political changes, such as changes in ethnic geography and international borders and shifts in the distribution of power or domestic political dynamics (Shelef, 2015; Mylonas & Shelef, 2017). In addition to the large body of research on territorial disputes and their settlement, a few studies have also examined the consequences of territorial change as one of the outcomes of territorial disputes, focusing primarily on the value of the transferred territory and the circumstances of the exchange. For example, Weede(1973) provides evidence that territorial changes can provoke armed conflict between states, if the transferred territory is contiguous to both of them. Similarly Goertz and Diehl(1992) find that conflict is more likely fol- lowing an exchange of territory that is proximate to both sides, or is considered part of the homeland by the losing party. tion to controlling it, or if it has decisive strategic advantages (Hensel & Mitchell, 2005).

15 2. Borders and conflict in the literature

Overall, recent studies have shown that territorial disputes are associated with a range of negative outcomes, and have provided key insights into their initiation, escalation and settlement. While most studies on territorial disputes have not paid specific attention to the role of borders as such, they generally treat borders and territory as the object of struggle in a zero-sum competition between states (Simmons, 2005). This view again closely corresponds with a realist perspective on international relations, according to which states compete over territory in order to ensure their survival in an anarchic world. In this context, borders are just a temporary expression of state power, which are likely to change following changes in the distribution of power. However, as Simmons(2005) has pointed out, the conventional view of borders as zero-sum cannot account for three historical developments. First, interstate war over territory has continually declined in recent history, up to the point where conquest and annexation have become almost extinct. Second, most borders have remained stable in the period following World War II. Third, in many cases in which changes did occur, territory has often been transferred peacefully. Therefore, while conventional theories may help explain patterns of conflict and instability, they cannot account for recent trends in border stability. Already 70 years ago, Fischer(1949, p. 221) expressed a very similar critique:

[The] geopolitical idea of boundaries cannot be accepted. Its almost exclusive emphasis on strategic and power values is a serious draw- back. Its overemphasis on dynamic change obscures factors working for stability.

In short, therefore, a one-sided focus on warfare and territorial competition is insufficient when it comes to understanding border stability, border change and their effects on international politics.

2.3.3 The territorial order

In a seminal article, Zacher(2001) has offered one of the most convincing ex- planations for the decline of conflict and the general stability of international borders since 1946. According to this view, the growing acceptance and in- stitutionalization of territorial integrity norms have led most states to respect existing boundaries. Territorial integrity norms are best defined as the common understanding that territorial acquisitions through the use of force are illegit- imate. This effectively rules out conquest, annexation and violent secession, which were generally viewed as more acceptable until well into the early 20th century (Zacher, 2001; Fazal, 2011; Goertz et al., 2016; Hathaway & Shapiro,

16 2. Borders and conflict in the literature

2017). As territorial integrity norms have become more firmly established in international law, multi-party agreements and other international institutions, norm violations are not only seen as inappropriate, but have also become in- creasingly costly for the offending state. In particular the enforcement of norms by powerful states and the threat of economic sanctions have been central in de- terring would-be revisionist states from violating international borders (Goertz et al., 2016; Hathaway & Shapiro, 2017). While most studies have emphasized the benefits of international efforts to "freeze" existing borders, some have argued that these efforts may also have had adverse effects. Most importantly, norms against conquest have eliminated competition over territory, which has left leaders of politically weak states with less incentives to invest in state capacity (Herbst, 2000; Atzili, 2012). In some instances, this may have increased the risk of civil war and other sources of regional instability. Similarly, Fazal and Griffiths(2014) argue that norms against conquest have ensured the survival of small and weak states that would not have withstood international competition in earlier times, which has made secessionism an increasingly attractive option for minority groups. Ironically therefore, efforts to stabilize international borders may have created new sources of instability by creating more favorable conditions for secession. While the overall effects and effectiveness of territorial integrity norms re- main somewhat debated, international norms certainly provide a compelling explanation for recent historical trends, such as the decline of violence and the relative stability of borders. However, international norms alone cannot convincingly account for regional differences in border stability. While territo- rial disputes and border change have indeed become extremely rare in many parts of the world, other regions appear to be more prone to revisionism and secession. To explain such differences, it seems more fruitful to examine borders and interstate relations on a local, rather than a global level.

2.3.4 Borders as institutions

The most recent wave of studies takes a new approach that treats borders themselves as institutions that structure international politics. As noted above, Simmons(2005) criticized conventional approaches that view territory and borders entirely through the lens of zero-sum competition. Instead, borders can be sites of both conflict and cooperation, as geographers have often noted. According to this view, borders that are clearly defined and mutually accepted can serve as institutions that constrain and coordinate the behavior of states and individuals on the ground, reducing uncertainty across a range of political

17 2. Borders and conflict in the literature and economic dimensions. In doing so, stable borders can generate significant benefits to states and populations on both sides: They reduce uncertainty in matters of policy, jurisdiction, and property rights, and can help to minimize international barriers to trade (Simmons, 2005; Schultz, 2015a). In contrast, disputed borders are inherently disruptive to economic activity, due to an increased risk of violent conflict, a potential militarization of the border and a general uncertainty about the future status of territory. The potential benefits of border stability create strong incentives for states to settle their borders, even if they disagree about the distribution of territory (Simmons, 2005; Carter & Goemans, 2011; Huth, Croco, & Appel, 2013; Schultz, 2015a). More formally, the institutional view suggests that border drawing is not simply the result of zero-sum competition over territory, but instead is best understood as a mixed-motive game, which combines elements of both conflict and cooperation (Schelling, 1980; Simmons, 2005; Goemans, 2006). States may disagree about their territorial claims, but both sides stand to benefit from settling on a definitive boundary. In support of this view, Simmons(2005) shows that border disputes entail substantial opportunity costs, as they lead to significant losses in bilateral trade. Building on the institutional perspective, Carter and Goemans(2011) further- more show that the "design" of borders shapes their effectiveness as institutions, which in turn affects the risk of conflict between them. For borders to be effec- tive, they need to be clearly defined and well-known by all relevant actors. New borders that follow previous administrative borders meet these criteria, which should help to reduce the potential for friction. In their analyses, the authors compare new borders that follow administrative divides with those that lack administrative precedent, and generally show that the former are less conflict- prone and less disruptive to trade than the latter (Carter & Goemans, 2011, 2018). Based on a similar argument, Goemans and Schultz(2017) show that African borders that follow recognizable geographic features, such as rivers and watershed lines, have been less prone to disputes than other types of borders. Finally, a recent study by Abramson and Carter(2016) further develops the institutional framework by considering the role of path dependence. As the benefits of settled borders increase over time, individuals become more likely to coordinate their behavior around them, which makes it increasingly attractive for others to do the same. Moreover, there is some evidence that borders have lasting effects on individual behavior and regional patterns of trade, which can persist long after the border itself has been erased from the map (Fischer, 1949; Wolf, 2005). This makes historical border precedents an attractive target for territorial claims. Not only are claims based on historical precedent easier

18 2. Borders and conflict in the literature to justify, but areas within previous borders are easier to integrate into the state, as populations on the ground may continue to organize their lives around them. In support of this argument, the authors show that within Europe, areas with historically unstable boundaries have remained more prone to territorial disputes in the long run.

2.4 Summary and contributions

This literature review has covered a lot of ground, and it is therefore useful to take stock before outlining the main contributions of this dissertation. I have traced back the origins of recent border studies to a classical literature in political geography, which extensively discussed questions of border drawing and border change. While this literature is often criticized as overly descriptive and outdated, it still offers a number of important conceptual tools and insights that remain relevant until today. In particular, early discussions about the importance of drawing well-defined and recognizable borders, and case studies on the impact of borders on the cultural landscape speak directly to questions studied in the more recent border literature. Building on this earlier literature could therefore help to improve our overall understanding of borders and their political implications. Within the international relations literature on borders and conflict, the insti- tutional approach has offered a number of new insights that have furthered our overall understanding of territorial conflict. First, in arguing that stable borders can serve as institutions that provide joint benefits, recent studies have shifted the focus from the defensive properties of borders to their role in coordinating peaceful relations between states. Overall, this view has offered new explana- tions for why most borders have remained stable, and for why many territorial transfers have been peaceful or involved some form of compromise. Second, recent studies have shown that the institutional design of borders shapes the risk of conflict between states, as borders that follow previous administrative divides or prominent natural features can more effectively coordinate relations between them. Third, the recent literature has provided evidence that border instability can persist, as historical border precedents can become attractive targets for new territorial claims. While recent studies have made significant progress on several fronts, a number of gaps in the literature remain. In particu- lar, I identify the following four research gaps, which the following chapters aim to fill:

19 2. Borders and conflict in the literature

When and where do borders change?

Previous studies have shown that the institutional design of borders affects the risk of conflict between states (Carter & Goemans, 2011; Goemans & Schultz, 2017). Moreover, historical border precedents help explain where new terri- torial disputes emerge (Abramson & Carter, 2016). However, the causes of border change, and the conditions under which they occur remain less well understood. How are borders created in the first place? And what accounts for their stability over time? Since borders are institutions that are endogenous to politics, it is crucial to study the process of border drawing and border change in further detail (Alesina & Spolaore, 2003). Chapter5 contributes to this effort by examining when and where border change occurs. Building on previous work, I develop a general argument that helps explain how states choose their borders, and consider how these initial choices may affect the durability of borders over time. I then test these arguments using CShapes 2.0, a new geospatial dataset on international borders since 1886, which I introduce in Chapter4.

Need for a global analysis on the consequences of border change

Abramson and Carter(2016) provide compelling evidence that historical border instability is likely to persist, as historical precedents can provide state leaders with attractive targets for new territorial claims. However, as their study focuses on pre-1815 border instability in Europe, it is unclear to what extent their findings apply to the other parts of the world, such as Asia and Africa. Arguably, the European case could be seen as an extreme case, due to its long history of intense territorial competition. Moreover, because states in many parts of the world did not have clear-cut territorial boundaries until quite recently, it is less clear whether historical precedents have the same relevance in other parts of the world. To further examine the persistence of border instability, Chapter6 builds on this previous analysis to examine the link between pre- World War II border instability and post-1946 territorial disputes on a global scale, using the CShapes 2.0 dataset.

Moving beyond interstate conflict

Recent studies on borders and conflict have focused exclusively on the interstate domain. However, most territorial conflicts today arise within states, rather than between them. Although domestic conflicts over territory are clearly distinct from international ones, both revolve around the creation of new international borders. As such, we may assume that borders play a similarly important role in the emergence of domestic territorial conflict. For example, domestic groups

20 2. Borders and conflict in the literature that aim to secede may prefer to rely on historical border precedents to bolster their claims, similar to states involved in territorial disputes. The fact that recent research on borders has largely neglected the domestic sphere reflects a general tendency in the literature to treat international and domestic conflicts as two entirely separate phenomena. This strict separation has often been criticized as arbitrary and misleading, as the two types of conflict often overlap and share a range of common determinants (Lake, 2003; K. S. Gleditsch, Salehyan, & Schultz, 2008; Cunningham & Lemke, 2013). Therefore, the study of borders may constitute an important area of overlap that can produce useful insights. In Chapter6, I contribute to this effort by considering how historical border instability may create motives and opportunities for domestic groups to secede. To test these arguments, I rely on data from GeoSDM, an ongoing project that maps the territories claimed by secessionist and irredentist movements since 1946.

Examining the spillover effects of border change

While the institutional literature has examined borders as local, bilateral institu- tions that coordinate relations between neighboring states, other studies have treated borders as multilateral institutions, which are part of a larger frame- work of rules and norms that make up the modern international order. There is widespread agreement within this literature that international rules and norms have played a key role in stabilizing borders in recent history, especially in the post-1946 period (Zacher, 2001; Goertz et al., 2016; Hathaway & Shapiro, 2017). One question that has received less attention is what happens if existing status quo-oriented rules are violated. For example, how do annexations or land-grabs by individual states affect the risk of further instability? Although it is fre- quently argued that individual border changes could set dangerous precedents that may destabilize borders more generally, the conditions under which this holds true are not well understood, and the spillover effects of border change have rarely been examined empirically. In Chapter7, I consider under what conditions border changes in one location may destabilize borders elsewhere, and test my argument in a dyadic analysis of border changes and territorial disputes worldwide since 1816.

21 2. Borders and conflict in the literature

22 Chapter 3

Main arguments

3.1 Introduction

In the previous chapter, I have argued that the existing literature has paid insufficient attention to the causes of border change, has not taken into account its effects on domestic conflict and has yet to examine the potential spillover effects of border change. This chapter develops the theoretical foundations that will help to fill these gaps. I start with a brief overview of the main arguments and assumptions that I lay out over the course of this chapter. Building on previous work, I conceptualize borders as institutions that coordinate the behavior of states, and which are part of a larger legal and normative framework that makes up the modern international order. One key implication of this view is that borders are not just the object of interstate competition, but can also provide mutual benefits to neighboring states. Clearly defined and mutually recognized borders help states to maximize authority over their territory while avoiding constant friction with their neighbors and provide political and legal certainty from which states can draw substantial benefits (Simmons, 2005). The first part of my argument considers how borders are created to begin with. I argue that most borders are the outcome of a bargaining process where states compete over territory but have a shared interest in reaching a stable settlement. To find agreements, states commonly rely on a set of established focal principles such as the use of natural features, historical precedents or cartographic referents as the basis for new borders (Schelling, 1980). These principles enable state leaders to overcome basic bargaining problems, but may also help to ensure that the new borders are well defined and recognizable on the ground, which contributes to their effectiveness and durability as institutions (Carter & Goemans, 2011).

23 3. Main arguments

To better understand the consequences of border change, I first consider what accounts for border stability over time, drawing on two general explana- tions that have been discussed in the recent literature. The first explanation is path dependence. Borders have lasting effects on the behavior and identities of people on the ground and, as a result, it becomes increasingly difficult to re- draw them, the longer they have remained in place (Abramson & Carter, 2016). The second explanation holds that borders have been kept in place by a set of rules and norms aimed at reducing uncertainty and instability in international politics (Zacher, 2001). Over time, international rules have become increasingly conservative, reducing the scope for border change. From these explanations, I derive two general arguments on the conse- quences of border change. I argue that if border stability is path dependent, the same logic also suggests that border instability may persist over long periods of time. In regions where borders have frequently shifted, existing borders are less firmly institutionalized, while historical precedents generate motives and opportunities for the leaders of states and sub-national groups to challenge existing borders (Murphy, 1990; Abramson & Carter, 2016). This suggests that territorial disputes may often be part of a vicious cycle of border instability. Turning to the role of international rules and norms, I argue that border insta- bility may become contagious, as individual border changes can set precedents that can encourage actors elsewhere to seek further border adjustments. Interna- tional rules and norms generally impose strong constraints on revisionism, but their strength and effectiveness often remains uncertain. Would-be revisionist states base their choice to initiate a dispute in part on the observed behavior of other, proximate states. In this regard, successful border changes nearby demonstrate the weakness of international constraints, suggesting favorable conditions for revisionism. My arguments build largely on previous work and are therefore not entirely new. However, in further developing existing ideas, I make two theoretical con- tributions. First, while previous research on borders has focused mostly on the interstate domain, I develop an argument that links historical border instability and domestic conflict, which emphasizes the relationship between borders and national identities. Second, while it is commonly argued that individual border changes could destabilize borders on a larger regional scale, I develop a more specific argument on how such spillover effects are likely to arise, focusing on regional and temporal variation the enforcement of international rules and norms. The remaining parts of this chapter are structured as follows: I begin with the task of conceptualizing borders and then discuss today’s borders within

24 3. Main arguments a larger historical context. Next, I introduce the model of border drawing as a bargaining process, which provides the basis for my overall framework. I then lay out my main arguments and conclude with a brief summary and an overview of the remaining chapters.

3.2 Definitions

In order to discuss the causes and consequences of border change, it is key to precisely define what borders are. Following conventional approaches, we can define borders as lines of separation between states that specify where one state ends and the other begins. More specifically, borders are abstract legal concepts that determine the limits of state sovereignty, and are usually defined in bilateral or multilateral agreements and represented by lines on maps and by physical demarcations on the ground (Brownlie & Burns, 1979; M. Anderson, 1996; O’Leary, 2001). The definition of borders as lines is often contrasted with the notion of fron- tiers,1 which are larger zones on the periphery of states, in which state authority gradually fades out. Most pre-modern states had frontiers instead of linear borders. In these states, rulers typically projected their authority outward from a center of control, while the frontier zone was a mostly unclaimed, unregulated and often sparsely inhabited space in which the state was largely absent (Kristof, 1959; Branch, 2016). Unlike borders, frontiers lacked a clear spatial extent or a legal definition, and instead simply reflected the de-facto reach of the state (Kristof, 1959). Geographers have often described an evolutionary process, in which frontier zones were gradually replaced by linear borders, as a result of population growth, state expansion, and competition over territory (Ratzel, 1897; Curzon, 1907; Pounds, 1951; Prescott, 1978). Although linear borders have replaced frontier zones in dividing state sovereignty, the latter continue to exist today as distinct political and socio- economic zones that surround each border, which are often referred to as borderlands (Sahlins, 1989; Baud & Van Schendel, 1997). As recent studies have shown, frontier zones remain central to our understanding of questions related to border security, state-society relations and national identity, which has led many scholars to move away from a narrow understanding of borders as lines, defining them instead more broadly as "institutional zones" (Gavrilis, 2008) or

1There remains some confusion in the literature, as some studies use the terms interchange- ably, and many languages lack separate terms for the two concepts. However, the general convention is to distinguish border lines from frontier zones (Prescott, 1978; Rankin & Schofield, 2004).

25 3. Main arguments as part of larger "socio-territorial constructs" (Agnew, 2008). For the purpose of this dissertation, a narrower focus on borders as "lines on a map" still offers a good starting point. After all, border drawing and border change are primarily about dividing up land along precise territorial lines. At the same time, this narrow definition is insufficient, as it underplays the extent to which borders shape domestic and international politics. Fundamentally, the modern international order is based on the mutual recognition of territorial sovereignty, the notion that states hold exclusive authority over the space within their borders (Herz, 1957; Kratochwil, 1986; Ruggie, 1993). In this regard, a precise definition of the territorial extent of states enables them to exercise their authority without provoking constant disputes with their neighbors in doing so (Herz, 1957). In principle, borders establish a clear separation between the domestic and international sphere, and rule out any undue external interference into inter- nal affairs. This enables even weak states to isolate themselves from external pressures to some degree (Herbst, 2000) and allows them to establish their au- thority "inward" from their borders rather than solely "outward" from the center (Branch, 2013). In the domestic sphere, borders determine which rules apply on the ground and which people belong to which state. More generally, they define the space in which most other political, social and economic interactions take place. In general, borders constrain and enable the behavior of states and individu- als, while also shaping their incentives and expectations. They are therefore best understood as institutions, most commonly defined as "the rules of the game" or "humanly-devised constraints that structure political, social and economic interactions" (North, 1990, p. 3). Considering their far-reaching implications, it would not be an exaggeration to state that borders are among the most impor- tant institutions in modern politics. They can be understood both as bilateral institutions that define the distribution of territory and sovereignty between states, and as multilateral institutions, which are often part of multi-party agree- ments, and are embedded in a set of rules and norms that govern international politics. As I will argue below, both perspectives have their own implications for understanding the causes and consequences of border change.

3.3 Borders in historical context

Before introducing my theoretical arguments, it is important to discuss today’s borders within a larger historical context. This will be helpful in particular when considering the historical validity of my arguments. Although linear

26 3. Main arguments borders are often seen as a self-evident feature of international politics, they are a fairly recent phenomenon and are almost certainly a European invention. As late as the end of the 19th century, many parts of the world still lacked clearly defined borders that separated state territories from each other. Pre-modern states were often based on pre-territorial forms of rule and the creation of linear borders was an essential step in the formation of modern territorial states. Some have traced the origins of borders back to the Roman Empire, which maintained a hierarchy of internal administrative divisions, whose borders were often delimited by stone markers on the ground (M. Anderson, 1996; Pounds, 1951). The outer limits of the Roman Empire were approximately defined by a set of natural and man-made barriers intended to keep out barbarians, but these demarcations were not borders in a strict sense, as the authority of the Roman Empire often extended far beyond them (Kratochwil, 1986; M. Anderson, 1996; O’Leary, 2001). Throughout the Middle Ages, the catholic church maintained a similar hierarchy of territorial subdivisions, in which bishops and priests held exclusive authority (M. Anderson, 1996). For the most part however, feudalism gave rise to forms of rule that were less explicitly defined in space. Political rule was based mainly on personal allegiances, rather than on exclusive territorial control. Villages and local elites often owed allegiance to multiple rulers and as a result, political authority and jurisdiction often overlapped in space (Agnew, 1994; M. Anderson, 1996; Branch, 2013). In addition, the territories that were held by rulers were often non-contiguous and poorly defined. Typically, the outer limits of kingdoms were frontiers: mostly uninhabited and unclaimed stretches of land or areas in which local rulers were granted far-reaching au- tonomy in exchange for defending the kingdom (Curzon, 1907; Pounds, 1951). Over time, population growth, local disputes over the use of resources, state expansion and competition over territory created the need for more precise territorial boundaries, which eventually circumscribed state territories as a whole (Pounds, 1951; Prescott, 1978). Although some rulers already to establish linear boundaries as early as the Middle Ages, it should be noted that these borders generally had little meaning on the ground, as they were usually contradicted by overlapping allegiances and feudal rights (Pounds, 1951; Branch, 2013). It was only after the old feudal and religious order was replaced by a system of mutually exclusive, sovereign territorial states that borders gained the significance they have today. When exactly this shift took place remains debated. While the literature has traditionally treated the Peace of Westphalia (1648) as the birth of modern territorial states (Herz, 1957; Holsti, 1991; Ruggie, 1993), others have argued that the final shift took place much later, following the French Revolution (1789)

27 3. Main arguments or the Congress of Vienna (1816). According to this second view, Westphalia only marked the beginning of a long transitional period in which modern notions of sovereignty and territoriality gradually matured, before they became firmly established (Osiander, 2001; Branch, 2013). The reasons for the shift are also the subject of an interesting debate. For example, Branch(2013) argues that the creation of modern territorial states was in part driven by the invention of modern cartography, which favored the representation of states as discrete territorial entities rather than a set of separate locations, as they were previously understood. According to this view, the increasing use of maps changed common understandings of space among rulers and the wider population, which eventually gave rise to modern notions of territorial sovereignty.2 On the other hand, a more conventional reading of history holds that persistent competition over territory gave rise to increasingly centralized states that expanded outward and absorbed their smaller competitors, resulting in a system of large states with borders that were "well-defined" and "well-defended" (Tilly, 1990, p. 88). In any case, most of Europe and the Americas were divided up along precise linear borders by the early 19th century. In contrast, borders remained largely absent in Africa and large parts of Asia for almost another century. Herbst (2000) notes that in pre-colonial Africa, state power was not based on control over territory, which was abundant, but instead rested on control over people and labor, which were scarce. Because African rulers usually did not compete directly over the control of territory, they had no need for precisely delimited boundaries. In Asia, statehood often took on similar forms as in pre-modern Europe in that states projected their power outward from their center of control and their influence gradually faded out within the state’s frontiers (Branch, 2013; Goettlich, 2018). This changed drastically during the colonial period. In most of Africa and parts of the Middle East and Asia, European colonizers established new borders that consolidated their territorial claims. In most instances, borders were the result of negotiations among rival colonial powers and were often drawn with little regard for pre-existing political structures or cultural differences.3 In other cases, borders resulted from - often highly unequal - negotiations between colonial powers and local states that were able to maintain some form of autonomy or independence (M. Anderson, 1996).

2An alternative view holds that the Westphalian order created the need for precise borders, which could only be established once map-making techniques were sufficiently advanced (Shaw, 1997). 3The extent to which colonial borders were arbitrary remains debated as some have argued that some African borders were only drawn after extensive surveying and exploration, and were drawn based on cultural differences on the ground (Brownlie & Burns, 1979; Englebert et al., 2002).

28 3. Main arguments

As a result of this overall development, the Western model of statehood was exported to the remaining parts of the world. By the late 19th century, the use of linear borders had become firmly institutionalized and any attempts by states to claim authority within vaguely defined frontier zones or spheres of influence were no longer seen as legitimate (Goettlich, 2018). Overall, the early 20th century marked the end of a long transition from early forms of statehood to today’s system of territorial states. This transition was mostly completed in Europe and the Western Hemisphere by the early 19th century but took much longer in Africa, the Middle East and large parts of Asia. The replacement of poorly defined frontier zones by increasingly precise linear borders also coincided with the gradual adoption of international rules and norms that make up the modern international order. Most borders were established by bilateral or multilateral treaties, which followed the basic principle that treaties are binding and cannot be easily overturned (Kacowicz, 1994). Throughout most of 19th century Europe, states furthermore cooperated in attempts to preserve existing borders in attempts to uphold the balance of power (Holsti, 1991). At around the same time, newly independent states in Latin America committed to preserving their inherited colonial borders based on the principle of uti possidetis, in an effort to reduce the risk of subsequent conflict. The same principle was later applied during the decolonization of Africa and is today still used to regulate secessions, ensuring that new borders follow previous administrative divisions. In the post-1946 period, these basic principles were complemented by territorial integrity norms, which rule out the acquisition of territory by force (Zacher, 2001). Although international laws and norms have often been violated, there is widespread agreement that they have played an essential role in pacifying international relations (Shaw, 1997; Zacher, 2001; Huth et al., 2013; Fazal & Griffiths, 2014; Goertz et al., 2016; Hathaway & Shapiro, 2017). This brief historical overview gives us a better sense of the historical validity of my theoretical arguments. Most obviously, my focus on territorial conflict as a dispute over where borders are drawn makes sense only in a world where states themselves are defined by linear borders. As we have seen, this has not been the case until quite recently. In addition, the view of borders as institutions that structure politics applies mostly to the modern international order, where much of international politics rests on the mutual recognition of territorial sovereignty. From this principle flow a set of rules and norms that shape and constrain relations between states (Shaw, 1997). Absent these principles, the institutional view of borders may lose much of its meaning. In general, we can assume that the further back we move in history, the more the institutional

29 3. Main arguments framework loses its validity. It is difficult to draw a sharp line between the modern state system and its predecessors. Nevertheless, we can assume that most of the components of the modern order were in place by the early 19th century, which is also the starting point of my empirical analyses.

3.4 Border drawing as a bargaining process

At the beginning of this chapter, I have defined borders as institutions that structure international politics. But how are these institutions created to begin with? In this section, I discuss border drawing and border change as part of a bargaining process between states. Throughout most of the discussion, I focus on a simple scenario, in which the leaders of two states compete over the allocation of a given territory. Both sides make competing territorial claims and aim to establish a border that settles their dispute. The term "bargaining" requires some clarification. While some prominent models examine armed conflict as the alternative to bargaining, or the result of bargaining failures (e.g., Fearon, 1995), I follow others that treat conflict as part of a larger bargaining process in which parties may rely on both peaceful and violent means to reach their goals (e.g., Schelling, 1980). In order to increase their payoff in the eventual settlement, states make offers and counter-offers but may also issue threats and occasionally resort to violence. However, although the strategic use of violence is a common part of the bargaining process, I argue that it is unlikely that states will attempt to "fight it out" in a total war, in which one state attempts to annihilate the other and seize its entire territory. Although there have been several examples in which states have in fact attempted to do this, it has been an exception in the modern state system. In general, such actions have increasingly been viewed as impermissible, and have often entailed negative consequences for the offending state. Moreover, we can assume that state leaders are usually not indifferent to the costs of war, and will therefore aim to avoid an uncontrolled escalation. There are also good reasons to assume that leaders generally do not have an unlimited appetite for more territory, as not all territories are equally valuable to each state and enlarging the state can also come at the cost of governing an increasingly diverse and potentially hostile population (Alesina & Spolaore, 2003; Schultz & Goemans, 2019). In short therefore, we can assume that bargaining typically involves limited claims over specific territories and that state leaders will generally seek to limit the use of violence. A second assumption is that even if states compete over territory, they can draw substantial benefits from establishing a well-defined border. The absence

30 3. Main arguments of such a border implies a constant risk of conflict, which may erupt over the territory itself, but can also arise from uncertainty or disagreement over where one state’s sovereignty ends and the domain of another state begins. Moreover, the lack of a clear and mutually agreed upon border creates political and legal uncertainty that can be detrimental to economic activity (Simmons, 2005). In settling their borders, states therefore stand to benefit from increases in cross- border trade and international investment. We can thus assume that even in the case of territorial conflict, states have at least some incentive to ultimately find a stable settlement. Once the division of territory is agreed upon, states structure their relations around the new boundary. However, their leaders may continually choose be- tween maintaining the existing border or demanding a reallocation of territory. The process can therefore be seen as a cycle in which previous agreements may be upended by new disputes over territory, as illustrated by Figure 3.1. As I will argue below, a leader’s decision to dispute existing borders is shaped largely by the changing political and economic benefits of the status quo, and by the perceived international constraints on revisionism.

Claim Bargaining Settlement

Figure 3.1: Territorial disputes and border drawing as a cycle.

3.4.1 Where to draw the line? Focal points and bargaining

Border drawing is best understood as a "mixed-motive" game, which combines elements of conflict and cooperation. In negotiating over territory, states have conflicting preferences, but aim to coordinate on a specific outcome (Schelling, 1980; Simmons, 2005). A key task in this process is defining the range of potential outcomes. Which allocations of territory are on the table? Which may be seen as acceptable? The main challenge in bargaining is that it involves considerable uncertainty. State leaders do not exactly know the other side’s preferences and cannot fully anticipate its behavior. In addition, one could in principle draw an infinite number of borders to divide up any given territory, and leaders cannot anticipate the consequences of each possible outcome (Carter & Goemans, 2011; Huth et al., 2013). To overcome this challenge, leaders are likely to fall back on a set of general principles that are commonly recognized as the most "obvious" foundations for new international borders. These are what Schelling(1980) has referred

31 3. Main arguments to as focal points: prominent features that are collectively seen as the most plausible or salient options. Focal points are derived from general principles that are collectively recognized as common knowledge. These principles help players to coordinate on a limited set of outcomes even in the complete absence of communication, because they represent the most obvious reference points. For example, in the context of price negotiations, individuals are more likely to settle on round numbers than on any other price, mainly because round numbers stand out for their simplicity (Schelling, 1980; Pope, Pope, & Sydnor, 2015). In the context of border drawing, Schelling observed that states have tended to settle on borders based on historical precedents, latitude lines, rivers or other prominent features of the landscape, arguing that "these features seem less important for their practical convenience than for their power to crystallize agreement" (Schelling, 1980, p.68) . Settlements based on focal principles are often fairly arbitrary and are not always to the best advantage of either side but are crucial in overcoming coordi- nation problems. If two states divide up a contested territory along a river, they may not receive the full amount of territory they desire, but this is compensated by their ability to reach an agreement at all. Focal principles furthermore enable states to make highly specific claims that signal to their opponents that their ter- ritorial ambitions are in fact limited. This is crucial, as we can generally assume that states are more willing to negotiate with an opponent that makes narrow and specific demands as opposed to one that makes unspecified demands for "more" territory (Schultz & Goemans, 2019). Similarly, if states offer territorial concessions, fixing these concessions to a focal point allows them to signal the limits to their willingness to compromise (Schelling, 1980). In sum, the overall assumption is that states that compete over territory rely heavily on focal principles as these allow them to narrow down the range of potential agreements, reducing the risk of deadlock. As such, states have incentives to claim specific territories whose borders align with prominent features and they are similarly more likely to settle on a new border if it follows such features. This of course does not solve the problem of "competing focal points," as states have an obvious incentive to propose specific outcomes that closely match their preferences. At a minimum however, the use of focal principles allows states to move from an infinite number of potential claims and borders to a limited set of the most likely outcomes. Finally, another key feature of focal principles is that they can facilitate coordination on a new border after it has been established (Carter & Goemans, 2011). In order to effectively coordinate the behavior of states and individuals, the exact location of a border needs to be clear to both states and individuals on

32 3. Main arguments the ground. If a new border follows a river or some known historical precedent, there is much less room for disagreement or misunderstandings about where the border actually lies, and states do not have to invest much additional effort in surveying and demarcating the new boundary. This also appears to be one of the main reasons why 19th century geographers advocated the use of natural features as the basis for new international borders. While scholars have generally emphasized the defensive advantages of natural barriers, they have often placed an even greater emphasis on their importance in defining borders that are "obvious", "indisputable" and capable of "instant and easy recognition" (Holdich, 1899; Curzon, 1907; Lyde, 1915; Brigham, 1919). According to this view, well-defined and recognizable borders support the peaceful coexistence of states by clearly establishing the limits of each state’s sovereignty. Clear-cut borders further help to reduce uncertainty in international and domestic politics and thereby increase the potential for joint gains (Simmons, 2005; Goemans & Schultz, 2017). Therefore, if state leaders rely on focal principles that result in clear territorial lines, they have a greater incentive to preserve borders once they have been established.

3.4.2 Focal points in history

A cursory look at history indeed suggests that states have often relied on focal principles in attempts to settle their borders. In fact, it seems much more difficult to find cases in which they have not done so. One of the oldest and most established principles is the use of natural features as the basis for border agreements. Some of the earliest border agreements in medieval Europe were based on rivers (Pounds, 1951). By the 17th century, it had become common practice in Europe to claim territory and establish borders based on rivers, mountains and other prominent natural features (Pounds, 1951, 1954; Sahlins, 1989; Goemans, 2006). As borders in Asia and Latin America were defined with increasing precision from the 18th century onwards, local states and colonial powers also widely relied on natural features in border agreements (Prescott & Triggs, 2008). Even in 19th and 20th century Africa, colonial powers often attempted to define borders based on rivers, mountains and the like, as evidenced by the negotiation texts and border agreements at the time (Brownlie & Burns, 1979; Prescott & Triggs, 2008). The use of historical precedents is another important and long-established principle. In many instances, states have established new borders in accordance with previous international borders or with current administrative divides. Pre- existing borders have the advantage that they are usually well known by states

33 3. Main arguments and individuals on the ground, and even the location of old border precedents can usually be inferred from maps, written sources, and demarcations on the ground. An early example of an agreement based on historical precedent is the treaty of the Pyrenees (1659), which approximately located the frontier between Spain and France in the mountain range, based on the ancient Roman division between the Spains and the Gauls (Sahlins, 1989). In general, the idea that new borders should follow pre-existing ones has become so established that it has become an essential component of international law today. As noted earlier, the principle of uti possidetis has played a central role in the decolonization of Latin America and later Africa, during which newly independent states committed to preserving their inherited colonial boundaries. In many recent cases of secession, the same principle has been applied to ensure that new international borders were established along previous administrative boundaries (Zacher, 2001; Carter & Goemans, 2011). During the colonial expansion and the scramble for Africa, the use of natural features and historical precedents had clear limitations. Colonial powers such as France and Britain lacked detailed knowledge of African geography and local conditions and existing African states generally did not have clear borders. In order to quickly establish their territorial claims on the continent, colonial powers widely settled on straight borders derived from latitude and longitude lines (Englebert et al., 2002; Brownlie & Burns, 1979). The same principle had been applied earlier during the colonization of the Americas where states similarly competed over territories they had little knowledge about.4 The rise of nationalism gave rise to yet another organizing principle for the drawing of boundaries. Based on the ideal of national self-determination, nationalism holds that the boundaries of the state should match the national territory (Gellner, 1983). As a result, borders that matched the perceived ge- ography of nations were seen as more legitimate or more desirable than other types of borders. Instances in which state borders did not match national ones have fuelled numerous disputes and violent conflicts throughout recent history. At the same time, states have also increasingly relied on cultural features in attempts to claim territory and settle borders, as shown for example by the Treaty of Berlin (1878),5 the Paris Peace Conference (1919), and several colonial border agreements in Africa and Asia (Fischer, 1949; Englebert et al., 2002; Prescott & Triggs, 2008). In short, states have relied on a diverse set of focal principles through-

4Branch(2013) notes that the first attempt to divide up territories along straight cartographic lines was made in the Treaty of Tordesillas (1494), which settled disputes over between Spain and Portugal over territory in Latin America. 5This refers to the treaty that followed the Russo-Turkish war and not the Congo Conference.

34 3. Main arguments out recent history in attempts to claim territory and establish new borders. Common practices in border drawing have varied substantially throughout history and across regional contexts, presumably because different principles were commonly seen as more salient, convenient or legitimate under given circumstances.

Natural Historical Cartographic Cultural

B ● B ● B ● B ●

A ● A ● A ● A ●

Figure 3.2: Focal points in border drawing. A and B indicate the capitals of two states that compete over territory. The lines indicate potential divisions, derived from alternative principles.

3.4.3 Limitations

So far, I have sketched out a simple bargaining model of border drawing, in which states compete but are generally willing to work towards an agreement that enables them to avoid costly conflict. It is important to note that this model has clear limitations, which are briefly discussed here. First, my argument rests on the general assumptions that state leaders do not have an unlimited demand for expansion and that they are not indifferent to the costs of warfare. There are of course prominent historical examples in which these two assumptions have not been met, the most infamous ones being Napoleon and Hitler. These leaders had a seemingly insatiable drive for conquest and were more than willing to rely on all-out warfare in pursuit of this goal. However, the overall historical record suggest that these were in fact very rare exceptions, at least within the modern state system (Holsti, 1991; Schultz & Goemans, 2019). Instead, it appears that most states have generally attempted to avoid an uncontrolled escalation of conflict. On the contrary, Schultz and Goemans(2019) point out that in a number of prominent cases, the victor in a war willingly ceded territory that was under its control. Such behavior is more consistent with the idea that states pursue specific rather than unlimited territorial goals. In short, therefore, although the main assumptions may hold

35 3. Main arguments in most cases, it is important to note that there are important exceptions to which my arguments do not apply. A second limitation is that although many borders are based on agreements that follow some focal principle, this is not always the case. A prominent example is the demilitarized zone between North and South Korea, which simply reflects the military frontline at the end of the Korean war.6 Two other examples include the Line of Control between India and Pakistan in the Kashmir region and the buffer zone that separates the North of Cyprus from its South, which were similarly established based on cease-fire lines. In these instances, borders were created primarily based on "facts on the ground," rather than efforts to coordinate on a given boundary. However, it is still worth noting that in many instances in which borders were redrawn as the result of violent conflict, peace agreements have in fact established new borders in accordance with specific focal principles, which often resulted in the re-establishment of previous borders or the creation of new ones based on natural features or administrative divides (Fischer, 1949; Prescott & Triggs, 2008; Carter & Goemans, 2011). Still, the fact that some borders were purely the result of conflict demonstrates the limitations of my argument. Third and lastly, the bargaining model outlined in this chapter is highly simplified as its main purpose is to explore the use of focal points to better understand how existing borders were established. As such, the model does not account for a number of well known bargaining problems such as informa- tion asymmetries, commitment problems and indivisibility, which have been discussed at length elsewhere in the literature. In particular indivisibility often constitutes a fundamental obstacle to the resolution of territorial disputes (Toft, 2003; Hassner, 2003; Goddard, 2006). If opposing parties attach great symbolic value to a given territory, any outcome other than exclusive control over the undivided territory may be seen as unacceptable. Indivisibility is undeniably important in explaining why territorial disputes frequently escalate and are often difficult to resolve. As such, indivisible territory may help to account for instances in which states cannot settle their boundaries. Although indivisibility poses less of a problem when it comes to explaining the durability of existing borders, it does constitute an important limitation of my argument.

6Originally, the Soviet Union and the United States had divided Korea along the 38th parallel north. Following the Korean war however, a new border was defined by the demilitarized zone, which corresponded to the frontline at the end of the war (Walker, 2011).

36 3. Main arguments

3.5 The path dependence of international borders

Once states have settled on the distribution of territory, their leaders continually re-evaluate whether to accept existing borders or to initiate a dispute in an effort to redraw the map. This decision is based largely on a cost-benefit calculation in which leaders weigh the costs and benefits of the status quo against those of entering an uncertain bargaining process. As has often been noted, border disputes are costly. They carry the inherent risk of conflict, and can entail reputational costs on the international stage (Schultz, 2015a). In particular in the post-1945 period, rules and norms against conquest have made it much more difficult for states to legitimate claims on another state’s territory (Zacher, 2001). Moreover, border disputes entail significant economic opportunity costs, which may further prevent states from challenging the status quo (Simmons, 2005). Despite the clear drawbacks associated with disputes, there are numerous scenarios in which states may still be tempted to challenge existing borders. For example, an unfavorable previous settlement may motivate leaders to re- claim what has been lost. In addition, changes in the distribution of power may encourage them to push for such adjustments. The discovery of valuable resources across the border may also cause states to reconsider previous agree- ments. Moreover, the rise of nationalist movements or a change in government could result in new disputes. Considering these many potential sources of friction, it seems puzzling that most borders have remained largely stable in the post-1945 period. Indeed, this development seems to contradict the realist understanding of international politics as a constant struggle over territory which implies that borders are inherently unstable (e.g., Ratzel, 1897; Gilpin, 1981; Mearsheimer, 1994). The recent stability of international borders may also seem puzzling based on prominent theories of nationalism which, at a minimum, suggest that many borders remain under constant pressure due to their poor alignment with cultural divides (e.g., Gellner, 1983; Hechter, 2000; O’Leary, 2001). What explains this puzzle? One likely explanation is that border stability is self-reinforcing. As borders remain in place for longer periods of time, they increasingly shape political, economic and societal interactions both within and across borders which makes it increasingly difficult and costly to alter them. Early studies in political geography have already pointed at such path- dependent effects. In one of the first detailed historical studies of the evolution of borders, Fischer(1949, p. 212) noted that:

"With our own eyes we can observe this hardening of originally

37 3. Main arguments

quite meaningless ’paper’ boundaries drawn at a conference table and destined to have no real existence for many years afterwards. In Africa are found the classical examples of such boundaries drawn through unexplored areas. But it is difficult to change them now that they have functioned for several decades."

In a series of detailed case studies, Fischer(1949) shows that new borders, once established, can have a profound influence on local patterns of economic exchange, infrastructure development, social networks and intermarriage, and can even shape religious affiliations and local national identities. Furthermore, the author notes that the effects of borders on the cultural landscape can persist long after a border has been removed from the map. Along similar lines, there is also evidence that old borders have lasting economic effects. For instance, a more recent study on domestic trade following Poland’s unification in 1918 shows that the old partition borders continued to influence trade flows long after they ceased to exist as international boundaries (Wolf, 2005). The institutional approach again offers a good framework to study the self- reinforcing effects of border stability. To reiterate, its basic assumption is that clearly defined and mutually agreed upon borders serve as institutions that shape the behavior of states and individuals. In doing so, stable borders provide a host of political and economic benefits that are difficult to realize if borders remain uncertain and disputed (Simmons, 2005). Therefore, as the economic benefits of border stability accumulate, we may also expect that incentives to keep existing borders in place increase over time. In addition to accumulating economic benefits that raise the costs of dis- putes, one can also argue more generally that border consolidation itself is path dependent (Abramson & Carter, 2016). The literature on political institutions more generally has provided key insights into self-reinforcing patterns of insti- tutional stability, which are commonly described as increasing returns processes (Pierson, 2000). Institutions may be the result of historical contingencies, but actors are increasingly likely to adjust their behavior to them, the longer they remain in place. As individuals and organizations increasingly "invest" into the status quo and adopt institution-specific forms of behavior in their everyday interactions, it becomes more costly and difficult to replace existing institu- tions, even if potential alternatives may be more efficient or preferable from a normative standpoint. As this process continues, actors will develop stronger expectations about the persistence of institutions, causing them to adjust their behavior in ways that further contribute to institutional stability (North, 1990; Pierson, 2000; Levitsky & Murillo, 2009). Applying this same logic to international borders, we may expect that

38 3. Main arguments individuals become more likely to adjust their social and economic behavior to existing borders, the longer they have remained intact. Over time, borders therefore have an increasingly profound impact on local patterns of trade and social interactions. As more individuals coordinate on existing borders, it becomes more attractive for others to do the same which further reinforces expectations of border stability (Abramson & Carter, 2016). Whereas economic factors may explain a large part of the path-dependent process, other mechanisms are likely to play an important role as well. In general terms, as borders define the space in which domestic politics play out and in which other institutions evolve, they create what North(1990, p.95) has called an "interdependent web" of institutions that becomes increasingly dense over time. It is easy to see how, after the establishment of a border, the divergent institutional development of states makes it increasingly costly and difficult to extract a piece of territory from one state and integrate it into another.7 Another crucial aspect of path dependence that has so far received little attention in the institutional literature is the impact of borders on national identities. As Fischer(1949) has noted, interactions across an established border are often limited, while populations that live within the same territorial units tend to develop a distinct set of habits and social values over time. People within the same political unit interact more frequently, and are part of the same domestic political sphere, which supports the idea that they share a common history (White, 2004). The belief in common origins in turn is one of the main building blocks of national identity. In short therefore, borders can therefore be seen as the one of the main "containers" in which national identities develop (White, 2000, 2004). As borders shape and reinforce the identities of people that live within them, they increasingly come to be viewed as legitimate. In addition to economic mechanisms and mechanisms of institutional devel- opment, identities thus constitute a third mechanism by which border stability can become self-reinforcing. Arguably, the impact of borders on identities may be even more persistent than their effects on institutional and economic patterns. Whereas trade networks may adjust relatively quickly to the new status quo8 and domestic institutions can in principle be replaced, national identities can be passed on through generations, and may even develop a life of their own, if they are reproduced as a result of domestic politics.

7This general problem may also help explain the findings of a recent study on German re-unification, which is shown to have led to an overall reduction in economic growth (Abadie, Diamond, & Hainmueller, 2010). 8In the case of Poland, Wolf(2005) finds that the old partition borders had a substantial effects on domestic trade patterns for up to 15 years after their removal, after which the effects declined.

39 3. Main arguments

3.5.1 The persistence of border instability

If border stability is self-reinforcing, the same logic also points to a more trou- bling scenario in which border instability can persist over long periods of time. In regions where borders have frequently shifted in the past, individuals have good reasons to doubt the persistence of the status quo, in particular if new borders were recently established. If the future of current borders remains uncertain, their societal impact remains limited as local populations are less likely to adjust to them. As a consequence, the social and economic costs of redrawing borders may remain comparatively low, which can reinforce expecta- tions of border instability (Abramson & Carter, 2016). In addition, the existence of border precedents in historically unstable regions may creates motives and opportunities for state leaders to engage in territorial disputes (Ibid.). As international law tends to support claims based on historical possession and precedents, demands for the restoration of previous borders are more likely to be seen as legitimate on the international stage. This in turn makes it easier for leaders to mobilize domestic and international support for their efforts to redraw the map (Murphy, 1990; Abramson & Carter, 2016). Moreover, because previous borders often have lasting effects on the behavior and identities of people on the ground, they may effectively compete with current borders as the commonly accepted divide. If previous borders continue to affect social interactions and patterns of trade, they can also more easily be reinstated (Abramson & Carter, 2016). Although the discussion has thus far focused on territorial competition be- tween states, we may also expect that historical instability can fuel territorial disputes within states. Most challenges to border stability today originate from groups within states which either aim to secede and establish their own state, or pursue the goal of seceding and uniting with their co-nationals across the border. Like states, the leaders of such groups have clear incentives to make specific territorial claims that have some prospect of succeeding. Moreover, as sub-national groups are especially dependent on the support and goodwill of other states, they have a strong interest in making claims they can portray as legitimate. As such, we may expect sub-national groups to comply with the uti possidetis norm and confine their claims to existing administrative divi- sions. In addition, border precedents may also become an attractive target for secessionist and irredentist claims. A partial explanation for this is that calls for the reinstatement of previous borders may be seen as less disruptive than the drawing of entirely new ones and such demands therefore have a greater appearance of legitimacy.

40 3. Main arguments

Overall, we can generally assume that past border instability increases the opportunities for state leaders and sub-national groups to claim territory and dispute borders. At the same time, historical instability may also provide them with powerful motives to challenge the status quo. This second aspect has been less extensively discussed in the recent literature but is no less important. Perhaps most importantly, it is likely that border changes increase the risk of nationalist tensions. In the age of nationalism, most territorial disputes are intertwined with questions of national identity. Nations are "imagined commu- nities" (B. Anderson, 2006) which strive for some form of self-determination within their own governance units. National identities are usually based on cultural ties, such as common ethnicity, language and religion, and a shared belief in the nation’s deep historical roots. Just as importantly, nations are territorial groups of people, defined in large part by their collective attachment to a specific territory, which they view as their homeland. Nations claim exclu- sive ownership over their homeland and generally aim to achieve their goal of self-determination within its boundaries (White, 2000; Shelef, 2015). If political borders do not match the national territory, this is likely to moti- vate attempts to redraw the political map (Gellner, 1983; Hechter, 2000). While most of the literature on nationalism has focused on "static" configurations of political borders and national units and their impact on conflict, a more specific focus on border change may help to better explain when and where disputes emerge. In many instances, border changes have reduced, rather than increased the perceived overlap between political and cultural borders, as illustrated by the examples of the Russians, the Kurds, the Serbs and many other groups in historically unstable regions (Cederman, Rüegger, & Schvitz, 2019). Intuitively, we may therefore expect that in regions where borders have frequently shifted in the past, leaders may be more preoccupied with recapturing "lost territory", uniting the nation with stranded populations abroad or correcting some other perceived historical injustices. However, the relationship between nationalism and border change is less than straight-forward. Contrary to what nationalists themselves often suggest, nations do not exist as fixed cultural units that determine the natural distri- bution of states. Instead, national identities evolve in large part as a result of long-term developments within existing borders (White, 2004; Darden, 2013).9 Nationalism may indeed create significant pressures for border change, but nations are themselves to a large degree shaped and defined by existing borders.

9As one striking example, Darden(2013) shows how members of the same ethnic group within Austria-Hungary came to develop separate national identities after they were divided by an internal administrative boundary and subjected to different nation-building policies.

41 3. Main arguments

Based on this view, we can distinguish between two plausible mechanisms that link past border changes with nationalist motives for territorial conflict. First, in some instances, groups that have already developed cohesive na- tional identities may "lose out" as a result of border changes. For example, border adjustments can lead to the partitioning of such groups that were previ- ously united, or place an entire group under "foreign rule" in the territory of another state. In these instances, it is easy to see how such disruptive border changes can motivate collective action against perceived historical injustices. One clear example is the collapse of the Soviet Union in 1991, which left millions of Russians on the "wrong" side of the border and was recently described by Vladimir Putin as the "great geopolitical disaster of the century."10 Another example is the Soviet occupation of the Baltic states half a century earlier, which ended the existence of Estonia, Lithuania and Latvia as independent nation- states and immediately generated resistance on behalf of the occupied groups. A second possibility is that groups may retroactively construct nationalist claims based on historical configurations (White, 2000; Goemans, 2006). For example, some groups may already share a general sense of cultural distinc- tiveness and a desire for self-rule but need to identify a specific territory that can serve as the basis for a future state (White, 2004). The act of defining and claiming a homeland territory can help groups to mobilize around a common goal but can also play an essential role in establishing and reinforcing national identities (White, 2000, 2004). Homelands provide groups with an easier way to establish and communicate group boundaries as compared to claims based on common culture or history alone (Goemans, 2006). They further help to strengthen a nation’s identity by defining the space within which its people live (or where they are supposed to live) and within which their group-specific rules, norms and customs apply (White, 2000; Goemans, 2006). Lastly, homelands make it easier for anonymous individuals to develop a sense of solidarity with each other, due to their shared attachment to the same geographic space (White, 2004; B. Anderson, 2006). Therefore, potential nationalist movements face the crucial task of defining the territory where they aim to realize their goal of self-determination. In order to mobilize their members around a common cause, groups need to settle on a homeland definition that each of its members can instantly recognize. In this regard, historical precedents may again serve as a prominent focal point (Goemans, 2006; Mylonas & Shelef, 2017).11 Previous borders are often

10See http://kremlin.ru/events/president/transcripts/22931 11In addition to historical precedents, nationalist movements also frequently rely on sub- national administrative boundaries as the template for a future state (Roeder, 2007). This may be in part due to the fact that administrative borders are common knowledge.

42 3. Main arguments known by individuals on the ground and territorial claims based on them are therefore more likely to be recognized by all concerned. Moreover, in defining the homeland based on prior states or imperial divisions, groups are able to promote claims about the nation’s deep historical roots, which can further support their case for self-determination.

As an illustration of this point, White(2000) details how Romanian national- ists in the late 19th century attempted to develop a meaningful definition of the homeland on which their future state would be built. At the time, Romanian elites were divided across the Ottoman provinces of Walachia, Moldova and Transylvania, and they faced the task of delimiting the nations territory and identifying its people. The case for Romanian nationhood could not be made convincingly based on cultural differences alone, which at the time were rel- atively diffuse. In response, Romanian leaders reached into the past to find a historical justification for their nationalist efforts. They settled on the Roman province of Dacia, which had existed until around 275 AD and covered roughly the same area as the three Ottoman provinces (White, 2000).

To be sure, nationalist claims based on ancient history are quite rare, and therefore the Romanian example may be seen as an extreme example. How- ever, it is easy to find other examples in which states and sub-national groups have relied on historical precedents to claim specific territories (Murphy, 1990; Minahan, 1996). Examples include Iraq’s claim on Kuwait, which was based on previous territorial divisions under Ottoman Rule, or Afghanistan’s claim on Pashtun regions in Pakistan, based on its territorial losses in 1893 (Day, 1982). Many Serbs continue to view Kosovo as an inseparable part of Serbia, due to its role in the collapse of the Dušan empire (White, 2000). Similarly separatist groups such as the Karenni in Myanmar, the Bakongo in Angola, the and the Ewe in Ghana have all tied their demands for independence to the re-establishment of pre-colonial kingdoms . Such historical arguments do not only serve to justify territorial claims, but also serve to boost national identities and mobilize support among the population.

To summarize, the lasting societal impact of borders suggests that border instability can become persistent. In historically unstable regions, existing borders are less firmly institutionalized and border precedents become a likely target of revisionist and secessionist claims. Historical precedents facilitate the justification of territorial claims and can more easily be reinstated as new borders. Moreover, previous borders can serve as focal points for the definition of national homelands which serve to further justify and motivate demands for territorial change.

43 3. Main arguments

3.6 Rules, norms and precedents

In the previous section, I have discussed path dependence as a partial explana- tion for border stability, based on which I have argued that border instability may become persistent as well. A second explanation for border stability is located at the international level and has its own implications for the conse- quences of border change. Today’s borders have been kept in place in part by a set of international rules and norms aimed at reducing uncertainty and insta- bility in international politics. The modern international system is built on the basic principle that states hold exclusive sovereignty over the territory within their boundaries (Herz, 1957). International law regarding borders developed mainly to ensure that states were able to exercise their sovereignty without engaging in constant strife with each other and served to define territorial units and rules of non-interference (Herz, 1957; Shaw, 1997; Simmons, 2019). Since the 19th century, a broader set of rules and norms developed that further limited the conditions under which border changes were viewed as permissible. The first norm of fixed borders emerged in Latin America following decolonization and was based on the principle of uti possidetis (Shaw, 1997; Zacher, 2001). This norm held that newly independent states were to preserve their inherited colonial borders and was later also applied in post-colonial Africa. In 19th-century Europe, states adopted an informal set of rules and practices regulating territorial transfers which were guided by coordinated efforts to preserve the balance of power (Kacowicz, 1994). Once leading states had settled on a distribution of territory that reflected the regional balance, their main goal was to preserve existing borders (Holsti, 1991). However, although 19th-century rules and norms served to reduce instability on a larger scale, border changes as a result of conquest and annexation were generally seen as legitimate at this time. This started to change in the aftermath of World War I, which gave rise to territorial integrity norms that stated that territorial changes by force were impermissible. These norms became a key part of international law following World War II and have since been entrenched in a large number of bilateral and multilateral agreements (Zacher, 2001; Hensel, Allison, & Khanani, 2009; Hathaway & Shapiro, 2017). In general, international rules and norms on borders have become increas- ingly conservative, reducing the scope for instability. This development has not been evenly spread, as attempts to establish and enforce principles of border stability have often concentrated within regions. For example, states in 19th century Europe or in post-colonial Latin America and Africa have each adopted their own set of rules and practices to deal with the threat of instability. In the

44 3. Main arguments absence of a single, overarching enforcement mechanism, the preservation of existing borders largely depends on the willingness of states to cooperate, or on the ability of powerful states to enforce existing rules. As such, international rules and norms can be and have often been violated. A key question that has been frequently discussed is what happens if states violate rules and norms that are believed to be essential for international stabil- ity. Individual rule violations have often prompted fears of setting precedents, here defined as actions and events that can serve as an example or justifica- tion for future behavior (Kier & Mercer, 1996, p. 79). In addition to setting precedents, rule violations may also reveal information about the costs of non- compliance (Carnegie & Carson, 2018). If these costs turn out to be lower than may have been assumed, this may encourage further rule-breaking behavior. For these reasons, many studies have argued that individual border changes may destabilize borders elsewhere as they could encourage states elsewhere to push for their own border adjustments (Zartman, 1966; Touval, 1972; Fearon, 2004). However, the conditions under which this may occur and the mechanisms by which spillover effects play out have not been discussed in much detail within the literature. To address this, it is useful to consider the behavior of potential revisionist states within their own regional context. Some states may seek to claim and seize part of a neighboring state’s territory, but their behavior is generally constrained by the costs of doing so. Engaging in border disputes does not only involve the inherent risk of violent conflict but can also entail significant costs on the international stage (Schultz, 2015a). These costs can range from reputational costs to international sanctions or even military intervention by third states. Would-be revisionists are therefore likely to weigh the costs of initiating a dispute against the benefits of obtaining the desired territory and their likelihood of success. As the enforcement of international rules and norms largely depends on other states, the strength and effectiveness of international constraints often remains uncertain. In some periods, the international system may be more permissive towards territorial change than in others, as powerful states lack the willingness or capacity to enforce existing rules (Ambrosio, 2001; Abramson & Carter, 2018). Similarly, constraints on revisionism may vary across regional contexts due to differences in regional institutions or the varying influence of status-quo oriented states in different parts of the world. To anticipate the costs of revisionism and its likelihood of success, state leaders are likely to pay close attention to recent interactions between other, proximate states that operate within the same regional context. If other states nearby were able to successfully

45 3. Main arguments redraw borders in their favor, this is likely to encourage state leaders to seek their own territorial revisions.

3.7 Summary and outlook

This chapter has layed out the theoretical foundations for the subsequent chap- ters. First, I have sketched out a simple model of border drawing as a bargaining process, in which states rely on focal principles as the basis for potential agree- ments. In Chapter5, I develop this argument further and consider how the use of different types of focal principles may affect the durability of international borders. I test these arguments using disaggregated data on border segments and their durability since 1886. The second main argument holds that border instability may persist within regions. I have argued that due to the weak consolidation of existing borders and the existence of border precedents, his- torically unstable regions are likely to experience further instability. Chapter6 examines this claim using disaggregated data on historical border instability and the location of territorial disputes. Third and lastly, I have argued that border instability may be contagious as border changes that violate interna- tional norms can set precedents that may encourage other states to challenge the status quo. In Chapter7, I test this argument using data on territorial changes and disputes since the 19th century.

46 Chapter 4

Mapping the international system: The CShapes 2.0 dataset

Based on a collaboration with Luc Girardin, Seraina Rüegger, Lars-Erik Cederman, Nils B. Weidmann and Kristian Skrede Gleditsch1

4.1 Introduction

This dissertation relies heavily on CShapes 2.0, a new historical GIS dataset that maps country borders and capitals from 1886 to the present. The new data builds on the original CShapes dataset created by Weidmann, Kuse, and Gleditsch(2010), which covers independent states from 1946 onwards. Version 2.0 offers two new features: First, it extends temporal coverage by tracing borders back to 1886, which followed the Berlin conference on the partition of Africa that marked the first point in time at which most territorial divisions of the world were defined. Second, the new dataset covers colonies and other dependencies that were previously not included, and thereby provides global coverage throughout most of recent history. In this chapter, I introduce the new dataset, starting with a discussion of the overall coding procedure. I then present some illustrative applications of the data and conclude with a discussion of the dataset’s contribution to the literature.

4.2 Coding procedure

The overall goal of CShapes 2.0 is to map international borders from 1886 to the present. For this period, the dataset represents countries and their borders

1We also thank Dragana Vidovic, Baris Ari, Nora Schmidlin, Lukas Dick, Nora Keller, Sarah Daescher, Nino Abzianidze and Simon Pressler for their excellent research assistance.

47 4. Mapping the international system: The CShapes 2.0 dataset

Figure 4.1: Preview of the CShapes 2.0 dataset in 1886 as GIS polygons. Each country polygon is linked to a row in an attribute table that contains further information, such as the time period during which the polygon is active, the country’s sovereignty status and the name and location of its capital. Our representation of countries as time-varying polygons is based on a set of specific coding rules, which we describe in more detail in the following sections.

4.2.1 Defining the universe of cases

In order to represent political units in space, we first have to define them. We consider two types of units: independent states and dependent territories. For the former category, most political science research has relied on two main datasets on independent states that use their own definitions of statehood: the Correlates of War (COW) list and the K. S. Gleditsch and Ward(1999) list of independent states (GW). The COW list was first introduced by Russett, Singer, and Small(1968), and covers the period from 1816 to the present. During this period, COW records all units that qualify as system members according to certain criteria.2 Although the COW list is widely used in international relations research, it has been frequently criticized for its conceptual problems, inconsistencies in its coding scheme, as well as a Eurocentric bias (K. S. Gleditsch & Ward, 1999; Griffiths &

2Specifically, these criteria include diplomatic recognition by Britain or France in the period before 1920, and membership of the League of Nations or the United Nations in the periods thereafter. In addition, the COW list also codes states as independent if they exceed a population threshold of 500,000 and maintain diplomatic ties to at least two major powers (Russett et al., 1968).

48 4. Mapping the international system: The CShapes 2.0 dataset

Butcher, 2013). In response, others have introduced alternative lists of states. Among these alternatives, the list by K. S. Gleditsch and Ward(1999) has been used most commonly. The GW list builds on COW and generally covers the same units during the same time period, but it defines independent statehood according to three basic criteria that are applied consistently across all cases: States must have relatively autonomous control over their territory, be recognized by other regional actors, and have a population over 250,000 at any point in time during the sample period (K. S. Gleditsch & Ward, 1999). The GW list covers the same set of states that are listed by COW, but also includes a number of additional units.3 Moreover, because the GW list generally uses less restrictive criteria for statehood, it often dates the independence of states much earlier than the COW list.4 In line with other major data projects (K. S. Gleditsch, 2002; Vogt et al., 2014, e.g), we base our coding of independent states entirely on the GW list rather than relying on the COW list. There are three main reasons for this: First, the GW coding scheme is more consistent in our view, as it applies the same set of criteria across all cases. Second, the GW list generally provides better coverage of states outside of the European system, as it does not use diplomatic recognition by France or Britain as the main criterion for statehood before World War I. Third, because the GW list covers many states for longer periods of time, it also includes all country-years covered by COW, with just a few minor exceptions. In other words, the COW list is therefore almost entirely subsumed by the GW list. It is important to note that our decision to rely exclusively on the GW coding marks a departure from the previous CShapes dataset, which was designed to be compatible with both the COW and GW state lists. This facilitated the use of CShapes with many other datasets, but required a complicated double coding scheme, which would have been difficult to maintain in our effort to backdate international borders to 1886. For this reason, we refrain from using the double coding scheme for the time being. However, although CShapes 2.0 is entirely based on the GW coding of independent states, we still provide access to a separate, COW-based version of the data for independent states after 1946 to ensure compatibility with COW for the post-World War II period. Following our definition of independent states, we now turn to their de- pendencies. We define as dependent territories those units that are under the

3For example, the GW list codes Tibet (1913-1950) and Orange Free State (1854-1910) as independent states, which are not part of the COW list. 4An example is Canada, which COW codes as an independent state from 1920 onwards, while GW sets its independence date to 1867.

49 4. Mapping the international system: The CShapes 2.0 dataset control of an independent state, but that are not considered part of its core territory. These are typically non-adjacent territories ruled as colonies or pro- tectorates. To gather information on dependencies, we rely on a second list of political entities, which is also taken from the COW project. The original COW list of political units covered both independent states and all other units that were classified as dependencies (Russett et al., 1968). These units were later removed from the main COW list, but others have relied on the initial coding to create a separate list of dependencies over time (Wyckoff, 1980; Bennett & Zitomersky, 1982). Our dataset relies on the latest version of this list, taken from the supplementary data from the Territorial Change dataset (Tir, Diehl, & Goertz, 1998). For each dependent territory, the COW list of dependencies lists the inde- pendent state it belongs to and indicates its status. For example, Hawaii is listed as a U.S.-colony from 1898 until 1959, when it became part of the United States. Nigeria is listed as British colony from 1914 until 1960, when it gained independence. Our coding of dependent territories includes four categories from the original COW list: (1) colonies, (2) protectorates, (3) international man- dates and (4) occupied territories.5 In cases where a dependent territory gained independent, we code its dependency status up until the date of independence recorded by the GW list. To ensure consistency with our coding of independent states, we have also narrowed down the list of dependencies by selecting only those units with a population greater than 250,000 during the sample period. Figure 4.2 shows the number of states and dependencies in our dataset over time.

4.2.2 The geographic extent of states

Having defined the units that make up the international system, our next task is to map their geographic extent. As in the original CShapes dataset, we code a state’s territory primarily based on its internationally recognized boundaries. In most cases, this means that we code borders as they were defined in bilateral and multilateral agreements and shown on contemporaneous maps. One challenge, however, is that some borders lack international recognition or remain the subject of a dispute. For example, Israel and Syria continue to dispute sovereignty over the Golan Heights, while India and Pakistan remain locked in a dispute over Jammu and Kashmir. In such instances, there may

5We exclude four other types of dependencies that were part of the original list: possessions, leased territory, neutral or demilitarized zones and claimed territory. Territories classified as pos- sessions and leases were generally too small to be included in our sample, while demilitarized zones and territorial claims do not match our coding of territorial units.

50 4. Mapping the international system: The CShapes 2.0 dataset 150

Type

100 Dependencies Independent States 50 0 1880 1900 1920 1940 1960 1980 2000 2020

Figure 4.2: States and dependencies over time be multiple competing descriptions and maps of the same territories.6 Instead of coding disputed territories separately, however, we assign them to a given state based on its de facto control over the region. In the case of the Golan Heights, this means that we assign the disputed territory to Israel, although its control over the region is not internationally recognized. In the case of Kashmir, we code the Line of Control as the existing border, although this border remains disputed by both India and Pakistan. In some cases, we lack clear evidence that any state exercised de facto control over a disputed region. The border between Oman and Saudi Arabia is a case in point. This border runs through mostly uninhabited desert areas and remained undefined until well into the 20th century. Negotiations between Britain and Saudi Arabia in 1935 failed, after which both sides continued to make conflicting claims, as shown on contemporaneous maps (Schofield, 2016). Saudi Arabia and Oman finally settled on a border agreement in 1990, which closely followed the initial border proposed by Britain (Peterson, 2020). Throughout the dispute, we found no evidence that either side successfully seized control of the disputed areas. In such instances, we simply backdate borders as they were eventually defined in the agreement that settled the dispute. Another related issue are de facto states, such as Abkhazia and South Ossetia, which declared independence from Georgia in the early 1990s, but have not received international recognition. Similar examples include Biafra’s attempted secession from Nigeria in 1967 or the Republic of Serbian Krajina that split

6One potential solution would be to code disputed territories as separate units that do not belong to any state. However, we view territorial disputes as an important subject in their own right, which are best dealt with as part of a separate data collection effort. A recent example of such a dataset is the Mapping Interstate Territorial Conflict (MITC) project by Schultz(2015a).

51 4. Mapping the international system: The CShapes 2.0 dataset from Croatia in 1991. Because these entities do not match our definition of independent states, we do not code them as separate units and instead assign their claimed territory to the host state they are located in. Lastly, an additional challenge in determining the geographic extent of states is that until the early 20th century, some political units lacked precise borders that enclosed their territories. This was the case especially in parts of Africa, the Middle East and East Asia, where borders long remained poorly defined or non-existent. In these regions, colonial powers and local rulers often gradually settled on the location of their borders in successive agreements (Brownlie & Burns, 1979). Although we would ideally be able to trace the gradual delineation of borders in these cases, our use of polygons in the dataset does not allow us to do so, since it requires us to represent states as closed spatial units.7 In cases where state borders remain partially undefined, we therefore add a placeholder polygon that represents the borders as were eventually defined. These polygons are flagged with a dummy variable to indicate that their borders remain uncertain during a given period, which enables users to remove or modify these observations if necessary.

Coding Territorial Changes

To account for changes in international borders over time, we furthermore need to precisely define what constitutes a territorial change. We distinguish between three types of territorial change, and have systematically gathered data for each type. First, territorial changes may occur due to the creation and dissolution of political units. An example of such a change is the collapse of the Austro- Hungarian empire in 1918 and the subsequent creation of several new states, including Austria and Hungary. A second type of territorial change occurs when states exchange sovereignty over territories as a whole. One example is Germany’s loss of German West Africa (Namibia) to South Africa under the Treaty of Versailles in 1919. In these instances, a territorial unit may change ownership, but its borders remain intact. Thirdly, territorial changes can occur if a part of a unit’s territory is transferred to another, which involves a re-drawing of borders between existing units. An example for this is Germany’s loss of Alsace-Lorraine to France in 1919, which also occurred under the Treaty of Versailles. To code changes due to the creation and dissolution of units, our dataset relies on the GW list of states and the COW list of dependencies, which record

7A potential alternative would be to represent states with partially undefined borders as lines, instead of using polygons. However, this would result in a less consistent data structure that would make it more difficult to compare units over time.

52 4. Mapping the international system: The CShapes 2.0 dataset the historical lifespans of political units. For transfers of sovereignty over units as a whole, we use the COW list of dependencies, which keeps track of changes in state sovereignty. For border adjustments between existing units, we rely mainly on the Territorial Change Dataset (Tir et al., 1998), which lists all territo- rial transfers that involved at least one independent state since 1816. For each change, the dataset indicates the gaining and losing side, and provides addi- tional information on the territory that changed hands, such as the territory’s name and its approximate size. For feasibility reasons, we have restricted our coding efforts to transfers of territory larger than 100 x 100 km, as done in the previous version of CShapes. Since the Territorial Change Dataset is based on the COW list of states, this excludes some territorial exchanges between entities that qualify as indepen- dent states according to GW, but not according to COW. Moreover, although the Territorial Change dataset records transfers of dependent territory among colonial powers, it does not account for border adjustments that occurred within the same colonial empire. Therefore, the dataset does not capture all instances of territorial change that are relevant for us. To address this, we gathered addi- tional information, relying mainly on the Encyclopedia of International Boundaries by Biger(1995) and another encyclopedia of African Boundaries by Brownlie and Burns(1979). Both sources provide detailed accounts of territorial changes and the location of borders after each change. In keeping with the coding rules of the Territorial Change Dataset, we used the precise date of treaties that confirm the reallocation of territory as the date of each change. Furthermore, our approach excludes wartime territorial changes, unless they were made permanent after the war. In these instances, we relied on the date of postwar agreements as the date of the change.

Geocoding

Using the information described in the previous section, we created a compre- hensive list of all relevant changes that have occurred since 1886. This list served as basis for the identification of country-periods during which the sovereign sta- tus, the capital and borders of a country remain unchanged. Conversely, any change in these attributes marks the beginning of a new period. After defining all country-periods, we gathered additional information on the location of terri- torial transfers, and on the historical circumstances under which each change took place. This information is summarized in our dataset’s documentation, which provides a detailed chronology of territorial changes per country and justifies our coding decisions in a number of ambiguous cases.

53 4. Mapping the international system: The CShapes 2.0 dataset

The identification of country-periods supported the collection of historical maps that depict the status quo of borders before each change. Figure 4.3 shows an example of such a map depicting borders in Europe in 1936. We geo-referenced these maps using GIS software and used them to draw and modify country borders. Border changes were coded in reverse chronological order. In other words, we started with the earliest observation in the original CShapes dataset, and adjusted country polygons to represent borders before each territorial change.8 For countries that did not experience any changes in the previous period, such as Switzerland or Spain, we simply backdated their borders to 1886. In cases where changes did occur, such as the border between France and Germany in 1919, we adjusted those portions of the border affected by each change in reverse chronological order. Figure 4.4 shows an example of the changing political map in Southeast Europe between 1886 and 1945.

4.3 Applications of the data

This section presents a number of illustrative applications of the new dataset, starting with a descriptive analysis of border changes and aggregate trends in state size. We then illustrate how CShapes 2.0 can be used in combination with other GIS data sources to compute new aggregate measures, using the GREG dataset on ethnic settlement areas as a specific example.

4.3.1 Global trends in border stability and state size

What can we learn from the new data? We start with a descriptive analysis of territorial changes in recent history, as recorded by our dataset. Figure 4.5 plots territorial changes, which are divided into three categories: The first panel shows changes resulting from the creation of new units, the second panel shows border adjustments among existing territorial units, and the third panel shows cases in which states exchanged sovereignty over territories as a whole. Overall, the data confirms that territorial changes have become increasingly rare in recent history. In the first two panels, we see an overall decline in the frequency of border changes from the late 19th century to the present. The only major exception to this trend were the dissolution of the Soviet Union and Yugoslavia in the 1990s, which entailed a large number of border changes, mostly due to the creation of new states. Transfers of sovereignty over entire territories

8In most cases, the earliest date recorded by CShapes 0.6 is January 1st, 1946. For colonies that became independent after 1946, coverage in the previous dataset started at their date of independence. The current version traces their borders back to the onset of colonization.

54 4. Mapping the international system: The CShapes 2.0 dataset

Figure 4.3: Map of Europe in 1936 used to code country borders. Source: Atlas Universel de Geographie (Vivien de Saint-Martin & Schrader, 1937)

Figure 4.4: Preview of the data: Border changes in Southeast Europe 1886-1946

55 4. Mapping the international system: The CShapes 2.0 dataset became somewhat more common in the early 20th century, and skyrocketed in the 1960s. This increase was almost entirely driven by decolonization, during which former colonizers formally transferred sovereignty to the governments of newly independent states. Another question that has received much attention in the literature is state size. Aside from a long-standing theoretical debate (e.g., Friedman, 1977; Tilly, 1990; Alesina & Spolaore, 2003), a number of studies have examined historical trends in the distribution of state sizes. For example, Lake and O’Mahony (2004) present global data on state sizes from 1816 to the present, showing that the average size of states increased throughout most of the 19th century, after which it has dropped continually until the present. In contrast, Abramson(2017) examines state sizes in Europe from 1100 to 1790, and finds that the typical state decreased in size throughout the entire period. Part of the discrepancy between these two findings may be explained by the different time periods and regions examined by both studies, and by their differing definitions of statehood. Another likely reason is more technical: Whereas the first study evaluates trends based on the average size of states in square kilometers, the second study examines log-transformed data. As Abramson(2017) shows, the distribution of state sizes is highly skewed, with a small number of extremely large states distorting the mean value. Log- transformed data are less sensitive to distortions by extreme values, and may therefore provide a better understanding of how the size of "typical" states evolved. To evaluate post-1886 trends in state sizes, we calculated the yearly size of country polygons in 1’000 square kilometers. Following Abramson’s ap- proach, we computed the yearly mean, median and upper and lower quartiles of state sizes, using both untransformed and logged data. The results are shown in Figure 4.6, which shows the untransformed data on the left, and the log- transformed data on the right. In line with Abramson(2017), our results show that the distribution of state sizes has remained highly skewed since 1886, as becomes clear from the untransformed data. The mean is consistently higher than the 75th percentile, which suggests that throughout the entire period, there have been a few extremely large states, compared to a majority of much smaller states. However, both the mean and skewness of the untransformed data have declined since the 1900s, while the 75th percentile has experienced a similar decline. This suggests that overall, states at the upper end of the distribution have declined in size, following a slight increase in the late 19th and early 20th century. The log-transformed data reveals a similar decline in state sizes, although the overall trend is less pronounced. The sharpest decline occurred in

56 4. Mapping the international system: The CShapes 2.0 dataset

Figure 4.5: Territorial changes over time as recorded by CShapes 2.0. Dashed lines indicate smoothed trend estimates

Figure 4.6: Trends in state size the second half of the 20th century, which is largely the result of decolonization and the dissolution of the Soviet Union and Yugoslavia. To summarize, our data points to an overall decline in state size throughout the 20th century. Most of this decline is driven by the contraction of the largest states in the international system, but even among "typical" states, we see a slight overall decline in state size.

4.3.2 "Right-sizing" the state?

In another example, we illustrate how CShapes 2.0 can be combined with other spatial data sources to further explore long-term historical developments. In particular, we use data on ethnic settlement areas to examine whether there

57 4. Mapping the international system: The CShapes 2.0 dataset has been a trend towards more ethnically homogeneous states, whose borders have become more closely aligned with ethnic divides. Several prominent theories have suggested that the rise of nationalism since the 19th century has created increasing pressures for border change, often with the goal to align state borders with cultural divides. Gellner(1983, p. 1) has defined nationalism as "a political principle, which holds that the political and the national unit should be congruent." Perceived violations of this principle may motivate groups to push for border changes (Hechter, 2000). From a state-centric perspective, O’Leary(2001) has argued that modern state leaders have incentives to engage in the "right-sizing" and "right-peopling" of states, which refers to a set of policies aimed at creating culturally homogenous states with cohesive national identities. One way to achieve this goal is via border adjustments that help reduce internal ethnic differences. To examine long-term trends in ethnic heterogeneity within states, we com- bined CShapes 2.0 with another GIS dataset on ethnic settlement territories, from which we derive time-varying measures of ethnic fractionalization. In this example, we used the GREG dataset (Weidmann, Rod, & Cederman, 2010), which locates ethnic groups in space based on the soviet Atlas Narodov Mira (ANM) from the 1960s. We overlayed the data on ethnic settlement territories with our time-varying border data, and extracted those portions of ethnic settle- ments that fall within a country at a given time, as illustrated in Figure 4.7. Next, we calculated the yearly size of ethnic settlement areas within each country, which we used as a rough proxy for the overall size of each ethnic group. We then calculated the relative sizes of ethnic groups within each country across time, and computed yearly ethnic fractionalization values.9 These values were then aggregated into global and regional averages. Figure 4.8 plots the results on a global level and by world regions. From 1886 until the early 20th century, global levels of ethnic fractionalization increased, as colonial powers created new territorial units in Africa, which encompassed ethnically diverse populations. Ethnic fractionalization values then started to decline in the early 20th century, a development that was largely driven by a wave of secessions from the Ottoman Empire, Austro-Hungary and Russia, which gave rise to many smaller, more homogenous states in Asia and Europe. The second substantial decline occurred in the 1990’s, as the Soviet Union and Yugoslavia dissolved. While the overall variation is relatively modest on a global scale, variation within and across regions is more substantial.

9We used the standard measure of ethnic fractionalization, based on the Herfindahl index, n 2 which is defined as follows: 1 − ∑i=1 s , where s denotes the size of each ethnic group’s settlement area, divided by the combined settlement area of all ethnic groups within the country

58 4. Mapping the international system: The CShapes 2.0 dataset

Figure 4.7: Intersecting historical borders with ethnic settlement areas

In the Western Hemisphere, very few major border changes have occurred since 1886, and therefore fractionalization has remained at around the same, relatively high level. In Europe, fractionalization has continually decreased since the early 20th century, especially following World Wars I and II. In Africa, ethnic fractionalization increased during colonization, but has not changed sub- stantially since then, in part because border changes have remained relatively rare after decolonization. Asia has experienced the largest overall drop in frac- tionalization in the early 20th century, which was mainly due to the dissolution of the Ottoman Empire and the partial disintegration of Russia following the Russian revolution, which gave birth to more homogeneous successor states. Lastly, Australia and Oceania experienced an increase in ethnic fractionaliza- tion in the late 19th century, which is mainly due to the establishment of new colonies.

To be sure, there are clear limitations to our approach to studying long- term trends in ethnic fractionalization. A first issue is that the size of ethnic settlement areas does not always correspond to population size, as there are sub- stantial variations in population density. In principle, this could be addressed by incorporating gridded population estimates to calculate population-based fractionalization estimates, which however goes beyond the scope of this illus- trative application. A second, more serious limitation is that the GREG dataset only offers a static "snapshot" of ethnic groups and ethnic settlement areas at certain point in time. The underlying atlas was originally drawn by Soviet ethnographers in the 1960s, and as such GREG does not account for changes in ethnic group definitions and settlement areas that may have occurred before or since then. To address this, we would require time-varying data on ethnic settlement areas that covers the entire period from 1886 to the present, which is currently not available. One option to partially address this issue is by using the GeoEPR dataset (Vogt et al., 2014), which does account for changes in ethnic

59 4. Mapping the international system: The CShapes 2.0 dataset

Figure 4.8: Estimated trends in ethnic fractionalization geography in the post-1946 period.10 However, the present example mainly serves an illustrative purpose, and we therefore do not seek to further improve the accuracy of our measure. Despite the limitations of our approach however, the current example still provides a general idea of global trends in ethnic fractionalization.

4.4 Comparison to existing datasets

As noted in the introduction, CShapes 2.0 improves significantly upon its predecessor by tracing borders further back in time, and by including dependent territories that were previously not covered. Whereas the previous version left out large parts of the world map under the colonial period, the new version provides complete coverage throughout most of recent history. Figure 4.9 illustrates this difference by comparing two maps of Africa in 1946, based on the previous and the new version of the dataset. It is important to note that CShapes 2.0 is not the only GIS dataset on interna- tional borders. For example, two other widely used sources include the Natural Earth and Global Administrative Areas (GADM) databases (Hijmans, 2012), which include both international and sub-national borders. These datasets

10It should be noted however that GeoEPR has another drawback, in that it only represents ethnic groups that are considered politically relevant at a particular point in time. Therefore, changes in ethnic group relevance could lead to changes in aggregate measures derived from GeoEPR polygons, as the one discussed here.

60 4. Mapping the international system: The CShapes 2.0 dataset however only cover current borders and do not account for historical border changes. Aside from these two sources, there are also a number of other histori- cal GIS datasets, some of which cover even longer periods of time. For example, the Euratlas project (Nuessli, 2010) maps states and other political entities in Europe from 1 AD to the present, but only provides periodical snapshots of the continent in 100-year intervals. The Centennia Atlas (Reed, 2008) covers Europe from the 11th century to the present, and provides a much higher temporal resolution of 5-week intervals. Other historical GIS projects include the CHGIS project (Lex Berman, 2001), which traces prefectures in pre-modern China to the present, and the Japan Historical GIS database (Lex Berman, 2017), which covers Japan between 1664 and 1820. Therefore, while some existing datasets offer much greater historical depth, they are limited to specific world regions. To the best of our knowledge, CShapes 2.0 is currently the only historical GIS dataset that maps borders across the globe since the late 19th century. Finally, while some existing datasets such as Euratlas or Centennia require users to purchase a license, CShapes 2.0 is made freely available for academic use and other non-commercial purposes.

CShapes 0.6 CShapes 2.0

Type

Independent States Dependencies

Figure 4.9: Comparing CShapes 0.6 and 2.0: Africa in 1946

4.5 Conclusion

This chapter has introduced CShapes 2.0, a new historical GIS dataset that maps the international system from 1886 to the present. The dataset is the most detailed and comprehensive of its kind as it provides both historical depth and global coverage. In a number of illustrative applications, we have examined the frequency of different types of territorial change over time and explored long- term trends in state sizes and ethnic fractionalization. Aside from these specific examples, the new dataset also enables researchers to calculate new measures,

61 4. Mapping the international system: The CShapes 2.0 dataset using historically accurate information on country borders and capitals. For example, the data can be used to calculate minimum distances between states or inter-capital distances, which are frequently used to construct spatial weights matrices in cross-country analyses. Most importantly, CShapes 2.0 enables new research on a number of topics that were so far difficult to study, such as the causes and consequences of border change or the expansion of colonial rule. In the subsequent chapters, I draw on the new dataset to study why some borders have been more stable than others and to examine the consequences of border change.

62 Chapter 5

Defining the outlines: Border drawing and border durability

To remove every subject of discord, every occasion for quarrel, one should mark with clarity and precision the limits of territories - Emmerich de Vattel(1758), The Law of Nations

5.1 Introduction

Why have some borders remained stable throughout recent history, while others have shifted? A number of recent studies have examined how border characteristics and border change affect the risk of conflict between states (Carter & Goemans, 2011; Abramson & Carter, 2016; Goemans & Schultz, 2017). However, the causes of border change, and more specifically, the conditions under which they occur remain understudied. An improved understanding of these questions is undoubtedly important from a policy perspective, as it enables us to identify those areas that are especially prone to instability and conflict. The question is also important from a theoretical and empirical perspective. Borders are the result of complex interactions that take place between and within states, and as such it is crucial to understand how they are created in the first place, and what accounts for their stability over time. Since borders are endogenous to politics, it is furthermore difficult to study the effects of border change in statistical analyses, which assume that a given "treatment" is randomly assigned among comparable units (King, Keohane, & Verba, 1994; Rosenbaum, 2010). If this condition is not met, any observed relationship between border changes and any outcome of interest may be spurious. Although I do not claim to have a comprehensive solution to this problem, a systematic analysis of when and where borders change occurs help

63 5. Defining the outlines: Border drawing and border durability to address this issue by uncovering potential omitted variables. In this chapter, I contribute to this goal by examining how the way in which borders were initially drawn affects their durability over time, building on the arguments introduced in Chapter3. For borders to effectively coordinate relations between states, they need to be of high focal quality. In other words, they need to be well-defined, unambiguous and recognizable on the ground (Carter & Goemans, 2011; Goemans & Schultz, 2017). Based on this argument, I argue that borders that follow clear-cut natural features or well known historical precedents are more likely to endure, while borders that follow straight lines are more likely to be redrawn. To test these claims, I use the CShapes 2.0 dataset that was introduced in Chapter4, which enables me to trace the durability of international borders from 1886 to the present. I focus on land borders between neighboring states as the unit of analysis, which I disaggregate into historical segments of comparable lengths. I then develop a set of measures to capture the alignment of borders with specific focal principles. First, I rely on fractal dimensions as a measure of border straightness. Second, I use other sources of GIS data to measure the alignment of borders with rivers, watershed lines, mountains and other geographical referents. Third, I use information on past border changes to measure a border’s alignment with historical precedents. In a series of survival analyses, I generally find strong support for my arguments: straight borders have been less likely to persist, while borders that follow rivers, watersheds or historical precedents have been more durable. Moreover, I find that not all of these effects remain constant over time. In particular, the negative effects of border straightness tends to decrease with time, suggesting that even initially "flawed" borders may become increasingly resistant to change. This chapter builds on previous research on borders in international rela- tions. In particular, recent studies have shown that new borders that follow pre-existing administrative boundaries or prominent natural features have been less prone to interstate conflict (Carter & Goemans, 2011; Goemans & Schultz, 2017). Following these studies, this chapter goes one step further by examining the durability and change of international borders as another important out- come in its own right. Whereas states have often disputed borders in an attempt to redraw the map, such attempts have often been unsuccessful. It is therefore important pay close attention to when and where border changes occur. In the following sections, I outline my arguments, which build on the theoretical discussion in Chapter3. I then turn to the empirical section, in which I detail the main research design considerations and the construction of my data, after which I discuss my results. I conclude with a discussion of limitations and

64 5. Defining the outlines: Border drawing and border durability avenues for future research.

5.2 Explaining border durability and change

In Chapter3, I have argued that borders are institutions that serve to coordinate relations between states. Borders are usually established as part of a bargaining process, in which states compete over territory but have a common interest in reaching a definitive settlement. Assuming that state leaders are willing to negotiate an agreement,1 they face a difficult coordination problem. One can in principle draw an infinite number of borders to divide up a given territory, and state leaders do not exactly know the other side’s true preferences and cannot well anticipate the other side’s behavior. To quickly find a settlement that helps minimize the risk of incessant conflict, leaders have an incentive to rely on focal principles. These are best understood as general rules that can be used to identify certain outcomes, or focal points, which are qualitatively distinct from all other options (Huth et al., 2013). In the context of border drawing, state leaders have incentives to follow a specific set of focal principles, such as the use of natural features or historical precedents, which help them to narrow the bargaining range down to one or multiple potential borders that "stand out" among an infinite number of potential divides (Schelling, 1980; Carter & Goemans, 2011). Throughout recent history and across regional contexts, states have applied a varying set of focal principles in border negotiations, presumably because different principles were seen as more salient, convenient or legitimate under different circumstances. Specifically, a large number of borders have been established along prominent natural features, such as rivers, mountains or wa- tershed lines, while others have been derived from historical or administrative precedents. In other instances, competing states have settled on borders that follow straight lines derived from cartographic referents. The main argument I outline here is that the initial use of these alternative principles has had varying implications for the stability and persistence of international borders. I gener- ally assume that state leaders are less inclined to push for the adjustment of borders that effectively coordinate the behavior and expectations of states and individuals (Goemans & Schultz, 2017). Such borders do not only help states to avoid constant friction with their neighbors, but generally provide legal certainty that can have important political and economic benefits (Simmons, 2005). Therefore, if borders are well-defined and recognizable on the ground,

1They may not be willing to do so if the stakes are seen as indivisible, see Chapter3.

65 5. Defining the outlines: Border drawing and border durability they should also be less likely to be redrawn. As previous studies have noted, this may be especially difficult to ensure if borders follow straight lines that were derived from latitude and longitude lines on the map (Curzon, 1907; Prescott, 1978; Goemans & Schultz, 2017). The use of cartographic principles may have seemed the most convenient or feasible in cases where states competed over territories that were largely unknown to them, as was the case during the Scramble for Africa. However, straight borders often generated substantial difficulties in practice, which were at times unanticipated.2 One obvious challenge is that such borders generally do not align with recognizable features of the landscape, which makes it more difficult for people on site to establish exactly where one state ends and the other begins, which laws apply where and which people belong to which state. For example, Prescott(1978) notes that colonial administrators in Africa often had great difficulties in determining the precise location of straight borders, especially if the border ran through difficult terrain. For the local population, it was even more difficult to comply with new borders that only existed as imaginary lines on the map. The author illustrates this his case study of the border between Rhodesia and Mozambique in the late 19th century, where uncertainty over the location of straight portions of the border led to numerous disputes between colonial powers and the local population over questions of jurisdiction, taxation and trespassing (Prescott & Triggs, 2008, p. 62). A second potential issue with straight borders is that they are also more likely to cut across existing demographic patterns, thereby separating mem- bers of previously united ethnic groups (Curzon, 1907; Englebert et al., 2002; Alesina et al., 2011). This second aspect is one of the main reasons why several recent studies have argued that straight borders are especially problematic (e.g., Englebert et al., 2002; Alesina et al., 2011). Indeed, there is evidence that the partitioning of ethnic groups by international borders can, under cer- tain circumstances, trigger conflicts and border disputes (Michalopoulos & Papaioannou, 2016; Goemans & Schultz, 2017; Cederman et al., 2019). In some cases, ethnic partition may give rise to irredentist efforts to unite the group into a common state. However, the relationship between borders, ethnic partition and nationalism is not as straight-forward as some studies have suggested. For example, Alesina et al.(2011, p. 247) argue that straight borders are problematic simply because they do not correspond with the "natural division of peoples" or with the "division of nationalities desired by the people on the ground".

2For example, Curzon(1907, p. 35) remarked that "[Straight] lines are very tempting to diplomatists, who in the happy irresponsibility of their office-chairs think nothing of intersecting rivers, lakes, and mountains, or of severing communities and tribes."

66 5. Defining the outlines: Border drawing and border durability

However, people are not naturally divided into fixed nationalities that deter- mine the optimal distribution of states. Instead, national identities are just as often shaped by existing borders as they can become a driving force for border change (White, 2004). This is not to deny that ethnic partition can indeed give rise to border instability and conflict. However, a better understanding of this relationship requires a more careful examination of the conditions under which this occurs, which goes beyond the scope of the present study.3 Instead, the present argument emphasizes that straight borders are problematic for a wholly different reason, which is unrelated to existing national identities. Compared to borders that are well-defined and unambiguous, straight borders are less effective in coordinating relationships between states, and states therefore have fewer incentives to preserve them:

Hypothesis 5.1 Straight borders are more likely to be redrawn than borders that follow irregular lines.

In comparison, borders derived from natural features have the decisive advantage that they stand out in the landscape, leaving much less room for con- fusion or disagreement about their location. In particular, rivers and watershed lines have often been described as useful foundations for international borders, as they serve as recognizable features. For example, the geographer and bound- ary commissioner Holdich(1899, p. 468) argued that rivers and watersheds make for effective boundaries that "every man can recognize from afar, and which he cannot mistake if he encounters [them]".4 It should be noted that even the use of supposedly clear-cut natural features can cause confusion over their exact location, as, for example, rivers have a certain width and can change their course over time (Prescott & Triggs, 2008; Goettlich, 2018). However, states have generally relied on principles that help them minimize such uncertainty, such as the general practice of using the thalweg as the precise location of river boundaries (Huth et al., 2013). Other natural features such as mountains, deserts or forests have long served as effective frontier zones in earlier times, where states did not possess linear borders and where frontiers often served as protective buffers against invasion (Prescott, 1978). Despite their defensive advantages, these continuous features of the landscape usually lack clear-cut boundaries, and therefore do not offer specific guiding principles on where to draw the line (Broek, 1943). In general, we can expect that natural features that help to define clear territorial lines also contribute to border stability:

3For example, some studies have shown that the relationship between ethnic partition and conflict risk depends in part on the partitioned group’s size and political power access (Cederman, Gleditsch, Salehyan, & Wucherpfennig, 2013; Goemans & Schultz, 2017) 4See also: Curzon(1907); Brigham(1919)

67 5. Defining the outlines: Border drawing and border durability

Hypothesis 5.2 Borders that follow clear-cut natural features are more likely to endure than borders that do not.

Like prominent natural features, historical precedents may also provide clear focal principles for the establishment of new borders. Geographers have long noted that previous borders often continue to leave visible marks on the surrounding cultural landscape, as shown for example by distinct infrastruc- tural and architectural patterns on each side of the divide (Hartshorne, 1936; Fischer, 1949). In some cases, previous borders continue to influence individual behavior and identities long after they have been erased from the map (Fischer, 1949; Wolf, 2005; Abramson & Carter, 2016). If previous borders were clearly demarcated, their location furthermore remains visible on the ground. We can therefore generally assume that the location of border precedents often remains common knowledge for long periods of time, and may expect the following:

Hypothesis 5.3 Borders that follow historical precedents are more likely to endure than borders that do not.

Although the present analysis does not address the role of cultural divides as an organizing principle,5 it is useful to briefly consider their use as potential boundaries as part of this discussion. As noted in Chapter3, the rise of national- ism did not only fuel an increasing number of territorial disputes and conflicts based on cultural claims but also led to growing efforts to establish new borders in accordance with ethno-linguistic divides. One major challenge here is that it is often difficult to draw borders based on culture alone, due the subjective and often fuzzy nature of ethnic and national identities, and because ethnic settlement patterns are often diffuse and overlap in many places. In line with this view, Prescott and Triggs(2008, p. 64) note that cultural borders "generally cause more disagreement than features of the landscape," both before and after they have been established. This also helps explain why even in many instances of secession, new borders followed pre-existing administrative boundaries, rather than ethnic divides (Carter & Goemans, 2011). To summarize, states have relied on a varying set of focal principles through- out modern history and across regional contexts, and the use of different princi- ples have resulted in borders with varying degrees of clarity. As states are more likely to benefit from borders that are clear-cut and recognizable, their leaders have greater incentives to preserve such borders, as opposed to borders whose exact location remains less clear to people on the ground. The latter applies in

5I do not examine the alignment of borders with cultural divides because I do not have data on ethnic geography for the period 1886-1946.

68 5. Defining the outlines: Border drawing and border durability particular to straight borders that exist mainly as lines on the map and are by themselves not visible in the landscape. Of course, states can overcome their coordination problems in the long run by physically demarcating their borders on the ground. At least historically however, such demarcation efforts have been very costly and time-consuming, and have often taken several decades to complete (Prescott & Triggs, 2008).6 This suggests that all else equal, straight- line borders have been the least effective in coordinating relations between states. While some focal principles may be more effective than others, it is important to note that they cannot always be neatly separated from each other, as they may often overlap in practice. For example, a river border may also correspond to a historical precedent, or a watershed boundary may also divide culturally distinct populations. Presumably, states have an incentive to align new borders with not just one, but multiple overlapping principles, in order to minimize their disruptive effects, and to facilitate coordination. Similarly, the focal quality of borders cannot always be separated from their defensive properties. Rivers may have helped states to coordinate on a given border, but they have historically also served as barriers against invasion. Watershed boundaries may serve to neatly divide up state territories, but the fact that they are often located in mountains may suggest that they were also chosen for strategic reasons. The main argument I am making here is not that the defensive value of borders is irrelevant. Rather, I argue that aside from their defensive properties, the coordinative value of certain borders constitutes another important explanation for why some borders have been more durable than others.

5.3 Research design and data

To test my arguments, I conduct a series of Cox proportional hazard analyses in which the outcome variable is the "survival" of a given border segment, while the explanatory variables capture its alignment with natural features, historical precedents and straight lines derived from cartographic referents. The analysis begins with extensive data preparation efforts, which I describe first, before discussing the variable selection and model specification.

6For example, Curzon(1907) notes that the surveying and demarcation of the straight border between the United States and Canada took over 50 years to complete.

69 5. Defining the outlines: Border drawing and border durability

5.3.1 Border segments

To analyze the durability of international borders, I rely on CShapes 2.0, a GIS dataset that maps international borders from 1886 to the present (Schvitz et al., 2019). A series of adjustments to the original dataset were made: Whereas CShapes represents countries as polygons that cover their entire territory, the present analysis focuses exclusively on boundary lines that divide neighboring states. To obtain these boundary lines, I use the original polygon data to create a time varying dataset of land borders between pairs of states. The next challenge is to distinguish those parts of an international border that have remained stable over time from those that have been redrawn. To this end, I divide borders up into historical segments, which I define as continuous portions of a border that have the exact same lifespan. To illustrate how these segments are created, consider Italy’s borders as an example. Following World War I, Italy annexed the South Tyrol region, as well as parts of Slovenia from Austria-Hungary. Meanwhile, Italy’s borders with France, Switzerland and parts of its Austrian border remained unchanged, as shown in panels A and B of Figure 5.1. To obtain historical border segments, I start by extracting the land borders between Italy and its neighbors at any given point in time from the polygon dataset, as shown in Figure 5.1 C. I then combine all historical boundary lines and extract the nodes at which they intersect. Using these nodes, I then split up all boundary lines into unique segments that have existed at any point in time during the sample period, as illustrated in Figure 5.1D. In a final step, I spatially join my border segments with the original CShapes data to determine the periods during which each border segment existed. Figure 5.2 illustrates part of the results, showing the age of international borders across Europe in 2018. It is important to note that the lifespan of each border is measured independently of the countries on each side of the border, to account for the fact that new states often inherited parts of pre-existing borders. For example, Ukraine was established as an independent state in 1991, following the collapse of the Soviet Union. While its border with Russia was established as a new international border in 1991, its borders with Poland are much older, and date back to the end of World War II, as can be seen on Figure 5.2. Having created historical border segments, one remaining challenge is that they exhibit vast differences in length. For example, the longest section of the US-Canadian border measures around 5680 km, while the entire border between Germany and Luxembourg is only 130 km long. Such variations are problematic for two reasons: First, because my approach splits borders that underwent changes into smaller segments by design, this means that the

70 5. Defining the outlines: Border drawing and border durability

Figure 5.1: From country polygons to border segments. Panels A and B: Borders before and after adjustments. C: Extracting boundary lines. D: Splitting lines into historical segments

Figure 5.2: Border segment age in 2018

71 5. Defining the outlines: Border drawing and border durability length of border segments is likely to correlate with their observed durability. Second, variations in length make it difficult to derive comparable measures of important characteristics, such as border straightness or their alignment with natural features. For example, a long section of an international border may follow a river in some parts, and may follow a straight line in others. Any attempt to measure these characteristics across the entire length of a border is likely to mask important variation that exists on finer scales of measurement. To address these challenges, I divide historical border segments into smaller sub-segments of the same length. In the main analysis, I set the maximum length of each border segment to 100 km.7

5.3.2 Measuring border straightness

One of the main variables in the present analysis is the straightness of inter- national borders. As noted in the previous discussion, some studies have used border straightness as an indicator for their "artificiality" or arbitrariness (Englebert et al., 2002; Alesina et al., 2011). More generally, I have argued that straight lines do not make for particularly effective borders, as they are generally not recognizable on the ground. The main question here is how we can distinguish straight borders from more "squiggly" lines. Some previous studies have used a categorical measure of border straightness (Englebert et al., 2002; Goemans & Schultz, 2017), which requires manual coding of straight vs. irregular borders based on some arbitrary cutoff criterion. In contrast, the present analysis relies on a continuous measure that is better able to capture the full range of variation between perfectly straight and increasingly irregular borders. This continuous measure is based on fractal dimension of borders, a metric that is commonly used to capture the complexity of objects. Unlike standard measures of dimensionality, which can only take on discrete values, fractal dimensions are continuous: a perfectly straight line has a dimension of 1, a plane has a dimension of 2, while irregular lines have a fractal dimension that falls somewhere in between. Following a previous study by Alesina et al.(2011), I measure the fractal dimension of border segments using the box-counting method. This method consists of fitting a plane to each line segment, which is then divided into a set of grids of varying resolutions.8 For each grid, the number of boxes that overlap with the line is counted, as illustrated in Figure

7In a follow-up to the main analysis, I replicate my results across a set of alternative datasets with lengths ranging from 25 to 400 km, to ensure that my findings are not an artefact of any given segment length. 8My box-counting approach uses the resolutions of 1 divided by 2, 4, 8, 16, 32, 64, 128 and 256.

72 5. Defining the outlines: Border drawing and border durability

5.3 A. This results in a dataset of box counts across different grid resolutions, which I then use to fit the following regression:

log(boxcount) = −βlog(scale)

Where −β reprsent the object’s fractal dimension. One challenge in using the box counting approach is that the resulting fractal dimension scores are often sensitive to random variations in grid placement and rotation (Bouda, Caplan, & Saiers, 2016).9 As a result, two identical border segments with slightly different angles may have very different fractal dimension scores. To account for this, I use an algorithm to find the optimal rotation of each border segment, so as to minimize its box counts on the highest-resolution grid.10. This results in consistent fractal dimension scores that are not influenced by the orientation of borders on the map. Figure 5.3 B and C show the resulting fractal dimension values for two example datasets using this method.

5.3.3 Natural features and historical precedents

In addition to the measure of border straightness, I create a series of mea- sures to capture each border’s alignment with key geographic features. First, I measure the proportion of each border segment that aligns with major wa- tershed boundaries, using data from the Global Runoff Data Centre(2007). 11 Second, I measure a border’s overlap with rivers and lakes, using geo-coded data from Natural Earth(2015) and Lehner and Döll(2004). Another variable captures whether a border segment runs through mountainous terrain, using high-resolution digital elevation data from Danielson and Gesh(2011). This rugged terrain measure is calculated within a 25 km buffer zone around each segment, using the standard deviation of each pixel’s elevation from sea level. In addition to these terrain-specific variables, I add a dummy variable that measures whether a given border segment has been re-established, i.e. whether it follows a previously existing border segment according to the CShapes 2.0 dataset. As noted previously, some border segments were erased from the map, but were later re-established at the exact same location. This variable can there- fore be seen as a rough proxy measure for whether borders follow historical precedents. It should be noted that this measure is clearly incomplete, as many historical precedents date back much further than 1886, which is the earliest

9Thanks also to Carl Müller-Crepon for pointing this out. 10This is done using a standard optimization function in R that uses the Brent algorithm. 11I calculate this variable by creating a minor buffer around each drainage divide, after which I intersect each border segment with the resulting buffers, I then measure the length of each intersection relative to the total length of each segment.

73 5. Defining the outlines: Border drawing and border durability

Figure 5.3: Calculating fractal dimension scores. A: Box counts across different grid resolutions. B: Fractal dimensions of increasingly squiggly lines. C: Fractal dimensions scores for Mali in 2018 (100 km segments) year covered by the CShapes dataset. In addition, many historical precedents were based on previous sub-national, rather than international borders (Carter & Goemans, 2011), for which I lack adequate data. Despite these limitations, the variable might still provide some insights on the role of historical precedents. To reduce the risk of omitted variable bias, I add a set of control variables that are likely to influence the way in which borders are drawn or that may otherwise affect the results. First, I add two logged measures of population density and agricultural suitability surrounding each segment, as borders are likely to be drawn differently in densely populated or valuable regions. These variables are measured within 25 km buffer zones around each segment, us- ing data from Klein Goldewijk, Beusen, De Vos, and Van Drecht(2011) and FAO/IIASA(2011). To account for the distinct nature of borders that were externally imposed, I include a dummy variable that indicates whether a border segment was established under colonial rule, as coded by CShapes. Finally, I add a logged measure of each segment’s length. Although the maximum length of each segment is set to 100 km in the main analysis, my approach of dividing borders into historical segments results in a number of segments that are much shorter. Shorter segments are by definition more common in areas that have experienced frequent border changes, which makes it important to account for differences in segment length.

5.3.4 Model specification

The main analyses consist of a set of Cox Proportional Hazard models, in which the dependent variable measures each border segment’s stability over time.

74 5. Defining the outlines: Border drawing and border durability

More precisely, the Cox model considers how long a given unit survives until it "fails", and estimates the impact of predictor variables on the risk of failure. Crucially, the Cox model accounts for the fact that many units have not failed during the observed period, but may do so in the future, by treating information on these outcomes as censored. My baseline model is written as follows:

0 hi(t) = h0(t)exp(βX ) + eg

Where hi(t) denotes the hazard rate for unit i, and h0(t) denotes the baseline hazard. Robust standard errors are clustered on the level of country-dyads, which are measured at the first point in time at which a border segment is observed.12

5.4 Results

A first set of analyses provides general support for my arguments, as shown in Table 5.1. Model 1 only includes the main explanatory variables, while Model 2 adds the controls. The two remaining models include gamma frailty terms, which are used to account for other unobserved heterogeneity that may render some groups of observations more failure-prone than others (Box-Steffensmeier & Jones, 2004).13 Model 3 adds a region-specific frailty term, while Model 4 accounts for frailty on the level of country-dyads. Note that all continuous variables have been standardized, in order to keep regression parameters on comparable scales. As a result, the coefficients for the continuous variables represent the effect of increasing that variable by one standard deviation (Eager, 2017). Across all models, I find that border segments with higher fractal dimension values have had a lower risk of failure during the observed period. In other words, the results show that "squiggly" border lines have overall been more durable than borders that follow straight lines, in support of Hypothesis 5.1. Furthermore, I find that borders that align with watershed lines are more likely to endure, as are borders that follow rivers. My findings also suggest that new borders that follow previous historical borders are more durable, but these results are not robustly significant across all models. However, it should

12Each border segment belongs to a country-dyad, which is defined by the pair of countries that shares a border at a given point in time. 13Frailty models also account for dependence among observations due to repeated observa- tions for the same unit (Box-Steffensmeier & Jones, 2004), which can occur if border segments are removed and later re-established.

75 5. Defining the outlines: Border drawing and border durability

Table 5.1: Cox PH models: Border durability

Model 1 Model 2 Model 3 Model 4 Fractal dimension −0.192∗∗∗ −0.191∗∗∗ −0.139∗∗∗ −0.141∗∗∗ (0.066)(0.066)(0.033)(0.039) Watershed align. −0.205∗∗∗ −0.230∗∗∗ −0.240∗∗∗ −0.165∗∗∗ (0.064)(0.064)(0.043)(0.052) River align. −0.184∗∗ −0.201∗∗∗ −0.205∗∗∗ −0.169∗∗∗ (0.082)(0.072)(0.035)(0.049) Reestablished −0.156 −0.454 −0.497∗∗∗ −1.813∗∗∗ (0.359)(0.334)(0.130)(0.155) Lake cover −0.089 −0.071∗ −0.056 (0.071)(0.039)(0.044) Ruggedness −0.793∗∗∗ −0.665∗∗∗ −0.369∗∗∗ (0.148)(0.067)(0.082) Pop. dens. (log) 1.156∗∗∗ 0.847∗∗∗ 0.463∗∗∗ (0.189)(0.091)(0.120) Agric. Suit (log) −0.353∗∗∗ −0.297∗∗∗ −0.037 (0.111)(0.059)(0.060) Pre-1886 0.387∗∗ 0.389∗∗∗ 0.854∗∗∗ (0.188)(0.070)(0.150) Colonial 0.288 0.034 −0.168 (0.202)(0.098)(0.247) Length, km (log) −0.053 −0.108∗∗∗ 0.104∗∗ (0.050)(0.036)(0.046) Frailty term No No Region Dyad AIC 18425.371 18138.832 18001.994 16138.595 Num. events 1166 1165 1165 1165 Num. obs. 3647 3638 3638 3638 Missings 249 258 258 258 Notes: Cox PH Models. Unit of analysis: Border segments (Max length: 100 km). All con- tinuous variables are standardized to facilitate interpretation. Regions consist of: West- ern Hemisphere, Europe, Africa, Asia, Australia and Oceania. Robust standard errors are clustered on the dyad level. Significance levels: ∗p<0.1, ∗∗p<0.05, ∗∗∗p<0.01

76 5. Defining the outlines: Border drawing and border durability be noted that cases in which borders were removed and re-established were relatively rare in the period since 1886, while I also lack geographic data on older historical precedents. Turning to the control variables, I find mixed evidence to suggest that lakes are associated with more durable borders, but the results also suggest that borders that run through mountainous terrain are more stable, while borders in densely populated areas and those established prior to 1886 are more prone to border change. To summarize, my results provide support for Hypothesis 5.2, which suggests that the alignment of borders with clear-cut natural features increases their durability over time. Lastly, the results do not consistently support Hypothesis 5.3, which states that new borders are more stable if they follow historical precedents. To enable a substantive interpretation of the main results, Figure 5.4 plots predicted survival curves across the minimum and maximum values of each explanatory variable, while holding all other variables at their mean or mode. Predicted survival rates were calculated based on the estimates from Model 2, which does not include a frailty term.14 The results show substantial differ- ences in the survival rates of straight and irregular borders across time. The differences are less pronounced, but still clear for borders that follow historical precedents, rivers or watershed lines.

5.4.1 Non-proportional hazards

So far, the analysis has relied on Cox proportional hazard models, which are based on the core assumption that the impact of a variable on the underlying hazard rate remains constant over time.15 If this is not the case, coefficient estimates may be biased, and significance tests might be misleading (Box- Steffensmeier, Reiter, & Zorn, 2003). As Licht(2011) points out, the proportional hazards assumption is frequently violated in social science applications, where theoretical arguments often already suggest non-constant hazards and effects. For example, if our goal is to explain the tenure of state leaders, we may expect that experienced politicians are better able to respond to threats to their survival than inexperienced ones. Similarly, path dependence arguments suggest that institutions can become more entrenched and resistant to external shocks as time passes (Licht, 2011). In the case of border durability, I have argued in Chapter2 that it redrawing borders may become increasingly costly and difficult as time passes. This could also mean that differences in border

14Frailty models produce group-specific survival curves and are therefore not well suited to estimate average survival rates (Box-Steffensmeier & Jones, 2004). 15It should be noted that other survival models, such as the Weibull model are based on the same assumption (Box-Steffensmeier & Jones, 2004)

77 5. Defining the outlines: Border drawing and border durability

Figure 5.4: Predicted survival curves for the main explanatory variables (95% confidence intervals) characteristics may lose relevance over time, as even borders that were initially "flawed" may eventually become more institutionalized, as actors on the ground learn to coordinate and adjust to them. The goal of this section is to examine and deal with the possibility of non- proportional hazards. I start by testing the proportional hazards assumption, using a residual-based statistical test developed by Grambsch and Thernau (1994). This shows that most variables in the models presented so far indeed violate the assumption.16 One way to address this is by estimating stratified Cox models, which incorporate separate baseline hazard rates across the levels of a variable that exhibits non-proportional hazards (Box-Steffensmeier & Jones, 2004). While this can alleviate inferential concerns, one drawback is that the effect of the stratification variable can no longer be estimated. As such, this approach boils down to treating non-proportional hazards as a nuisance that needs to be accounted for and does not allow one to examine their substan- tively interesting properties. However, stratified Cox models are still useful to evaluate the overall sensitivity of the results to non-proportional hazards, and are therefore used in a first follow-up analysis. Table 5.2 shows the results

16The results of the non-proportionality tests are shown in the appendix to this chapter.

78 5. Defining the outlines: Border drawing and border durability of a first set of stratified Cox models. Models 1, 2 and 3 are stratified by three of the binary controls for pre-1886 borders, colonial borders and reestablished borders. In addition, Models 4 and 5 are stratified by regions and dyads. All of the main findings reported in Table 5.2 remain robust throughout these alternative specifications. However, a subsequent test reveals that although stratification has helped to reduce non-proportionality, some variables still violate the proportional hazards assumption. An alternative approach is to directly model the relationship between pre- dictor variables and time. This enables one to examine changes in substantive effects over time. This is useful insofar as changes in the effects of certain focal principles changes over time would clearly be of theoretical interest. Follow- ing Box-Steffensmeier and Jones(2004), I estimate a model that includes all predictor variables and interact those variables that violate the proportional hazards assumption with the natural logarithm of time. I then use a simulation- based method proposed by Licht(2011) to calculate and visualize changes in substantive effects over time.17 Figure 5.5 plots the results of this exercise for the four main explanatory vari- ables. The y-axis shows the change in the effect from switching from the highest to the lowest value of a predictor variable. Negative effects denote a lower risk of "failure". Interestingly, I find that the effect of border straightness indeed changes with time: Irregular borders, i.e. those with higher fractal dimension values, initially had a much lower risk of failure than straight borders, but this difference decreases over time and becomes indistinguishable from zero after around 40 years. This suggests that the negative effects of border straightness tend to decrease as time passes. This is generally consistent with my argument presented in Chapter3, which stated that border stability is path dependent. As such, we may assume that even borders that were initially "flawed" may become more firmly established over time, as these results suggest. In contrast, I do not find clear evidence of time-varying effects for the other variables. Borders that follow watershed boundaries or rivers are less likely to experience border change and this effect remains mostly constant over time. I find a similar effect for borders that were reestablished, although their initial rate of failure appears to be high.18

17I use the simPH R package by Gandrud(2015), which implements the method introduced by Licht(2011). 18This counter-intuitive finding may be explained by a subset of cases in which borders were removed and reinstated within a short period of time, as for example in the short-lived union of Mali and Senegal, which existed for less than a year in 1959.

79 5. Defining the outlines: Border drawing and border durability

Table 5.2: Stratified Cox Models: Border durability

Model 1 Model 2 Model 3 Model 4 Model 5 Fractal dimension −0.195∗∗∗ −0.187∗∗∗ −0.192∗∗∗ −0.136∗∗ −0.103∗∗ (0.065)(0.065)(0.066)(0.068)(0.051) Watershed align. −0.234∗∗∗ −0.228∗∗∗ −0.228∗∗∗ −0.236∗∗∗ −0.119∗ (0.064)(0.064)(0.064)(0.068)(0.066) River align. −0.205∗∗∗ −0.197∗∗∗ −0.194∗∗∗ −0.209∗∗∗ −0.124∗ (0.071)(0.073)(0.073)(0.073)(0.073) Reestablished −0.409 −0.503 −0.450 −1.472∗∗∗ (0.345)(0.317)(0.373)(0.433) Lake cover −0.089 −0.085 −0.088 −0.064 −0.033 (0.071)(0.072)(0.072)(0.065)(0.040) Ruggedness −0.795∗∗∗ −0.787∗∗∗ −0.798∗∗∗ −0.670∗∗∗ −0.306∗∗∗ (0.151)(0.145)(0.151)(0.139)(0.108) Pop. dens. (log) 1.143∗∗∗ 1.169∗∗∗ 1.158∗∗∗ 0.860∗∗∗ 0.321∗ (0.190)(0.187)(0.191)(0.195)(0.178) Agric. Suit (log) −0.339∗∗∗ −0.371∗∗∗ −0.349∗∗∗ −0.314∗∗∗ 0.037 (0.110)(0.114)(0.112)(0.104)(0.055) Pre-1886 0.332∗ 0.367∗ 0.336∗ 0.673 (0.185)(0.188)(0.189)(0.445) Colonial 0.325 0.283 0.010 −2.464∗∗∗ (0.208)(0.203)(0.277)(0.454) Length, km (log) −0.055 −0.062 −0.032 −0.125∗∗ 0.103∗∗ (0.049)(0.051)(0.052)(0.054)(0.046) Strata Pre − 1886 Colonial Reestabl. Region Dyad AIC 16533.260 16559.487 17657.509 14581.769 5091.452 Num. events 1165 1165 1165 1165 1165 Num. obs. 3638 3638 3638 3638 3638 Missings 258 258 258 258 258 Notes: Cox PH Models. Unit of analysis: Border segments (Max length: 100 km). All continuous variables are standardized to facilitate interpretation. Regions consist of: Western Hemisphere, Europe, Africa, Asia, Australia and Oceania. Robust standard errors are clustered on the dyad level. Significance levels: ∗p<0.1, ∗∗p<0.05, ∗∗∗p<0.01

80 5. Defining the outlines: Border drawing and border durability

Figure 5.5: Non-proportional hazards: Simulated first differences based on log- time interactions. Plots indicate how the impact of predictor variables changes with time.

81 5. Defining the outlines: Border drawing and border durability

5.4.2 Robustness

To further asses the robustness of my findings, I carry out two additional tests, which are discussed here and are shown in the appendix. First, I replicate the main analysis shown in Table 5.1 across a range of alternative datasets, in which the maximum length of each border segment ranges from 25 to 400 kilometers. This allows me to evaluate whether my findings are sensitive to random changes in the length of border segments used in the analysis. The results, summarized in Figure A4, suggest that this is generally not case. Specifically, the estimates for border straightness and their alignment with watershed lines and rivers remain significant across all alternative segment lengths. The estimates for reestablished borders also remain consistent but are not robustly significant, as most estimates do not reach the 0.05 significance threshold and one estimate fails to reach the 0.1 threshold. In a second robustness test, I replicate the main analysis in a series of simple OLS models, in which the unit of analysis is the segment-year and the dependent variable records the "failure" of a given border segment. To be sure, OLS is not the most suitable model for this type of analysis, as it does not account for right-censoring. Furthermore, the dependent variable in this particular analysis is binary, and some authors have cautioned against using OLS with discrete-outcome variables (Wooldridge, 2015). However, the main advantage of this approach is that it allows me to include a large number of fixed effects, in order to to account for observed and unobserved heterogeneity among observations. Therefore, I estimate four OLS models that include different combinations of dyad and year fixed effects to account for both between-unit heterogeneity and temporal shocks. Furthermore, I include the cubic polyonomials of time since the last border change, to account for serial correlation within segments that experience multiple "failures."19, The results are generally consistent with my previous findings, as shown in Table A5. The only exception concerns the estimates for reestablished borders, which fail to reach the 0.1 significance threshold in one of the models.

5.5 Discussion and conclusion

In this chapter, I have examined how the way in which borders were drawn influences their stability over time. I have argued that borders are rarely drawn arbitrarily, but instead are established as part of a bargaining process between

19This occurs with some border segments that are removed from the map and are reinstated at a later point in time.

82 5. Defining the outlines: Border drawing and border durability states, in which parties rely on commonly recognized focal principles in order to reach a definitive outcome. Such principles include the use of natural features, historical precedents or cartographic referents as the basis for new international borders, which have all been used to varying degrees across different historical and regional contexts. I have argued that not all principles have resulted in equally effective boundaries, as not all can ensure that borders are unambigu- ously defined on the ground. If the drawing of borders leaves uncertainty about their actual location, this increases the potential for friction and uncertainty and reduces the potential for cooperation and positive-sum benefits. As a result, leaders may have less incentives to uphold existing boundaries. Based on this argument, I have argued that straight borders that were de- rived from cartographic principles have historically been less likely to endure, while borders that follow clear-cut natural features and historical precedents have been more likely to persist. Using disaggregated spatial data on the loca- tion and timing of border changes since 1886, I generally find strong support for my arguments. The only finding that is not robustly significant across all models relates to the role of historical precedents. As I have noted however, the present analysis has relied on incomplete information on border precedents, and does not account for precedents that existed prior to 1886, or for the role of adminis- trative precedents. An interesting task for future research therefore would be to study the role of border precedents in more detail, using more comprehensive historical data on previous international and internal boundaries. It is important to note that the present analysis has a number of other limitations. Like most other research on borders, one main challenge is that borders are endogenous to international politics. As such, certain political dynamics that result in different types of borders may also have an influence on the risk of subsequent border changes. For example, some colonial powers may have used straight-line borders simply a temporary place-holders meant to establish their territorial claims, and may have planned to renegotiate them as soon as they effectively controlled the territory. Alternatively, states that agree on settling their borders along a river or a historical precedent may simply be more willing to cooperate to begin with, which should also decrease the risk of future instability. These issues are difficult to resolve and call for a more extensive examination of how and where borders are drawn in the first place, which would be another important task for future research. Despite these limitations, the present chapter offers a number of new findings that help explain when and where border changes are most likely to occur. In the subsequent to chapters, I will explore how such changes, if they occur, may affect the potential for further instability and conflict.

83 5. Defining the outlines: Border drawing and border durability

84 Chapter 6

Vicious cycles: The persistence of border instability

6.1 Introduction

Challenges to border stability remain one of the most pressing issues in inter- national politics today. Many states remain locked in disputes over territorial sovereignty and the location of their borders, which often give rise to serious tensions between them (Schultz, 2015a). The past decades have also seen a steady rise of separatism across the world, with many groups calling for in- dependent statehood (Griffiths, 2016; Sambanis et al., 2018). As most states are reluctant to give up sovereignty over parts of their territory, secessionist disputes often escalate into full-scale civil wars. Where do these territorial disputes come from? And where are they most likely to emerge? In this chapter, I examine the historical roots of territorial conflict between and within states. While a large part of the literature has focused on violent forms of territorial conflict, less is known about the origins of the disputes that drive these conflicts. Building on the arguments introduced in Chapter3, I argue that many disputes we observe today are the direct result of past border instability. In regions where borders have been historically unstable, the leaders of states and sub-national groups have greater motives and opportunities to claim territory and dispute borders as compared to historically stable regions. In turn, an increased potential for disputes also raises the risk of further border changes. To assess the persistence of border instability, I examine how pre-World War II border changes relate to post-1946 patterns of dispute and instability. I use the CShapes 2.0 dataset on international borders introduced in Chapter4, which I combine with two additional data sources on the location of interstate

85 6. Vicious cycles: The persistence of border instability and domestic territorial disputes.1 The latter data source is new and part of an ongoing data collection effort, which I introduce as part of this chapter. In an analysis conducted on the level of grid cells, I find that past border changes are indeed associated with an increased risk of interstate territorial disputes, secessionist and irredentist claims, and further instances of border change. These findings are robust to a range of additional tests, which include random variations in the size and shape of grid cells used in the analysis, a reanalysis of the data within regional sub-samples, matching and spatial filtering. This chapter contributes to a growing body of research on the origins of territorial conflict. In particular, I build on a recent study by Abramson and Carter(2016), which shows that historical border precedents help explain the emergence of territorial disputes in Europe.2 In the present analysis, I go be- yond previous research in presenting the first global analysis of the persistence of border instability. In addition, whereas previous research on borders has focused entirely on the interstate domain, the present analysis how historical border instability affects the risk of domestic territorial disputes. In doing so, I also contribute to recent studies that have examined the emergence of separatist disputes before they turn violent (“Lost Autonomy, Nationalism and Sepa- ratism”, 2014; Bartuseviˇcius& Gleditsch, 2019; Germann & Sambanis, 2019). While these studies have shown that political exclusion and lost autonomy help explain demands for secession, this chapter shows that past border instability constitute another important explanation. In the following sections, I review the main arguments developed in Chapter 3, from which I derive the hypotheses to be tested in this chapter. I then discuss the research design and data used to test my hypotheses, followed by the presentation of my results. The final section concludes by discussing some of the limitations of this chapter, and outlines the main tasks for future research.

6.2 Border instability and its consequences

In Chapter3, I have argued that border instability can persist over long periods of time, as past instability generates motives and opportunities for the leaders of states and sub-national groups to push for renewed border adjustments. I briefly restate the general argument here, before deriving my main hypotheses. My argument builds on the idea that stable borders are institutions that structure

1In this chapter, I use the term domestic territorial disputes to refer to secessionist and irreden- tist disputes that arise within states, and that may or may not be violent. 2The authors use geocoded data on border changes in Europe between 1648 and 1789 and examine their effects on post-1816 dispute emergence.

86 6. Vicious cycles: The persistence of border instability the behavior of states and individuals. In doing so, stable borders reduce the potential for friction between states and reduce uncertainty, which can generate wide-ranging political and economic benefits for states on both sides (Simmons, 2005; Carter & Goemans, 2011; Schultz, 2015a). Based on this idea, I have argued that border stability is path dependent. Part of this is explained by the profound impact that borders have on the institutional development of states and the social and economic interactions of people on the ground (Fischer, 1949; Wolf, 2005; Abramson & Carter, 2016). In large part due to their far-reaching social, economic and political implications, borders can be seen as the "containers" within which national identities develop (White, 2000). These wide-ranging effects make it easy to see how it can become increasingly costly and difficult to extract a territory from one state and integrate it into another, the longer current borders have remained in place. At the same time, the path-dependent nature of border stability also suggests that border instability can become persistent. First, borders in historically unstable regions are less firmly institutionalized, which should make it easier for states to redraw the political map. Second, the existence of border precedents in these regions creates opportunities for leaders to claim territory. Calls for the restoration of old borders are more likely to be seen as legitimate on the international stage than demands for entirely new borders, which suggests that border precedents make it easier for state leaders to mobilize international and domestic support for their claims (Murphy, 1990; Abramson & Carter, 2016; Carter, 2016). In addition, as previous borders often continue to influence the behavior and identities of people on the ground, they may effectively compete with current borders as the commonly accepted divide. Because of their institutional history, previous borders can also easily be reinstated, which should make them more likely targets for territorial claims (Abramson & Carter, 2016). Based on this logic, we may expect that state leaders are more likely to initiate disputes in historically unstable regions:

Hypothesis 6.1 Historical border instability increases the risk of interstate territorial disputes.

While previous studies have examined how borders shape relations between states, less attention has been paid to their impact on domestic politics. In Chapter3, I have argued that we may expect a similar relationship between historical border instability and domestic territorial disputes. Like states, the leaders of potential secessionist movements have incentives to make territorial claims that have some prospect of succeeding. If groups can tie their demands for self-determination to a historical precedent, it is more likely that they can

87 6. Vicious cycles: The persistence of border instability portray their demands as legitimate, which increases their chances of securing domestic and international support. Aside from the importance of legitimizing claims, I have also argued that border changes can generate grievances among members of groups that have "lost out" as a result of the change. Some ethnic groups have become increas- ingly divided due to border adjustments or were placed under "foreign rule" in the territory of another state. Such events are likely to be viewed as a se- vere injustice, which may motivate demands for secession or reunification (Cederman et al., 2019).3 As a second possibility, groups may also retroactively construct nationalist claims based on past historical formations. In this scenario, border precedents enable groups to establish claims that are tied to previous states, kingdoms or imperial legacies, which can serve to motivate demands for self-determination. Historical claims are fairly common among today’s separatist movements, as illustrated by the Karenni in Myanmar, the Bakongo in Angola or the Ewe in Ghana, who have all called for the restoration of pre- colonial kingdoms (Minahan, 1996; Roth, 2015). Although such claims are often based on questionable historical evidence, they serve a clear political purpose. First, they serve to underline a group’s proclaimed identity and historical roots, strengthening its case for self-determination (White, 2000). Second, historical borders may make it easier for groups to develop a shared definition of the homeland territory they are willing to fight for (Goemans, 2006). These two aspects facilitate a group’s mobilization around a common goal and can play an essential role in establishing and reinforcing national identities. In historically unstable regions, border precedents may therefore provide groups with the motives and means to claim territory and push for border adjustments. We may therefore expect the following:

Hypothesis 6.2 Historical border instability increases the risk of domestic territorial disputes.

Whereas the present discussion has so far suggested that past border insta- bility increases the potential for new territorial disputes, it is unclear whether this translates into actual instances of border change. In particular since the end of World War II, states have cooperated internationally and regionally to preserve existing borders, often precisely because their own borders were seen as vulnerable to internal and external challenges (Touval, 1972; Herbst, 1989; Zacher, 2001). If these efforts have generally been effective, it could be that despite a greater risk of disputes in historically unstable regions, borders have

3This argument presumes that potential secessionist groups already share cohesive national identities, which is not always the case.

88 6. Vicious cycles: The persistence of border instability remained mostly stable following World War II. It is therefore useful to test the following hypothesis:

Hypothesis 6.3 Historical border instability increases the risk of further instances of border change.

6.3 Research design and data

This section presents the empirical design and data used to test my hypotheses. I start with an overview of the main inferential challenges I aim to address in the present analysis, after which I discuss the construction of my data and the specification of my main models.

6.3.1 Inferential challenges

Examining the relationship between border change and territorial disputes is challenging, as it involves a whole range of inferential obstacles. Specifically, I identify four major challenges in terms of research design, which I label as non-random assignment, reverse causality, unit change and spatial dependence.

Non-random assignment

A first challenge is that border changes are by no means random events, as they are endogenous to politics. More specifically, border changes and conflict are often part of the very same process, which suggests that regions that experience territorial conflicts may also have been more prone to border changes to begin with, for reasons that may or may not be observed in the data. Ideally, such problems can be addressed using an instrumental variable, which serves as a strong predictor of the treatment, but is otherwise unrelated to the outcome. By leveraging instruments, it is possible to isolate the exogenous effect of a treat- ment from the influence of any potential confounders (Sovey & Green, 2011). Unfortunately, valid instruments are often very hard to find, as it is usually difficult to ensure that an instrument holds up to the exclusion restriction. At least in this case, it seems highly unlikely that there is a variable that strongly predicts border change but has no effect whatsoever on territorial conflict. For lack of a valid instrument, I adopt a three-fold approach. First, I use standard regression models in the main analysis, in which I aim to control for the influence of confounders as far as possible. To this end, I build on the findings from Chapter5, which examines the conditions under which border changes occur, and provides useful insights into geographic attributes that help

89 6. Vicious cycles: The persistence of border instability explain border change. Second, I repeat the main analysis in a sub-sample of the data that consists of Africa alone. The African case is often cited as an example in which borders were drawn almost entirely at random, and therefore arguably resembles a natural experiment, at least with regard to the way in which borders were initially drawn (Michalopoulos & Papaioannou, 2016; Goemans & Schultz, 2017).4 A re-analysis in this particular region should therefore help to further reduce observed and unobserved confounding (Rosenbaum, 2010). In a third step, I use matching to further reduce confounding by ensuring that my units of analysis are more comparable with regard to the observed covariates.

Reverse causality

A second, related challenge is that the causal arrow between border change and conflict may point both ways. While border changes may increase the risk of territorial conflict, the opposite may be true as well. For example, it is at least plausible to assume that some borders may have been adjusted in anticipation of impending conflicts. This makes it difficult to determine the direction of causal effects. Although I have no airtight solution for this problem, it can be addressed in part by creating a "temporal buffer" between the explanatory variable and the outcome, to ensure that the former is always observed prior to the latter. I do this by relying on a cross-sectional research design, in which I examine the effect of border changes that occurred prior to the outbreak of World War II on conflicts in the post-1946 period. It should be noted that by excluding border changes during and after World War II, it is likely that I underestimate the overall effects of border instability. However, the key advantage of this approach is that it helps to address the most obvious problems of reverse causality.

Unit change

A third problem in this particular analysis is that border changes may not only affect the outcome, but can also redefine the units of analysis themselves. Border changes often coincide with the creation or breakup of states, which makes it difficult to study the effects of such changes in a conventional state-centric research design. For example, we may assume that some disputes in today’s Balkan region can be traced back to instability under the Austro-Hungarian or Ottoman Empires, but such lasting effects cannot easily be captured in a

4Although some authors have argued that the randomness of African borders has been overstated (Brownlie & Burns, 1979; Englebert et al., 2002), it is clear that borders on the African continent are more exogenous than in other world regions.

90 6. Vicious cycles: The persistence of border instability conventional research design in which variables are coded on the level of states. In general, it is difficult to examine the effects of historical variables that operate across long periods of time, and across the formation and breakup of states. A similar problem exists for other sub-national entities, such as administrative units or ethnic groups, which are generally defined within existing states. The question then is how to address this problem. Previous studies on long-term historical developments have adopted two general approaches to the problem of unit change: A first set of studies uses currently existing states as their spatial referents, and projects their borders backward in time to study the effects of historical variables on present-day outcomes (Wimmer & Min, 2006; Fearon & Laitin, 2014; Dincecco, Fenske, & Onorato, 2019). Although this seems like a fairly straight-forward solution, it has its own drawbacks that make it unsuitable for the purpose of this analysis. For one, projecting existing states backward in time does not solve the issue that the state units themselves are often created as part of the processes being studied, which can make it very difficult to interpret statistical findings. More generally, aggregating historical data to a fixed "grid" of contemporary states may be useful for some purposes, but is entirely unsuitable if the goal is to study historical border changes, as it would defeat the purpose of the study. Lastly, a general issue with country-level studies is that they are likely to "over- aggregate" information (Cederman & Gleditsch, 2009). Border changes and territorial disputes are both highly local phenomena, and analyzing them on the level of states may blur or discard important patterns of sub-national variation (Goemans & Schultz, 2017). A second approach sidesteps the issues of backward-projection altogether, by examining long-term historical processes on the level of arbitrary grid- cells (e.g., Besley & Reynal-Querol, 2014; Depetris-Chauvin, 2015; Abramson & Carter, 2016). This approach consists of dividing up an area into cells of a given size, and then measuring all variables of interest within these cells. The advantage is that the resulting units remain constant over time, and are completely exogenous to the independent and dependent variables. In the main analyses, I follow this approach, using arbitrary grid-cells as my unit of analysis. Despite its advantages, the cell-based approach also comes with its own challenges. For one, because grid cells are completely arbitrary, it is difficult is to determine their "optimal" size. At what geographic scale do processes such as border change and territorial conflict play out? Using cells that are too small risks artificially inflating the number of observations, which could lead to double-counting of the same observations, drive down standard errors and increase the risk of false positives (Schutte & Donnay, 2014). Moreover, because

91 6. Vicious cycles: The persistence of border instability smaller cells lie close together, it is more likely that observations are spatially correlated, which could introduce further distortions. On the other hand, using cells that are too large increases the risk of over-aggregating information, and is more likely to result in spatial measurement errors. Another, related problem is that spatial measurements, and inferences based on them may be sensitive to the size and shape of units in which measurements are made. This problem is known as the Modifiable Areal Unit Problem (MAUP), as discussed most prominently by Openshaw(1984). In order to address both of these challenges, it is necessary to define an adequate size for grid cells, and to ensure that the findings are robust to changes in the size and shape of units. To this end, I set the size of each grid cell to 50’000 square kilometers in the main analysis, which is smaller than the median size of today’s states, but larger than the median size of disputed areas. To rule out potential distortions due to MAUP, I replicate the main analyses across a set of alternative grids, in which cell sizes range from 20’000 to 80’000 square kilometers.

Spatial dependence

A fourth challenge is that the results of the present analysis may be biased if it fails to account for spatial dependence. While standard regression models make the assumption that individual units are independent, this is often not the case. As units such as states, groups or individuals interact in space, it is likely that they influence each other in numerous ways. As previous studies have shown, ignoring such dependencies can bias estimates and can produce overconfident standard error estimates, which in turn can lead to false inferences (Wakefield, 2003; Ward & Gleditsch, 2007; Franzese & Hays, 2007). In dealing with spatial dependence, we can either treat it as the subject of interest and use the appropriate tools to study them, or treat it as a nuisance, and attempt to remove its influence on the results. I here take the latter approach, for which I rely on spatial filtering. This is a relatively new method that uses a spatial variant of principle component analysis, which is used to capture the spatial dependence components in the model in a set of vectors that can be added to the model to “control” for spatial dependence (Dray, S Legendre & Peres-Neto, 2006; Tiefelsdorf & Griffith, 2007).

6.3.2 Data

To test my arguments, I rely on geospatial data on border change and territorial disputes. Data on borders is taken from the CShapes 2.0 dataset, which covers all major border changes since 1886. Data on territorial disputes between states

92 6. Vicious cycles: The persistence of border instability is taken from the Mapping Interstate Territorial Conflict (MITC) dataset by Schultz(2015b). The MITC dataset is based on a list of disputes that was originally compiled by Huth and Allee(2002) and covers the period 1946-2001. The authors define disputes as situations that involve "either a disagreement between states over where their common homeland or colonial borders should be fixed, or, more fundamentally, the dispute entails one country contesting the right of another country even to exercise sovereignty over some or all of its homeland or colonial territory." (Huth, 1996, p. 19).5 To provide a preview of the MITC dataset, Figure 6.1 (left) plots interstate territorial disputes in South Asia since 1946. For territorial disputes within states, I rely on GeoSDM, a new dataset that is part of an ongoing data collection effort. GeoSDM maps territorial claims made by self-determination movements between 1946 and 2012, based on the SDM dataset by Sambanis et al.(2018). SDM covers a total of 466 movements across the world that have advanced claims that range from moderate demands for increased autonomy within the state to outright secession or irredentism.6 The goal of GeoSDM is to precisely identify and geocode the territories claimed by self-determination movements in domestic territorial disputes. The coding procedure and the contents of the dataset are described in more detail in the appendix. The current version of the dataset covers all areas claimed by groups that have either called for the creation of their own state, or have sought to secede and unite with another state. The analysis is therefore limited to se- cessionist and irredentist claims, which are also most directly relevant to my argument. Figure 6.1 (right) shows a preview of GeoSDM, showing secessionist and irredentist claims in South Asia since 1946.

6.3.3 Creating grid cells

As noted above, I conduct my main analyses on the level of grid cells. Previous studies have used square grid cells defined within a planar coordinate system (e.g., Besley & Reynal-Querol, 2014; Depetris-Chauvin, 2015; Abramson & Carter, 2016). One drawback of this approach is that any attempt to project the globe onto a 2-dimensional surface involves considerable distortion. As a result, grid cells created on a plane can greatly vary in their actual geographic

5In this chapter, I focus exclusively on cases in which the core territory of states is the subject of a dispute, which MITC codes as homeland disputes. This makes up the majority of all disputes covered by the data, but MITC also records some colonial disputes that were dropped from the analysis. 6Claims can also vary within the movement over time.

93 6. Vicious cycles: The persistence of border instability

Figure 6.1: Territorial disputes in South Asia since 1946. Left: Interstate ter- ritorial disputes. Right: Domestic territorial disputes. Dark colors indicate overlapping claims. size.7 To avoid major variation in cell sizes, I use a Voronoi tessellation method that defines cells within their local projection. The Voronoi method uses a set of points to divide a plane into a set of polygons that contain one point each, while ensuring that every location within the polygon is closer to that point than to any other point. Using a set of local Universal Transverse Mercator projections, I create a grid of points that are separated by approximately equal distances, which results in a grid of equal-area polygons. Figure 6.2 shows the grid of cells used in the main analysis, which is set to a size of 50’000 square kilometers. In the appendix, I provide a more detailed illustration of the Voronoi tessellation method.

Figure 6.2: Equal-area Voronoi cells (50’000 km2)

7For example, when using a 1 by 1 degree rectangular grid, a cell located at the equator covers an area that is roughly 7.5 times as large as a cell located at the 83rd parallel north, the northernmost point covered by CShapes 2.0.

94 6. Vicious cycles: The persistence of border instability

6.3.4 Measuring border change

A key task in this study is to measure historical border instability on the level of grid cells, which can be done in several ways. For example, Abramson and Carter(2016) develop a measure of historical border variability, which records variation in the density of international borders within a grid cell over a given historical period. This measure is obtained by overlaying land borders with grid cells, and then calculating border density within each cell in every year. The yearly measures of border density are then used to calculate the overall variability of borders during a given period. Although this approach offers a useful operationalization of border change, one drawback of the resulting measure is that it only picks up border changes for cells that overlap with current or previous borders. For all other cells, border density and variability are set to zero. However, many cells that do not coincide with international borders may still be affected by historical border changes as they may belong to larger territories that were transferred from one state to another. As a result, the border variability measure is likely to underestimate the full extent of territorial changes. To address this, I rely on a simple count variable that records a change each time a cell is transferred from one state to another or if part of the cell is transferred. This results in a more comprehensive measure of border change that is also easier to interpret than the density-based measure. In one of the robustness tests however, I also rely on the border variability measure to see whether it yields the same results.

6.3.5 Model specification

In the main analyses, I use a set of simple cross-sectional models that regress post-1946 disputes on pre-World War II border instability. The basic model takes the following form:

Yi = βXi + γZi + ec

The dependent variable records post-1946 territorial disputes for each grid cell i.8 The main explanatory variable X measures border instability in the pre-World War II period. Z is a vector of control variables. To account for dependencies between cells that fall into the same states, all models use robust

8To code disputes on the level of grid cells, I overlay the grid cells with disputed areas, and code dummy and count measures based on their spatial overlaps. To reduce distortions due to spatial measurement errors, I only code disputes for cells that cover at least 10 percent of the disputed area.

95 6. Vicious cycles: The persistence of border instability standard errors that are clustered on the country level in 1946.9 My selection of control variables is guided by two main concerns. First, one likely source of omitted variable bias is that certain territories are contested for economic or strategic reasons, which could drive repeated conflicts and border adjustments within the same areas (Abramson & Carter, 2016). Second, another potential source of confounding is that borders in some regions are more durable than others, for example due to favorable geographic conditions, which should make both border change and disputes less likely. To address these concerns, I rely in part on the findings from Chapter5, which examines the drivers of border change, but also include a set of additional confounders that are relevant in this particular context. The first variable, Historical border density is used to distinguish between cells that overlap with currently existing or previous historical borders and cells that do not. This is an important distinction, as areas located at a border are likely more prone to territorial changes and disputes than cells that were at all times far removed from any international border. Following Abramson and Carter(2016), I code this variable by calculating the density of borders within a cell in each year, and then average these values over the entire pre-WWII period. A second variable records the log of a cell’s estimated Population density at the start of the historical period, based on gridded historical population data from Klein Goldewijk et al.(2011). I further include a measure of a cell’s Ethnic Fractionalization, which I calculate using ethnic settlement areas from the GREG dataset (Weidmann, Rod, & Cederman, 2010) and population estimates from the HYDE dataset.10 To account for the strategic and economic value of territory, I include three measures of Agricultural Suitability (FAO/IIASA, 2011), as well as a count for the number of Mineral Deposits (USGS, 2005), and Oil Fields (Lujala, Rød, & Thieme, 2007). I also add a measure of Terrain Ruggedness, using data by (Danielson & Gesh, 2011). I code a logged variable for the length of Watershed boundaries in kilometers (Lehner, Verdin, & Jarvis, 2006), and calculate a logged measure of River length (Natural Earth, 2015). Another dummy variable records whether a cell was under Colonial rule at any given time, as coded by CShapes 2.0. To account for any omitted spatial variables, I add controls for each cell’s location, using the Longitude and Latitude of each cell’s centroid. Lastly, I estimate an additional model that includes region fixed effects, in order to further reduce omitted variable bias.11 Summary statistics of

9Cells are coded as part of the country they share the largest spatial overlap with. 10In each year, I intersect GREG polygons with country borders and aggregate gridded population estimates to the resulting polygons. This allows me to compute population-weighted ethnic fractionalization estimates, using geographic data. 11Regions are defined as follows: Western Hemisphere, Europe, Africa, Asia, Australia and

96 6. Vicious cycles: The persistence of border instability all variables are shown in Table B1 in the appendix to this chapter.

6.4 Results

Having presented my research design, I now turn to the results. I start with a discussion of the three main analyses used to test my hypotheses, after which I summarize the robustness tests that were conducted in addition to the main analysis.

6.4.1 Interstate territorial disputes

The first analysis examines the relationship between historical instability and interstate territorial disputes, as shown in Table 6.1. Models 1-3 are logit models in which the dependent variable records the incidence of interstate territorial disputes after 1946. Models 4 and 5 are negative binomial models that use the total number of disputes as the dependent variable. In all models, I find that pre-World War II border instability is associated with a higher likelihood of ter- ritorial disputes between states in the post-war period, as stated by Hypothesis 6.1. Aside from the mere presence or absence of disputes, Models 4 and 5 also show that the total number of disputes also tends to be higher in historically unstable regions. Moreover, a comparison between Models 1, 2 and 3 shows that adding controls and fixed effects does not substantively alter the coefficient estimates for the explanatory variable, which can be seen as an indication that the threat of omitted variable bias may not be too severe in this case (Altonji, Elder, & Taber, 2005). Turning to the control variables, the estimate for historical border density is positive and significant, indicating that disputes are more likely to erupt in areas located at current or previously existing borders. I find some evidence that densely populated areas are more dispute-prone, but this effect disappears in the models that include region fixed effects. Somewhat surprisingly, I find that areas with greater agricultural suitability are less likely to be disputed, but I have no clear theoretical explanation for this finding. The results also suggest that areas in rugged terrain or those containing oil fields have a higher dispute propensity, although the latter estimate is not significant across all models. Finally, disputes are more likely in areas formerly under colonial rule, which is perhaps not surprising considering that many newly independent states have inherited borders that were already disputed among colonial powers (Day, 1982).

Oceania.

97 6. Vicious cycles: The persistence of border instability

Table 6.1: Past border instability and interstate territorial disputes

(1) (2) (3) (4) (5) Incidence Incidence Incidence Count Count Past border chg. 0.425∗∗∗ 0.375∗∗∗ 0.392∗∗∗ 0.408∗∗∗ 0.402∗∗∗ (0.110)(0.110)(0.117)(0.094)(0.096) Hist. border dens. 0.272∗∗∗ 0.289∗∗∗ 0.271∗∗∗ 0.240∗∗∗ 0.204∗∗∗ (0.041)(0.035)(0.035)(0.029)(0.030) Population density, log 0.341∗∗∗ 0.138 0.320∗∗∗ 0.085 (0.114)(0.132)(0.118)(0.130) ELF (Greg) −0.059 −0.243 0.331 0.242 (0.473)(0.444)(0.379)(0.336) Agr. suitability −0.258∗∗∗ −0.167∗ −0.254∗∗∗ −0.142∗ (0.082)(0.092)(0.062)(0.074) Mineral deposits −0.066 −0.056 −0.042 −0.033 (0.048)(0.050)(0.058)(0.052) Oil fields 0.130∗ 0.109 0.182∗∗∗ 0.146∗∗ (0.077)(0.079)(0.058)(0.063) Terrain ruggedness, log 0.364∗∗ 0.277∗ 0.423∗∗ 0.332∗∗ (0.144)(0.150)(0.168)(0.152) Watershed length, log 0.006 −0.008 −0.013 −0.018 (0.037)(0.038)(0.036)(0.035) River length, log −0.050 −0.077∗ −0.031 −0.040 (0.049)(0.047)(0.054)(0.046) Colonial past 0.920∗∗ 1.398∗∗∗ 0.863∗ 1.306∗∗∗ (0.368)(0.392)(0.483)(0.384) Latitude 0.008 −0.011 0.009 −0.013 (0.007)(0.008)(0.008)(0.008) Longitude 0.004 −0.001 0.003 −0.006 (0.002)(0.005)(0.002)(0.004) Intercept −3.822∗∗∗ −5.806∗∗∗ −5.633∗∗∗ −5.993∗∗∗ −6.390∗∗∗ (0.320)(0.997)(1.152)(1.213)(1.183) Region FE no no yes no yes Log Likelihood −512.478 −466.147 −447.036 −616.739 −589.395 Observations 2776 2776 2776 2776 2776 Notes: Logit Models (1-3), Negative Binomial Models (4-5). Unit of analysis: Grid cells (50’000 Sq Km). Regions consist of: Western Hemisphere, Europe, Africa, Asia, Australia and Oceania. Robust standard errors clustered on the country level in 1946. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

98 6. Vicious cycles: The persistence of border instability

Table 6.2: Past border instability and domestic territorial disputes

(1) (2) (3) (4) (5) Incidence Incidence Incidence Count Count Past border chg. 0.381∗∗∗ 0.368∗∗∗ 0.324∗∗∗ 0.324∗∗∗ 0.278∗∗∗ (0.123)(0.111)(0.119)(0.079)(0.082) Hist. border dens. 0.179∗∗∗ 0.106∗∗∗ 0.089∗∗ 0.089∗∗∗ 0.070∗∗ (0.034)(0.033)(0.036)(0.028)(0.029) Population density, log 0.534∗∗∗ 0.347∗∗∗ 0.457∗∗∗ 0.296∗∗∗ (0.123)(0.130)(0.097)(0.098) ELF (Greg) 1.540∗∗∗ 1.523∗∗∗ 1.360∗∗∗ 1.284∗∗∗ (0.366)(0.345)(0.383)(0.353) Agr. suitability −0.076 −0.011 −0.051 0.014 (0.088)(0.088)(0.079)(0.081) Mineral deposits 0.009 0.005 0.009 0.003 (0.024)(0.024)(0.026)(0.028) Oil fields 0.084 0.051 0.065 0.039 (0.060)(0.063)(0.050)(0.053) Terrain ruggedness, log 0.227∗∗ 0.258∗∗ 0.221∗∗∗ 0.241∗∗∗ (0.113)(0.114)(0.080)(0.085) Watershed length, log 0.030 0.020 0.059∗∗ 0.051∗ (0.031)(0.031)(0.028)(0.026) River length, log −0.024 −0.006 −0.037 −0.026 (0.047)(0.043)(0.044)(0.038) Colonial past 0.727∗∗ 0.857∗∗ 0.669∗∗ 0.739∗∗ (0.321)(0.409)(0.286)(0.371) Latitude 0.002 −0.012 0.003 −0.009 (0.006)(0.008)(0.005)(0.007) Longitude −0.000 −0.012∗∗∗ 0.001 −0.009∗∗∗ (0.002)(0.004)(0.002)(0.003) Intercept −2.667∗∗∗ −5.154∗∗∗ −6.740∗∗∗ −5.028∗∗∗ −6.464∗∗∗ (0.225)(0.790)(1.004)(0.640)(0.885) Region FE no no yes no yes Log Likelihood −899.036 −781.762 −752.550 −1029.738 −1003.743 Observations 2776 2776 2776 2776 2776 Notes: Logit Models (1-3), Negative Binomial Models (4-5). Unit of analysis: Grid cells (50’000 Sq Km). Regions consist of: Western Hemisphere, Europe, Africa, Asia, Australia and Oceania. Robust standard errors clustered on the country level in 1946. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

99 6. Vicious cycles: The persistence of border instability

Table 6.3: Past border instability and border changes

(1) (2) (3) (4) (5) Incidence Incidence Incidence Count Count Past border chg. 0.399∗∗ 0.428∗∗ 0.589∗∗∗ 0.333∗∗∗ 0.378∗∗∗ (0.161)(0.174)(0.197)(0.104)(0.101) Hist. border dens. 0.120∗∗∗ 0.101∗∗ 0.104∗∗ 0.092∗∗∗ 0.085∗∗∗ (0.041)(0.041)(0.045)(0.027)(0.030) Population density, log 0.064 −0.104 0.073 −0.076 (0.139)(0.114)(0.111)(0.090) ELF (Greg) 0.174 −0.055 0.029 −0.172 (0.486)(0.464)(0.348)(0.329) Agr. suitability 0.024 0.086 0.011 0.080 (0.135)(0.130)(0.096)(0.095) Mineral deposits 0.039 0.059∗ 0.039∗ 0.046∗∗ (0.037)(0.033)(0.023)(0.020) Oil fields 0.079 0.040 0.074∗ 0.037 (0.077)(0.078)(0.043)(0.044) Terrain ruggedness, log 0.170 0.142 0.103 0.092 (0.255)(0.193)(0.175)(0.135) Watershed length, log 0.079 0.070 0.050 0.041 (0.050)(0.050)(0.038)(0.036) River length, log 0.000 0.008 −0.009 0.002 (0.110)(0.098)(0.076)(0.067) Colonial past 1.433∗ 1.842∗∗∗ 1.169∗∗ 1.372∗∗∗ (0.785)(0.656)(0.501)(0.404) Latitude 0.034∗∗∗ 0.022 0.024∗∗∗ 0.013 (0.012)(0.014)(0.008)(0.010) Longitude 0.003 −0.012∗ 0.003 −0.009∗∗∗ (0.003)(0.006)(0.003)(0.003) Intercept −2.076∗∗∗ −5.170∗∗∗ −6.464∗∗∗ −4.170∗∗∗ −5.420∗∗∗ (0.369)(1.655)(1.054)(1.115)(0.821) Controls no yes yes yes yes Region FE no no yes no yes Log Likelihood −1175.211 −1050.254 −989.600 −1371.911 −1318.761 Observations 2776 2776 2776 2776 2776 Notes: Logit Models (1-3), Negative Binomial Models (4-5). Unit of analysis: Grid cells (50’000 Sq Km). Regions consist of: Western Hemisphere, Europe, Africa, Asia, Australia and Oceania. Robust standard errors clustered on the country level in 1946. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

100 6. Vicious cycles: The persistence of border instability

6.4.2 Domestic territorial disputes

Having found overall support for Hypothesis 6.1, which stated that past border instability increases the risk of interstate territorial disputes, I now turn to its effects on domestic territorial disputes. As stated by Hypothesis 6.2, historically unstable borders should also increase the risk of territorial disputes within states. To test this claim, I replicate the analysis shown in the previous section but use the incidence and number of domestic territorial disputes as dependent variables. Table 6.2 shows the results. In support of Hypothesis 6.2, I find that regions with greater border instability in the pre-World War II period have been more prone to secessionist and irredentist disputes in the post-war period. A comparison of the coefficient estimates in Models 1, 2 and 3 again shows that the inclusion of controls and fixed effects does not substantially alter the results, which alleviates some concerns regarding omitted variable bias. The estimates for the control variables also reveal a number of interesting insights. As in the previous analysis, regions that coincide with current or previous borders have been more prone to domestic disputes, as shown by the estimated coefficients for historical border density. Moreover, I find that secessionist and irredentist disputes are more likely to emerge in densely populated areas, as well as in ethnically heterogeneous regions and in areas with rugged terrain. I also find that secessionist disputes are more likely in areas that were formerly under colonial rule. Lastly, I find some evidence that cells that coincide with watershed boundaries are more likely to experience domestic disputes, but this finding is not robustly significant across all models.

6.4.3 Border change

In a third analysis, I examine whether past border instability is also associated with an increased risk of future border changes. I replicate the previous analysis, but use the incidence and number of post-World War II border changes as the dependent variable. The results are shown in Table 6.3. In support of Hypothesis 6.3, I find that historically unstable regions have had an increased likelihood of border change in the post-1946 period. Turning to the control variables, I find that territorial changes have occurred more frequently in areas that were formerly under colonial rule, and find only mixed evidence that border changes are more likely in areas that contain valuable minerals. To allow a substantive interpretation of my main findings, I calculate the predicted probabilities of the post-1946 incidence of interstate and domestic- level disputes and border changes as a function of pre-World War II border

101 6. Vicious cycles: The persistence of border instability instability, holding all other variables at their mean or mode.12 Predictions are limited to cells with up to 5 border changes, as only a small number of cells has experienced a higher number. Figure 6.3 shows the results. As the number of past border changes increases, we see a substantial increase in the risk of interstate and domestic territorial disputes over the entire post-1946 period. We see a similar increase in the risk of post-1946 border changes, although the confidence intervals are quite large, suggesting that the true effect size may not be as pronounced.

Figure 6.3: Predicted probabilities of main outcome variables.

6.4.4 Robustness

To assess the robustness of my results, I conduct a number of additional analyses, which are shown in the appendix and are briefly discussed here. First, I replicate the main findings using two alternative measures of border instability. The first is a replication of the border variability measure used by Abramson and Carter (2016), which records variation in border density over the same period.13 The second is a logged measure of the number of years since the last border change between 1886 and 1939.14 The results, shown in Table 6.3, are overall consistent with my previous findings: Cells with greater historical border variability have been more likely to experience interstate and domestic territorial disputes, and have also experienced more border changes in the post-1946 period. I further

12Binary variables are set to their mode, continuous variables are set to their mean. Predicted probabilities and 95% confidence intervals are calculated using 1000 simulated draws from a multivariate normal distribution (King, Tomz, & Wittenberg, 2000), based on the estimates and clustered standard errors from Model 2 in each table. 13Following Abramson and Carter(2016), is use the following formula to calculate border variability within a given cell over period T: q T B.var = ∑t=1 Density − Density. 14For cells that did not experience any changes, the count starts in 1886.

102 6. Vicious cycles: The persistence of border instability

find that the risk of interstate and domestic territorial disputes declines as the time since the last border change increases. This second finding does not only confirm the previous results, but also provides some support for the idea that the risk of territorial disputes decreases the longer borders have remained in place. I find a similar relationship between the time since the last border change and post-1946 border changes, but the corresponding coefficient fails to reach conventional levels of significance.

Figure 6.4: Coefficient estimates for past border changes across varying cell sizes

In a second robustness test, I check whether my findings are sensitive to changes in the size and shape of grid cells used in the analysis. I use a series of alternative datasets in which cell sizes range from 20’000 to 80’000 square kilometers and then estimate a logit model that includes all controls and fixed effects with each dataset. The resulting coefficient estimates are shown in Figure 6.4, which confirms that the estimates remain robust across all cell sizes. In a third analysis, I further deal with potential bias due to the non-random nature of border change. Specifically, I replicate the main analysis in a sub- sample of the data that consists of Africa only. As noted earlier, the African case arguably resembles a natural experiment due to the relatively arbitrary way in which its borders were drawn. Re-analyzing the data in this particular case may therefore help to remove observed and unobserved confounding (Rosenbaum, 2010). The results remain consistent across all three outcome variables, as shown in Table B3 in the appendix. In addition, I repeat the analysis in another sub-sample of the data, which consisting of all regions except for Europe. This is useful to account for the possibility that the findings are largely driven by a small number of cases in Europe, which has experienced the most intense history of territorial change and conflict. The results remain robust across all sub-samples, as shown in Table B4. A fourth analysis uses matching to further deal with observed heterogeneity between observations. Matching is a useful way to reduce confounding, as it

103 6. Vicious cycles: The persistence of border instability creates a smaller subset of the data in which "treated" and "control" units are as similar as possible with regard to the observed covariates.15 Figure B2 summa- rizes the results of the matching procedure, showing a clear improvement in covariate balance after matching. A reanalysis of the data within the matched sample supports the previous findings, as shown in Table B4. In a fifth robustness test, I further address concerns of omitted variable bias. One main concern in the present analysis is that border changes and territorial disputes may both be driven by recurring conflict affecting the same areas. As such, the persistence of border instability may simply be the result of conflict persistence. Ideally, I would be able to account for this using fine- grained data on the location of past conflicts prior to 1946. Unfortunately, such data is currently unavailable. As an alternative, I rely on historical data on interstate and domestic conflicts from K. Gleditsch(2004), which I use to code past conflicts on the level of country polygons between 1886 and 1946.16 This results in a time-varying dataset that provides approximate information about the location of previous wars. I then use this dataset to code a measure of the logged number of war years within each grid cell, which I use as another control variable. The results, shown in Table B5 are robust to the inclusion of this additional control. In the sixth and final analysis, I use spatial filtering to account for spatial dependence. The goal of this method is to capture the spatial dependence components in the model as a set of synthetic covariates, which can then be added to the model to effectively account for spatial dependence. This is done by conducting a principal component analysis on a spatial connectivity matrix. This procedure returns a set of eigenvectors with the largest achievable Moran’s I coefficient of spatial autocorrelation.17 As with most spatial econometric tech- niques, spatial filtering works best for linear models with continuous outcome variables, and the implementation for discrete-outcome models is much more demanding, and the appropriate methods are still being developed. I therefore use spatial filtering in combination with OLS models, which treat discrete out- comes as if they were continuous. Although this is not an ideal solution, it is still preferable to ignoring spatial dependence altogether and can still provide

15In this case, the treated units are cells that have experienced at least one border change prior to World War II. I use the genetic matching method introduced by Diamond and J. Sekhon (2013), which uses a search algorithm to maximize covariate balance based on the propensity score and Mahalanobis distance metrics. 16I assign wars to the country in which fighting has taken place, which requires some additional coding in the case of interstate conflicts. 17Dray, S Legendre and Peres-Neto(2006) and (Tiefelsdorf & Griffith, 2007) have shown that adding these eigenvectors to standard statistical models can effectively remove spatial autocorrelation from the model.

104 6. Vicious cycles: The persistence of border instability useful insights into the degree to which it may drive the results. I use a simple connectivity matrix that links each cell to its primary and secondary neighbors, as illustrated in Figure B3. I then use this matrix to conduct spatial filtering,18 after which I replicate the main findings. The results, shown in Table B7 further support my overall findings.

6.5 Conclusion

This chapter has presented the first global assessment of the persistence of border instability, using geocoded data on international borders since 1886. In line with my arguments, I find that regions with unstable borders prior to World War II have remained more prone to interstate and domestic territorial disputes in the post-1946 period. Moreover, these areas have also remained more likely to experience further border changes. My findings are robust to a large number of additional tests that help alleviate concerns relating to omitted variable bias, spatial dependence or the arbitrary definition of my units of analysis. Although the finding that past border changes are related to future disputes and instability may seem obvious, I argue that it is nonetheless important to empirically assess the persistence of border instability. In doing so, we are better able to assess whether some regions are more prone to territorial conflict than others. Moreover, it also contributes to explaining the emergence of territorial disputes. Most importantly, while some studies have argued that border changes can, under certain circumstances, lead to stable outcomes (e.g. Tir, 2003), the findings presented in this chapter suggest that border changes have, at least on average, been associated with lasting instability. This chapter has only scratched the surface when it comes to studying the effects of border instability. As such, there are a number of important limitations that remain to be addressed. First, I have sketched out a number of causal mechanisms that link past border instability to disputes, but have not tested any of these mechanisms. Ideally, we would like to know not only if instability persists, but also how this effect arises. Second, I have not sufficiently examined the temporal dimensions of my arguments. The path dependent logic of border stability suggests that border precedents that have been in place for long periods of time are more likely to have lasting effects, while the overall importance of previous borders is likely to decline as the time since their removal passes

18I use a program developed by (Murakami & Griffith, 2018), which extracts a set of eigen- vectors from an unweighted connectivity matrix, and then adds them to the model, using an algorithm that optimizes the AIC measure of model fit, while minimizing multicollinearity. Overall, the models include between 63 and 70 eigenvectors.

105 6. Vicious cycles: The persistence of border instability

(Abramson & Carter, 2016). Although this chapter has provided some evidence for temporal effects in showing that the risk of disputes decreases with the time since the last border change, more needs to be done to fully unpack these temporal dynamics. Third, I have focused almost entirely on territorial disputes as the outcome of interest, but have yet to examine the violent escalation of these disputes. For example, it would be useful to examine whether disputes in historically unstable regions are more likely to escalate than disputes in regions that have remained mostly stable. These are all important issues that I hope to address in future research.

106 Chapter 7

Dangerous precedents? Border change, spillover and new territorial disputes

7.1 Introduction

Is border instability contagious? A common concern in international politics is that territorial changes in one region could destabilize borders elsewhere, as they may set precedents that encourage other actors to challenge the status quo. For example, when Russia invaded Ukraine and annexed Crimea in 2014, observers feared that these clear violations of international rules could lower the bar for territorial conflicts elsewhere (Burke-White, 2014). Interestingly, Russia sought to justify its own actions by comparing them to Western support for the independence of Kosovo, arguing that the latter had set a precedent for the secession of Crimea (Ambrosio, 2016).1 More recently, the United States government faced intense criticism over its decision to recognize Israel’s an- nexation of the Golan heights, as critics argued that this move may set another troubling precedent with potentially destabilizing consequences (Borger, 2019). The question of how border changes affect regional stability has often been discussed in the literature. A prevailing view holds that the international system is governed by a set of rules and norms that are aimed at preserving the territorial status quo (Zacher, 2001; Atzili, 2012; Goertz et al., 2016). Based on this view, we may expect that actions that violate international rules may encourage further rule-breaking behavior, which could destabilize the system as a whole. If one secessionist minority is granted independence, this may

1Russia has argued that it has not annexed Crimea but the peninsula voted to secede, after which it decided to join Russia.

107 7. Border change, spillover and new territorial disputes encourage other aggrieved groups to call for a state of their own (Fearon, 2004; Walter, 2009). Similarly, if one state successfully redraws borders in its favor, others may attempt to do the same (Touval, 1972). Despite the prevalence of such claims, the potential spillover effects of border change have rarely been examined empirically. As a result, we lack a clear understanding of whether, and under what conditions border change may fuel regional instability. While some studies have examined domestic "domino effects" in the context of secessionist conflict2 (Hale, 2000; Forsberg, 2013; Bormann & Savun, 2018), the broader regional implications of border change remain understudied. This chapter aims to help fill this gap, by examining how border changes between one pair of states may affect the onset of territorial disputes between states elsewhere. I argue that international rules and norms generally impose strong con- straints on territorial revisionism, but the strength of these constraints may vary across time and space, and is in many cases uncertain. Would-be revisionist states base their decision to initiate a dispute on the perceived costs of territo- rial conflict, as well as on their perceived chances of successfully redrawing borders. They base this assessment primarily on recent interactions between other, proximate states that face similar circumstances. Successful instances of border change nearby demonstrate the weakness of international constraints, and can also be used as examples to justify new territorial claims. However, I argue that not all territorial changes are equally disruptive to the international order. In particular, I distinguish between conflictual territorial changes that occurred against the will of the losing side and cooperative changes that were based on mutual agreement. My overall expectation is that conflictual instances of border change are more likely to have destabilizing consequences, as such changes are most at odds with international rules and norms. To test these hypotheses, I use data on territorial changes and interstate territorial disputes from 1816 to 2001 in a dyadic analysis. Drawing on existing data sources, I code a variable that distinguishes between conflictual and coop- erative instances of territorial change. I measure the spillover effects of border change using a series of spatial lags that are based on two complementary definitions of dyadic interdependence. In line with my hypotheses, I find that territorial changes between one pair of states indeed increase the risk of new territorial disputes among other states within the neighborhood. Moreover, this effect is largely driven by conflictual instances of border change, as opposed to

2The term "domino effects" refers to demonstration effects that can emerge as a result of government concessions to a separatist group’s demands, which can inspire other groups to confront the government with similar demands.

108 7. Border change, spillover and new territorial disputes cooperative exchanges of territory. The main findings are robust to the inclusion of dyad- and year fixed effects. However, the results also indicate that the risk of spillover has not remained constant throughout history, as I only find consistent evidence of spillover in the pre-World War II period. In the remaining sections of this chapter, I discuss the relevant literature and present my argument, after which I discuss my research design and present my findings. I conclude with a discussion of the main limitations and tasks for further research.

7.2 Dangerous precedents

Most of international politics is interdependent, in that the behavior of some actors shapes the decision-making of others. Although research on territorial conflict has rarely focused specifically on such indirect effects, they play a cen- tral role in many theories of conflict. For example, the deterrence literature is mainly concerned with how states can prevent their opponents from attacking them or others by issuing credible threats of retaliation (Schelling, 1980; Huth, 1988). According to this view, states aim to maintain a strong reputation to influence the behavior of their opponents. To a large degree, a state’s reputation is derived from its past record in dealing with adversaries (Huth, 1988). The same logic has also been applied to the context of secessionist conflict. Most prominently, Walter(2009) argues that governments are unlikely to grant con- cessions to secessionist groups if they face other potential challengers down the road. If a state concedes to the demands of one group, this may trigger new demands from other groups that perceive the government as weak (Hale, 2000; Walter, 2009; Forsberg, 2013). As a result, the assumption is that states have an incentive to fight their secessionist challengers rather than to accommodate them, even if the territory at stake is not seen as especially valuable (Walter, 2009). Based on similar assumptions, others have argued that states are likely to cooperate with each other to protect themselves from secessionist contagion effects. In particular, it is frequently argued that states that are vulnerable to disintegration are unlikely to support secessions abroad, as these could encourage secession attempts by minorities at home (Saideman, 1997). In the case of Africa, it has been argued that the leaders of newly independent states adopted a set of rules and norms to prevent border changes across the region, in order to ensure their own political survival. Following independence, Africa was viewed as especially prone to territorial conflict, due to the arbitrary nature of its borders, as well as the large ethnic heterogeneity and weak social cohesion of its states. By committing to preserving their inherited colonial borders under

109 7. Border change, spillover and new territorial disputes the uti possidetis norm, African leaders may therefore have avoided a dangerous spiral of revisionism and secession (Zartman, 1966; Touval, 1972; Herbst, 1989; Zacher, 2001). Rather than focusing on the reputation of individual states alone, this second argument emphasizes how shared rules and norms shape the behavior and expectations of states.

More generally, attempts to establish a rules-based territorial order can be traced even further back in time. Similar to states in post-colonial Africa, states in 19th century Latin America agreed to preserve existing colonial borders following their independence. Most newly independent states controlled only parts of their vast territories, and feared annexation attempts based on the prin- ciple of effective occupation. By mutually recognizing their inherited borders, they sought to prevent widespread conflict and to rule out a potential return of European colonialism (Shaw, 1997, p. 98). At around the same time, states in Europe increasingly cooperated in attempts to preserve the regional balance of power and to prevent large-scale conflict (Claude, 1989; Holsti, 1991). This was done most prominently under the Concert of Europe of 1816, which shaped much of international politics throughout the 19th century. The Concert devised a carefully calibrated system of major powers and neutral buffer states, in which any unilateral attempt to redraw borders was seen as a potential threat to the regional order. To preserve the regional balance, states agreed on a set of rules on legitimate state behavior, and committed to act jointly against attempts of revisionism (Holsti, 1991).

Although 19th-century rules and norms may have generally served to reduce the risk of major instability, the overall system was still relatively permissive towards border change, as states viewed war as a legitimate tool of foreign policy and treated annexation and conquest as legitimate outcomes of war. This began to change in the interwar period, as states increasingly adopted rules against territorial changes by force. In particular following World War II, the terms under which border changes were seen as acceptable were defined in increasingly restrictive terms. The new set of international rules was enforced primarily via the threat of sanctions against rule-breakers, and by a policy of non-recognition of de facto annexations and secessions (Zacher, 2001; Goertz et al., 2016; Hathaway & Shapiro, 2017).

Overall, there is widespread agreement that existing borders have been kept in place in part by international rules and norms (Zacher, 2001; Goertz et al., 2016; Fazal & Griffiths, 2014; Hathaway & Shapiro, 2017). Generally speaking, much of international politics is guided by conventions, defined as shared understandings of acceptable and unacceptable behavior that shape common

110 7. Border change, spillover and new territorial disputes expectations about future behavior (Kier & Mercer, 1996).3 Conventions may be written into law, but can also be based on informal norms and customs. Importantly, conventions can be created or destroyed by precedents, defined as "an act or statement that serves, or is intended to serve as an example, reason, or justification for a later one" (Kier & Mercer, 1996, p. 80). A clear violation of international rules can therefore set a precedent, as it may undermine a shared understanding of legitimate behavior, and may be used to justify other, similar behavior. Moreover, states that successfully challenge existing rules may demonstrate to others that the costs of rule-breaking behavior may not be as high as previously assumed. This explains why scholars have frequently argued that border changes could have far-reaching destabilizing consequences. In the case of Africa, Zartman(1966, p. 109) argued that:

"The greatest deterrent to territorial revisionism has been the fear of opening a Pandora’s box. If any one boundary is seriously ques- tioned, why not all the boundaries in Western Africa?"

Following the wave of ethnic conflict in the 1990s, there were widespread proposals to redraw state borders to end civil wars, which gave rise to similar concerns. While some authors claimed that territorial adjustments may be the only feasible way to secure peace (Kaufmann, 1996; Chapman & Roeder, 2007), others have warned that such a policy could set dangerous precedents that may provoke new conflicts elsewhere (Fearon, 2004, e.g.). Rewarding violent groups with a state of their own may provide incentives to other disadvantaged minorities to take up arms. Beyond the risk of new civil wars, Fearon(2004, p. 394) warns that such border adjustments could destabilize international politics more generally:

"Ad hoc use of partition to solve civil wars would undermine an implicit and relatively stable bargain among the major powers, in place since the 1950s. The bargain rests on the expectation that if any one major power seeks to change interstate borders by force, others may follow, to the detriment of the first."

While many authors have expressed general concerns that border changes could destabilize the system as a whole, it still remains largely unclear to what extent and under what conditions this actually holds true. As Kier and

3A convention, which shapes expectations about the behavior of actors based on shared rules and norms, is different from a reputation, which shapes expectations about one particular actor’s behavior, based on its previous actions (Kier & Mercer, 1996). However, conventions can be tied to the reputation of individual actors that commit to enforcing existing rules and norms.

111 7. Border change, spillover and new territorial disputes

Mercer(1996) point out, it is difficult to know which rule violations may set a precedent and which remain an isolated episode. Despite the importance of these questions, only very few studies have systematically studied the spillover effects of border instability. Some studies have examined the contagion of secessionist disputes within states, examining how government concessions to a secessionist minority affect the risk of conflict with other minorities (Hale, 2000; Forsberg, 2013; Bormann & Savun, 2018). However, existing theories suggest that spillover effects may also drive territorial conflicts between states. In this chapter, I make a first attempt at filling this gap, by examining how border changes between one pair of states shape the risk of territorial conflict between states elsewhere.

7.3 The spillover effects of border change

This chapter aims to explain the onset of interstate territorial disputes, which are defined as explicit disagreements between states over territorial sovereignty or over the location of a border (Huth, 1996; Hensel & Mitchell, 2017). Territorial disputes are known to be the main driving force of violent conflict between states, and have furthermore been linked to other adverse outcomes, such as enduring rivalries and losses in international trade (Holsti, 1991; Vasquez, 1993; Goertz & Diehl, 1992; Kocs, 1995; Huth, 1996; Vasquez & Henehan, 2001; Simmons, 2005). While most research has focused on the dynamics and consequences of territorial disputes, fewer studies have paid closer attention to their causes (Huth, 1996; Abramson & Carter, 2016; Goemans & Schultz, 2017; Abramson & Carter, 2018). Among these studies, most have examined state characteristics or location-specific factors as explanations for disputes, but have paid less attention to the international context in which disputes emerge. One recent exception is a study by Abramson and Carter(2018), which shows that disputes are more likely to erupt during periods of systemic instability that are characterized by shifts in the power balance between major powers. In this chapter, I provide another explanation that emphasizes the interde- pendent nature of territorial conflict, arguing that border changes between one pair of states can increase the risk of new territorial disputes elsewhere. My argument focuses primarily on the decision of a potential challenger state to initiate a dispute with one of its neighbors. A state may decide to do so for various reasons, including the symbolic or material value of territory, domestic political considerations, historical border precedents or the absence of clearly defined and recognizable boundaries (Abramson & Carter, 2016; Goemans & Schultz, 2017). Whatever a state’s motives may be, the potential benefits of

112 7. Border change, spillover and new territorial disputes claiming and obtaining territory are often outweighed by the substantial costs of territorial conflict. In addition to the direct risk of war and economic oppor- tunity costs, territorial disputes can entail reputation costs on the international stage (Schultz, 2015a). As noted, the international system is generally averse to actions that un- dermine the territorial integrity of its members. At a minimum, disputes can therefore entail reputation costs for the challenger state. In more extreme cases, a state’s illegitimate behavior can provoke costly sanctions or even military intervention by third states. However, the international community does not respond to rule violations in a clear and consistent way. Absent an overarching enforcement mechanism, the strength and effectiveness of international rules and norms depends largely on the willingness of states to cooperate, and the ability of powerful states to enforce existing rules. In some periods, states may be less willing or able to enforce existing rules (Ambrosio, 2001; Abramson & Carter, 2018). Likewise, constraints on revisionism may vary across regions, due to differences in regional institutions or the varying influence of powerful states in different parts of the world. For would-be revisionists therefore, the costs of initiating a dispute, and the likelihood of successfully redrawing borders remain largely uncertain. One key source of information is provided by recent interactions between other states in the neighborhood, which operate under similar constraints. As such, successful instances of border change nearby can serve as precedents that can be used to justify new disputes, while they also provide information about the expected costs and feasibility of border adjustments. If potential revisionists view the international environment as favorable, they may feel encouraged to act fast (Abramson & Carter, 2018). Therefore, we may generally expect that border changes have the potential to destabilize borders elsewhere:

Hypothesis 7.1 States are more likely to initiate a territorial dispute with their neigh- bors following territorial changes between other neighboring states than absent such changes.

As noted earlier, however, not all territorial changes are created equal. Some changes are the result of bilateral agreements, international arbitration or plebiscites, while others stem from unilateral attempts at conquest, annexation or secession. We can therefore broadly distinguish between cooperative and con- flictual instances of border change. Examples of the first type include the United States’ purchase of Alaska from Russia in 1867, or the peaceful dissolution of the Kingdom of Norway and Sweden in 1905. Examples of the second, conflict- ual type are Germany’s annexation of Alsace-Lorraine from France in 1870, or

113 7. Border change, spillover and new territorial disputes

Kosovo’s secession from Serbia in 2008. In general, the international system has been much more permissive regarding cooperative border changes, which were generally seen as the exclusive affair of the states involved. In contrast, conflictual border changes were often viewed as disruptive or illegitimate, as they were at odds with international rules. As a result, we may expect that overall, conflictual border changes have a greater potential to set dangerous precedents with destabilizing effects:

Hypothesis 7.2 States are more likely to initiate a territorial dispute with their neigh- bors following conflictual territorial changes between other neighboring states than absent such changes.

At this point, it is important to recall that international rules themselves have changed substantially over the course of recent history. Previous studies have distinguished between an old world order that was in place until World War I, and a new order that began to emerge in the interwar period and became firmly established after World War II (Zacher, 2001; Hathaway & Shapiro, 2017). Under the old system, states often committed to certain rules to prevent major instability, but the overall system was much more permissive towards territorial change, even if it occurred under the use of force (Zacher, 2001; Hathaway & Shapiro, 2017). Both systems therefore created different incentive structures that arguably had their own implications for the risk of spillover. Therefore, it is important to pay closer attention to the historical context in which states operate. For this reason, I test my hypotheses both from a long- term historical perspective, and from a more context-specific approach that distinguishes between different historical periods.

7.4 Research design and data

To test my hypotheses, I employ a time-series cross-sectional research design, using dyadic data on interstate territorial disputes by Hensel and Mitchell (2017), and data on territorial changes by Tir et al.(1998). The combined dataset covers border changes and territorial disputes across the world from 1816 to 2001.4 I conduct my analyses on the level of directed dyads, focusing on yearly interactions within pairs of neighboring states. The dependent variable records the initiation of a territorial dispute by a state against its neighbor in a given year. The main explanatory variable is a spatial lag that captures recent territorial

4The ICOW dataset (Hensel & Mitchell, 2017) is the most comprehensive dataset on territo- rial disputes between states, but only provides coverage through 2001. Therefore, the analysis is limited to this period.

114 7. Border change, spillover and new territorial disputes changes between other states in the neighborhood. Specifically, I distinguish between two types of dyadic dependence, which I define as dependence among neighboring dyads and dependence among common member dyads as explained in more detail below. These two dependence structures are used to generate spatial lags on the outcome and explanatory variables, which are then incorporated into the analysis.

7.4.1 Unit of analysis

The first task is choosing the right unit of analysis. Possible choices include a monadic research design that examines the behavior of individual states, a dyadic approach that examines interactions between pairs of states and a directed dyad design, which examines the behavior of a "source" state towards a "target" state in each dyad. I consider the directed dyad approach as the most suitable for two main reasons: First, both the outcome and explanatory variables are inherently dyadic. Territorial disputes occur almost without exception between two states, as do most transfers of territory, which suggests that a monadic design may not be optimal. Second, because my goal is to explain dispute initiation, rather than the mere presence or absence of disputes, a directed dyadic design is preferable to undirected dyads, as only the former identifies the challenger in a territorial dispute.5 Another important concern is the selection of cases to analyze. Many pre- vious studies have included all dyads in the international system by default, resulting in exceedingly large datasets. In any year, the number of dyadic linkages among N states is given by N(N − 1). It is not uncommon for studies to use all dyad-years from 1816 to the present, which would result in 1’912’350 directed dyad-year observations through 2016. Although the computational challenges in dealing with such large datasets are much smaller today, large dyadic datasets can still introduce other problems, such as an increased poten- tial for bias due to dependencies among observations (Minhas, Hoff, & Ward, 2016), or estimation problems due to a small number of "positive" outcomes compared to a vast number of dyadic observations (King & Zeng, 2001; Cook, 2014). An alternative approach assumes that not all dyads are equally relevant, and hence some can be omitted from the analysis (Maoz & Russett, 1993). When studying territorial conflict, we can reasonably argue that border disputes between Ecuador and Bhutan, or Luxembourg and Samoa are highly unlikely,

5Using directed dyads means that each pair of states is observed twice in any given year. As such, dyad ij, in which country i is the initiator and j the target of a potential dispute also appears as dyad ji, in which the roles are reversed.

115 7. Border change, spillover and new territorial disputes if not logically impossible, and hence such dyads can safely be ignored.6 I therefore follow the second approach and restrict the analysis to relevant dyads only. These can be defined as those dyads in which the outcome of interest can plausibly occur. As I am interested in conflicts between states regarding the location of their boundaries, relevant dyads can here be equated to contiguous states. Drawing on the Correlates of War Direct Contiguity dataset (Stinnett, Tir, Diehl, Schafer, & Gochman, 2002), I define contiguous states as those states that share a land boundary or that are separated by 24 miles of water or less, which I argue best represents the true set of potential cases of territorial conflict between neighbors. The resulting dataset contains 52’210 directed dyad-year observations between 1816 and 2001.

7.4.2 Capturing dyadic interdependence

Dyadic designs are very common in international relations research, where they have been used to study outcomes such as bilateral trade, alliance formation and conflict. At the same time, the widespread use of dyads has also attracted sustained criticism over the years (Signorino, 1999; Hoff & Ward, 2004; Poast, 2010; Cranmer & Desmarais, 2016; Minhas et al., 2016; Poast, 2016). Most im- portantly, most statistical models assume independence among observations, which is already problematic for time-series data that contain repeated obser- vations. This is even more of an issue for dyadic data, which include multiple observations for the same states as part of different dyads. As Beck, Katz, and Tucker(1998) note, analyzing such data without sufficiently accounting for dependence between observations can result in biased estimates and overly optimistic inferences. Although much progress has been made when it comes to accounting for temporal dependence within units (Beck et al., 1998; Carter & Signorino, 2010), cross-sectional dependence between different units still often re- mains unaccounted for, although it poses very similar inferential threats (Ward & Gleditsch, 2007; Cook, 2014). More generally, ignoring dyadic dependencies equals assuming that states act and interact in isolation from each other, which fundamentally contradicts our basic understanding of international politics. For example, most observers would assume that interactions between France and Britain are in some way related to interactions between France and a third state, such as Germany. Similarly, we may expect that relations between France and Britain are to some degree shaped by relations between another pair of

6A study by Lemke and Reed(2001) discusses and evaluates concerns that restricting the sample to relevant dyads alone may induce measurement error and selection bias. The authors generally find that these concerns are negligible, concluding that relevant dyad usage offers a useful alternative to analyzing vast dyadic datasets.

116 7. Border change, spillover and new territorial disputes states, such as Germany and Russia. Depending on the goal of our analysis, we may view such dependencies as a nuisance that needs to be controlled for, or they may be seen as the subject of interest. In either case, empirical models should account for spatial dependence between units as far as possible. As most existing dyadic research has failed to do so, some researchers have come to reject dyadic designs entirely, opting instead for more elaborate network-based approaches (Cranmer & Desmarais, 2016). However, it is worth noting that although problems of cross-sectional de- pendence may be especially pronounced in dyadic data, they also exist in other contexts, such as country- or group-level analyses. Here too, ignoring spatial dependencies can bias findings, while investigating them may lead to impor- tant new insights. In principle therefore, the literature’s recommendations on addressing dependence among individual units can also be applied to spatial dependence in a dyadic context (Neumayer & Plümper, 2010b). Most impor- tantly however, there are still situations in which a dyadic design is clearly preferable to its alternatives. If the theory to be tested is dyadic, then a dyadic analysis seems the most natural choice. This is clearly the case for territorial conflict, which is typically treated as an interaction between two states. For these reasons, the present analysis follows the example of previous studies that continue to use dyads, but aim to account for dyadic interdependencies as much as possible (Elkins, Guzman, & Simmons, 2006; Neumayer & Plümper, 2010b; Poast, 2016). This requires me to carefully consider how dyads may influence each other in the context of this study. How do border changes in one dyad influence the risk of territorial conflict in other dyads? Which patterns of dyadic interdepen- dence should be viewed as a nuisance, and which can be seen as substance? For the purpose of this study, I adopt a relatively simple approach, which distinguishes between two types of dyadic dependence, namely dependence among neighboring dyads and dependence among common member dyads. Both dependencies are based on relationships between the source state in a dyad and other dyads. The source state is the challenger in a potential dispute and therefore is the main actor of interest. The main focus of this analysis is on linkages between neighboring dyads. According to this definition, each dyad ij is related to other dyads km or jm, which are part of source state i’s neighborhood but do not include i as a member. In this scenario, the expectation is that the behavior of i towards its neighbor j is shaped by previous interactions between states k and m. This corresponds to my main Hypotheses 7.1 and 7.2, which state that states are more likely to initiate a dispute with their neighbors following territorial changes between

117 7. Border change, spillover and new territorial disputes

k k

i j i j

m m

Figure 7.1: Two types of dyadic dependence. Arrows indicate directed dyads between states (i, j, k, m). (1) Neighboring dyads: Dyad ij is influenced by other dyads in which i is not a member. (2) Common member dyads: Dyad ij is influenced by other dyads that include i as a member. My argument focuses on dependence between neighboring dyads.

Figure 7.2: Mapping dyadic dependence. Example: Nigeria

118 7. Border change, spillover and new territorial disputes other, neighboring states. A type of second linkage exists between dyads ij and other dyads that include source state i as a common member. In this case, we would expect that state i’s behavior toward j is shaped by its relationships with other states k 6= j. One reason to expect this is that states are likely to adopt relatively consistent foreign policies toward different states to begin with (Ward & Gleditsch, 2007). Another reason is that states may learn from recent interactions with some states, which may influence their behavior toward others. For example, a state that recently gained territory from one of its neighbors may be emboldened to challenge other neighbors with revisionist demands. Although this linkage is not directly relevant to my argument, I aim to account for it in the analysis, in order to capture dyadic dependencies as far as possible. My overall argument centers on spillover effects within neighborhoods. The assumption is that states are part of distinct regional systems, in which neighboring states operate under similar circumstances, and may therefore respond similarly to changes that occur within the region. For example, it seems more reasonable to expect that a border change between India and Pakistan would influence the behavior of China towards its neighbors than it would have any influence on Brazil’s foreign policies. Although it is certainly possible that some territorial changes have far-reaching consequences that extend beyond neighborhoods, I focus more specifically on neighborhood-specific contagion effects. To define common member dyads, I therefore link each dyad ij to other dyads that include source state i and other direct neighbors k 6= j. For neighboring dyads, I link each dyad ij to other dyads between i’s neighbors (k, m, n, ...) and their own direct neighbors (p, q, r, ...) Figure 7.2 illustrates this approach in the example of Nigeria.

7.4.3 Generating spatial lags

Having defined the overall structure of the neighborhoods in which spillover effects may operate, the next step is to generate spatial lags that capture these effects. Spatial lags provide aggregate measures of a variable among a unit’s neighbors, which are generated using a spatial weights matrix W. For N units, W is defined as a N × N matrix, which indicates the presence or absence of linkages between all units. Elements that represent connected units are assigned some non-zero value, while elements representing unconnected units are set to zero. The spatial weights matrix is then typically used to pre-calculate a weighted average or sum of a variable, or is directly incorporated into a model that includes spatial lags on the outcome variable, explanatory variable, or error

119 7. Border change, spillover and new territorial disputes term (Anselin, 2001). The case becomes more complicated when considering spatial effects with dyadic data. Here, the spatial weights matrix is defined by (N × N)2 for N states, which can also be described as a 4-adic dataset (Neumayer & Plümper, 2010b; Poast, 2010). For time-series data, the matrix extends into a N × N × T block-diagonal matrix across time periods T. For most applications, the spatial weights matrix alone would quickly become too large to handle.7 As a solution to this problem, I adopt an approach developed by Neumayer and Plümper (2010a), which consists of parsing through a "virtual" connectivity matrix, and selecting only those dyads that are linked to a given dyad ij according to one’s connectivity criteria. For each dyad and every year, the subset of connected dyads is then used to calculate the spatial effect variables. The next question is which variables require spatial lags. The literature generally distinguishes between three types of spatial effects, each of which requires its own modeling strategy (Neumayer & Plümper, 2010a). A first type of effect arises from dependence in outcome variables. For example, a prominent explanation for democratization is that democracy itself tends to diffuse within neighborhoods (K. S. Gleditsch & Ward, 2006). Such effects can be estimated with a model that incorporates a spatial lag on the outcome variable (Ward & Gleditsch, 2007). A second type of effect arises from dependence on explanatory variables. In this scenario, the value of a "treatment" Xi for unit i may also influence an outcome Yj for unit j. For example, studies have shown that environmental regulations in country i may also improve the health of citizens in neighboring country j, which has not implemented the same standards (Brueckner, 2003). Estimating such effects requires models that include a spatial lag on X, as discussed at length by Vega and Elhorst(2015). Third and lastly, spatial effects may be driven by dependence on the error term, which can result from failure to account for the two other types of dependence, or from omitted variables that cluster in space (Neumayer & Plümper, 2010a). For example, a possible alternative explanation to conflict diffusion is that some country characteristics that make countries more conflict-prone cluster in regions (Buhaug & Gleditsch, 2008). In short, spatial dependence can arise from dependence on the outcome, the explanatory variable, or both (Poast, 2016). In addition, it may also simply be an artefact of spatially clustered omitted variables, or the common exposure of neighboring units to external shocks. The main goal of this chapter is to

7For example, if we specify a inter-dyadic spatial weights matrix for all states in the international system from 1816 to 2001, the number of elements reaches ≈ 2.8e10. If we only consider contiguous dyads during the same period, the number still reaches ≈ 2.3e7

120 7. Border change, spillover and new territorial disputes examine spatial dependence on the explanatory variable, as I am interested in how border changes affect the risk of disputes elsewhere. However, in order to test for this type of dependence, it is important to account for alternative channels of spatial dependence as well, as these may otherwise drive the results (Vega & Elhorst, 2015). To this end, my models include spatial lags on both the explanatory and outcome variable and furthermore account for relevant dyad-level characteristics that may cluster regionally. Finally, one remaining question is how to measure and weigh spatial effects. Studies typically assign weights to spatial linkages based on the idea that nearby units are more influential than remote ones. The strength of each connectivity can be defined by geographic proximity, or based on non-geographic criteria, such as institutional similarity or bilateral trade (Ward & Gleditsch, 2007; Beck, Gleditsch, & Beardsley, 2006). When using a weighted connectivity matrix, the estimated spatial effects account for the varying degrees of connectedness and influence among units (Zhukov & Stewart, 2013). However, my analysis is already restricted to groups of contiguous states, and I have no strong reasons to assume that some dyadic linkages are more important than others in explain- ing territorial conflict. Therefore, I assign equal weights to each inter-dyadic linkage. Another question is how to aggregate the values of neighboring units into a single variable. Most studies calculate spatial lags as a weighted sum or average of neighboring observations, but others have also used a simple dummy variable for the presence or absence of an outcome among neighbors (e.g., Buhaug & Gleditsch, 2008). In the main analyses, I adopt the second ap- proach, which corresponds to a "weakest-link" specification: If any of a dyad’s neighboring dyads experience a territorial change, this is expected to increase the risk of dispute initiation.

7.4.4 Data

Having outlined my overall approach to capturing spillover effects, I now turn to the main data sources used in the analysis. The dependent variable measures the initiation of a new territorial dispute, according to the ICOW dataset (Hensel & Mitchell, 2017). In line with my definition, ICOW defines territorial disputes as an "explicit contention between official government representatives regarding sovereignty over a specific piece of territory." (Hensel & Mitchell, 2017, p. 127). The dataset further distinguishes between disputes over territories claimed as part of a state’s core territory (homeland), and disputes over colonial territories abroad (dependent territory). In this chapter, I focus entirely on the former category and thus exclude disputes over colonial dependencies, which are

121 7. Border change, spillover and new territorial disputes conceptually distinct and not necessarily relevant to my argument. Figure 7.2 shows the distribution of interstate territorial disputes over time, which reveals some interesting patterns. First, the number of active disputes has overall increased in recent history, which is explained in part by the onset of new disputes, and in part by the persistence of old ones (Huth, 1996; Schultz, 2015a). Second, disputes tend to cluster in time, erupting more frequently in certain periods than in others. For the explanatory variable, I rely on the Territorial Change dataset by Tir et al.(1998), which records all exchanges of territory involving at least one inde- pendent state since 1816. Besides listing the territories and the parties involved in each change, the dataset also records information on the process of territorial change, distinguishing between categories such as conquest, annexation, ces- sion, secession and international mandates. Furthermore, the dataset records whether or not a change was the result of violent conflict, and whether it coin- cided with the creation or dissolution of an independent state. Based in part on the information provided by Tir et al.(1998) and based on additional research, I code a dummy variable to distinguish cooperative from conflictual border changes.8 The former category includes cases in which territory changes hands as a result of bilateral agreements, purchases, mutually recognized plebiscites or international arbitration, as shown in Table C8 in the appendix to this chapter. Figure 7.4 shows the distribution of both types of territorial change since 1816. Overall, there has been a substantial decrease in conflictual exchanges of terri- tory throughout the 20th century, and in particular since 1945. This decrease is remarkable considering that the overall number of states, and therefore the potential for disputes between them, has increased over the same period. By comparison, cooperative territorial transfers have become more common in the post-1945 period. Most of these changes however are related to decolonization and the dissolution of the Soviet Union, which mostly fall into the cooperative category. One implication of my argument is that territorial disputes are likely to cluster in space and time. If border changes indeed give rise to new territorial disputes nearby, we would expect disputes to erupt more frequently in certain regions and at certain times than others. Absent any evidence of such cluster- ing, there would be no good reason to assume that spatial processes such as contagion drive territorial disputes. As a first step therefore, I examine regional trends in more detail. Figure 7.5 plots disputes over homeland territory by world regions over time. I broadly distinguish between 5 world regions, and assign disputes to regions based on the location of the challenger state. Disputes

8The coding procedure is described in more detail in the appendix to this chapter.

122 7. Border change, spillover and new territorial disputes

Figure 7.3: ICOW territorial disputes 1816-2001: Incidence and Onsets

Figure 7.4: Conflictual and cooperative territorial changes over time. Dashed lines indicate smoothed trend estimates.

123 7. Border change, spillover and new territorial disputes are shown as a percentage of all dyads within each region. Based on the data, it becomes clear that there are substantial differences across world regions, both in terms of the overall frequency of disputes and in region-specific time trends.9 In the Americas, the number of disputes continually increased until 1900, and then experienced a sustained drop, followed by another increase in the 1970s. In Europe, territorial disputes sky-rocketed during and after World War I, and decreased just as sharply thereafter. The 1990s saw another increase in disputes, mostly among the successor states of the former Soviet Union and Yugoslavia. In Africa, disputes increased sharply in the 1960s, when most African states became independent. Upon independence, many states inherited unresolved border disputes from their former colonial rulers. In Asia, disputes remained relatively rare throughout the 19th and early 20th century, but they experienced a steep increase following World War II, which continued through the 1970s. Most of this increase is driven by numerous disputes between the Soviet Union and its neighbors across the continent, as well as many disputes involving China, India and Pakistan. Finally, disputes in Australia and Oceania remained absent until 1975, after which two disputes emerged over small pacific islands.

7.4.5 Model specification

Having introduced the main data sources, I now turn to the analyses. To test my hypotheses, I estimate binary time-series cross-sectional models that take the following general form:

Yij = ρ1Wterr.chgkm + ρ2Wdisputekm + β1Xij + β2Z + eij

Yij indicates territorial dispute initiation by state i in dyad ij. W denotes the various spatial lags on the explanatory and outcome variables for dyads

Figure 7.5: Territorial disputes by region

9The number of disputes shown here is proportional to the number of neighboring dyads within each region.

124 7. Border change, spillover and new territorial disputes km 6= ij, with their effects indicated by ρ. X is a vector of dyad-level control variables. Z is a vector of system-level controls. Standard errors are clustered on the dyad level. The main explanatory variable is a spatial lag that captures territorial changes among neighboring dyads, and is derived from the Territorial Change Dataset (Tir et al., 1998). I only record exchanges involving each state’s core territory and exclude exchanges that involved colonial dependencies.10 All observations for territorial changes are lagged by one year. However, because territorial changes are likely to have more lasting effects, I extend this lag to include each of the five preceding years. To account for the second type of spatial dependence, I generate a second spatial lag that records territorial changes among common-member dyads, using the same procedure as for the first spatial lag. As noted previously, spatial dependence frequently arises from dependence in outcomes, rather than explanatory variables. In order to control for alterna- tive channels of dependence, I use two additional spatial lags on the dependent variable, which record active disputes among neighboring dyads and disputes among common-member dyads in the previous year and its 5 preceding years. In addition to these spatial lags, I also include a set of dyad-specific control variables that plausibly affect the risk of disputes within dyads. These include a time-lag for territorial changes within each dyad in the previous 5 years, as well as a time-lag for active territorial disputes. I also include a dummy variable for joint democracy, based on the Polity IV dataset (Marshall, Gurr, & Jaggers, 2014)11 and a variable indicating the difference in military capabilities between a dyad’s members, using the composite index of national capabilities (CINC) from the Correlates of War database (Singer, 1987). Further dyad-level controls record the number of past militarized interstate disputes, a count variable for a dyad’s total number of contiguous neighboring states, as well as a logged measure of dyad age in years. To account for temporal dependence among disputes, I also include three variables for the cubic polynomial of time since the last dispute (Carter & Signorino, 2010). The full model furthermore includes a set of system-level controls intended to absorb the effects of global trends and temporal shocks. As a large number of territorial disputes emerged during and after the two World Wars, I add a dummy variable for all years that coincided with these wars. To account for broader global trends in border instability, I further include a variable for the global number of border changes in the previous year. Finally, I include

10This is done using the dataset’s "gaintype" and "losetype" variables. 11Dyads are coded as democratic if their lowest Polity IV democracy score is equal to or greater than 6.

125 7. Border change, spillover and new territorial disputes a measure of systemic instability, which records shifts in the distribution of power among the most powerful states (Gunitsky, 2014). In a recent study, Abramson and Carter(2018) use this measure to explain the onset of territorial disputes. The variable records yearly changes in military capabilities among states coded major powers, using the CINC measure. The summary statistics of all variables are shown in Table C1 in the appendix. Finally, to account for temporal dependence within dyads, I include the cubic polynomials of time since the last dispute (Carter & Signorino, 2010).

7.5 Results

Table 7.1 shows the results of a first set of models, which test for potential spillover effects of any type of border change. Model 1 only includes the main explanatory variables, while Models 2 and 3 introduce dyadic and system-level control variables respectively. In addition, Model 4 includes region fixed effects to account for further unobserved heterogeneity between regions. In line with Hypothesis 7.1, I find that recent territorial changes in a dyad’s neighborhood increase the likelihood of territorial dispute onset within that dyad. The co- efficients are significant and consistent across all models. The coefficients for common member dyads point to a similar effect for recent territorial changes that involved the challenger state and other neighboring states. Perhaps not sur- prisingly, the results also indicate that recent territorial changes within a dyad also increase the risk of territorial disputes within that same dyad. While the incidence of territorial disputes among neighboring dyads appears to have no discernible effect on dispute onset, I find a positive effect for disputes involving common member dyads. This suggests that the source state in a directed dyad is more likely to initiate a territorial dispute if it is already engaged in disputes with other neighbors. The results for the dyadic and system-level control variables are generally in line with previous findings in the literature. Joint democracy decreases the likelihood of a dispute, while a larger number of previous militarized conflicts is associated with an increased risk of dispute onset. As expected, the two world wars explain a large portion of the overall variation in disputes. Finally, the number of global territorial changes in the previous year also has a positive and significant effect on the risk of territorial dispute initiation. This is an interesting finding by itself, and suggests that spillover effects may not be limited to the direct neighborhood of states only. A second set of models replicates the previous analyses, but distinguishes between conflictual and cooperative instances of border change. The results

126 7. Border change, spillover and new territorial disputes are shown in Table 7.2. As stated by Hypothesis 7.2, I find that conflictual border changes within a dyad’s neighborhood are associated with a greater risk of territorial dispute onset for that dyad. The findings are again estimated consistently across all models and are significant at the 0.05 level. In contrast, the estimates for cooperative border changes within the neighborhood are also positive, but far from significant. Models 1 and 2 suggest that conflictual border changes within common member dyads have a positive effect on dispute onset, but this effect is no longer significant after inclusion of system-level controls and region-fixed effects. As expected, conflictual border changes within a given dyad increases the risk of dispute initiation within that same dyad. Perhaps more surprisingly, the results also suggest that cooperative border changes within a dyad increase the risk of a dispute in that same dyad. The remaining estimates for the control variables are all in line with the findings of the previous analysis. To interpret the substantive importance of the results, I use the estimates of Models 1 through 3 to simulate differences in the predicted probability of a dispute onset when the explanatory variable "switches" from 0 to 1, following the simulation-based approach outlined by King et al.(2000). 12 I compare two hypothetical cases in which a dyad experiences a border change in its neighborhood and in which it does not, and set all other variables to their sample mean, or mode. Figure 7.6a shows the results for all border changes in the first panel, and for conflictual and cooperative changes in the second and third panels. Following neighboring border changes the risk of dispute initiation increases by approximately 0.2 percent in any given year, plus or minus approximately 0.1 percent. This effect is approximately the same for territorial changes overall, and conflictual changes in particular. By contrast, cooperative border changes have a smaller positive effect, but this effect is not clearly distinguishable from zero.13

7.5.1 Comparing pre- and post-1945 spillover effects

As noted earlier, the international system has undergone substantial changes in recent history. Following Hathaway and Shapiro(2017), we can broadly distinguish between an old world order that existed until World War I, and a new order that became firmly established after 1945. Although the old order

12I simulate point estimates and confidence intervals based on 1000 random draws from a multivariate normal distribution, based on the estimates and clustered standard errors from Model 3 in each table. All other variables are set to their sample mean or mode. 13The overall effect sizes seem very small, but are comparable to the effects of other important variables in the model, such as joint democracy.

127 7. Border change, spillover and new territorial disputes

Table 7.1: Logit models: Territorial dispute onset, 1816-2001

Model 1 Model 2 Model 3 Model 4 W: TC (neighborhood) 0.58∗∗∗ 0.60∗∗∗ 0.53∗∗∗ 0.43∗∗∗ (0.15)(0.16)(0.16)(0.16) W: TC (common member dyads) 0.58∗∗∗ 0.52∗∗∗ 0.47∗∗∗ 0.46∗∗∗ (0.17)(0.17)(0.18)(0.18) W: Disputes (neighborhood) −0.22 −0.26 −0.29 −0.37 (0.33)(0.35)(0.35)(0.34) W: Disputes (common member dyads) 0.11 0.10 0.08 0.08 (0.15)(0.15)(0.14)(0.14) TC (dyad) 1.44∗∗∗ 1.40∗∗∗ 1.29∗∗∗ 1.32∗∗∗ (0.24)(0.25)(0.26)(0.25) Ongoing dispute (dyad) −0.31 −0.28 −0.34 −0.36 (0.26)(0.31)(0.29)(0.29) Joint democracy −1.03∗∗∗ −0.94∗∗ −0.85∗∗ (0.37)(0.37)(0.37) Rel. capabilities −0.01 0.09 −0.35 (1.10)(1.04)(1.00) Past MIDs 0.10∗∗∗ 0.08∗∗ 0.06∗∗ (0.03)(0.03)(0.03) Dyad age (log) −0.16 −0.17 −0.09 (0.12)(0.11)(0.11) Neighbors (count) 0.00 0.02 −0.00 (0.02)(0.02)(0.02) World wars 1.45∗∗∗ 1.54∗∗∗ (0.24)(0.25) TC (world) 0.06∗∗∗ 0.06∗∗∗ (0.01)(0.01) Systemic instability 13.16 24.66∗ (14.34)(14.28) Intercept −5.23∗∗∗ −4.88∗∗∗ −5.34∗∗∗ −6.03∗∗∗ (0.40)(0.44)(0.44)(0.47) t, t2, t3 yes yes yes yes Region FE no no no yes Log Likelihood −1217.51 −1172.39 −1145.19 −1134.17 Observations 44681 43834 43834 43834 Notes: Unit of analysis: directed dyad-years (contiguous states only). Dyads consist of a potential challenger and a target state in a territorial dispute. W denotes spatial lags on variables among common-member and neighboring dyads. TC is short for territorial changes. Main explanatory variables in bold. Robust standard errors clustered on the (undirected) dyad level. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

128 7. Border change, spillover and new territorial disputes

Table 7.2: Logit models: Territorial dispute onset, 1816-2001

Model 1 Model 2 Model 3 Model 4 W: TC conflict (neighborhood) 0.55∗∗∗ 0.54∗∗∗ 0.41∗∗ 0.31∗∗ (0.16)(0.17)(0.16)(0.16) W: TC coop. (neighborhood) 0.24 0.24 0.27 0.24 (0.18)(0.18)(0.18)(0.18) W: TC conflict (common member dyads) 0.59∗∗∗ 0.51∗∗ 0.43∗∗ 0.40∗ (0.21)(0.21)(0.21)(0.21) W: TC coop. (common member dyads) −0.08 −0.04 0.04 0.11 (0.26)(0.27)(0.25)(0.24) W: Disputes (neighborhood) −0.18 −0.20 −0.23 −0.31 (0.33)(0.35)(0.35)(0.33) W: Disputes (common member dyads) 0.16 0.14 0.10 0.09 (0.15)(0.15)(0.14)(0.14) TC Conflict (dyad) 1.34∗∗∗ 1.27∗∗∗ 1.16∗∗∗ 1.19∗∗∗ (0.27)(0.28)(0.29)(0.28) TC Coop. (dyad) 0.94∗∗∗ 0.98∗∗∗ 0.99∗∗∗ 1.07∗∗∗ (0.34)(0.34)(0.34)(0.33) Ongoing dispute (dyad) −0.26 −0.27 −0.35 −0.38 (0.26)(0.31)(0.30)(0.29) Joint democracy −1.02∗∗∗ −0.94∗∗ −0.85∗∗ (0.37)(0.37)(0.37) Rel. capabilities 0.18 0.27 −0.23 (1.10)(1.05)(1.02) Past MIDs 0.09∗∗∗ 0.08∗∗ 0.06∗∗ (0.03)(0.03)(0.03) Dyad age (log) −0.14 −0.15 −0.08 (0.12)(0.11)(0.11) Neighbors (count) 0.00 0.02 −0.00 (0.02)(0.02)(0.02) World wars 1.48∗∗∗ 1.57∗∗∗ (0.24)(0.25) TC (world) 0.06∗∗∗ 0.06∗∗∗ (0.01)(0.01) Systemic instability 8.72 21.30 (14.65)(14.53) Intercept −5.20∗∗∗ −4.88∗∗∗ −5.32∗∗∗ −5.97∗∗∗ (0.40)(0.44)(0.44)(0.47) t, t2, t3 yes yes yes yes Region FE no no no yes Log Likelihood −1217.61 −1173.90 −1147.28 −1137.02 Observations 44681 43834 43834 43834 Notes: Unit of analysis: directed dyad-years (contiguous states only). Dyads consist of a potential challenger and a target state in a territorial dispute. W denotes spatial lags on variables among common-member and neighboring dyads. TC is short for territorial changes. Main explanatory variables in bold. Robust standard errors clustered on the (undirected) dyad level. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

129 7. Border change, spillover and new territorial disputes did have its own status quo-oriented rules and norms, it was overall much more permissive towards territorial change, as conquest and annexation were generally seen as legitimate. In contrast, the post-war order was defined by much stricter rules and norms regarding territorial change, especially regarding border changes by force (Zacher, 2001; Hathaway & Shapiro, 2017). It is likely that these two different sets of rules and norms have had their own implications for the risk of spillover. To examine this, I divide the dataset into two periods, the first one extending from 1816 to 1945 and the second from 1946 to 2001, and repeat the main analyses within these sub-samples. The full results are shown in the appendix to this chapter, specifically in Tables C2, C4, C3 and C5. In addition, Figures 7.6b and 7.6c summarize the main results, showing differences in predicted probabilities for the main explanatory variables. The results show a clear increase in the risk of territorial disputes following regional border changes in the pre-1946 period, which is approximately twice as large as the effects that were estimated for the entire period. In line with the previous analysis, I find that conflictual border changes have a positive and significant effect, but find no significant effect for cooperative exchanges of territory. In contrast, the results for the post-1945 period are much less consistent. Here, I find no clear evidence that regional border changes have had any impact on the risk of dispute onset. This suggests that the results that were estimated across the whole period are mainly driven by observations between 1816 and 1946. Strikingly, the estimated effects and the results from Table C4 suggest that in the postwar period, cooperative exchanges of territory are associated with a greater risk of territorial disputes across the neighborhood. Upon closer exami- nation, it appears that this effect is driven mainly by cases of decolonization after 1946, most of which fall under the "cooperative" category outlined above. Many colonies "inherited" unresolved border disputes from their former colo- nial rulers, which resurfaced as new disputes in the dataset. Finally, although I do not find evidence for spillover effects within neighborhoods following World War II, the number of border changes worldwide still appears to increase the risk of territorial disputes, as shown in Tables C3 and C5. This suggests that spillover effects may still operate today on a larger international scale, which should be investigated further. To assess the robustness of my findings, I conduct a set of additional analyses. The main concern are potential omitted variables that may drive the results. In particular, the findings may be explained by omitted variables on the dyad-level, or by global trends or temporal shocks that may drive both a large number of border changes and disputes. To account for this, I repeat the analyses in a

130 7. Border change, spillover and new territorial disputes

(a) 1816-2001

(b) 1816-1945

(c) 1946-2001

Figure 7.6: Differences in the probability of territorial dispute onset between no changes and one territorial change among neighboring dyads, estimated across different periods. Lines indicate 95% confidence intervals

131 7. Border change, spillover and new territorial disputes set of Linear Probability Models (LPM) that include a large number of group fixed effects. A first set of models include dyad fixed effects, while a second set of models include year fixed effects. In both cases, the main findings remain robust, as shown in Tables C6 and C7.

7.6 Conclusion

In this chapter, I have presented the first systematic test of a common claim in the literature, which holds that individual border changes may destabilize borders on a larger regional scale. Using information on territorial changes and disputes between neighboring states from 1816 to 2001, I find some support for this general argument. More precisely, I find that states are more likely to initiate a dispute with their neighbors following territorial changes between other states in their neighborhood. I furthermore find that this relationship is driven mainly by conflictual instances of border change that stem from conquest, annexation and unilateral secession, as opposed to cooperative exchanges of territory following bilateral agreements, purchases, arbitration and the like. While my overall findings remain robust across a set of alternative specifications that include region, dyad and year fixed effects, I also find that regional spillover effects are mostly limited to the pre-World War II period. By contrast, the evidence for spillover is less conclusive in the post-1945 period, where much stricter rules and norms regarding territorial change were put in place. It should be noted that the present analyses have clear limitations that remain to be addressed in further research. First, to further assess the robustness of my findings, it will be necessary to conduct additional tests. Most importantly, these tests should replicate the main findings with an alternative set of spatial lags and using alternative time windows to measure past border changes. Aside from spillover between direct neighbors, broader regional and even global spillover effects need to be investigated further. Until that point, my findings should be seen as preliminary, and are the result of a first attempt to unpack spillover effects among dyads. A second limitation in the current analysis is that I have implicitly assumed that all states are equally susceptible to spillover effects. This is clearly not the case. Not all states are revisionists, and in fact only relatively few may actively seek to redraw the map. We would therefore expect that only states with existing motives to initiate a dispute are likely to respond to precedents set by border changes elsewhere. Such motives could relate to recent territorial losses or the presence of ethnic kin across the border. Third and lastly, while this analysis has focused entirely on the onset of territorial disputes, it may be worthwhile to examine dispute escalation as

132 7. Border change, spillover and new territorial disputes another outcome of interest. Many disputes remain active for long periods of time, and tend to erupt in armed confrontations infrequently, or not at all (Schultz, 2015a). In these contexts, potential spillover effects of border instability may provide another important explanation.

133 7. Border change, spillover and new territorial disputes

134 Chapter 8

Conclusion

Contrary to predictions of a "borderless" world (Ohmae, 1991), borders will remain defining feature of international politics for the foreseeable future. This dissertation shows that the way in which borders were drawn matters for their stability over time and provides evidence that past border instability increases the risk of renewed disputes and border change. Moreover, I find preliminary evidence that border instability may be contagious, as border changes between one pair of states may increase the risk of new territorial disputes between other, proximate states. This section briefly summarizes the main contributions of this dissertation, highlights its limitations and discusses avenues for future research. It concludes with a brief discussion of policy implications.

8.1 Summary and main contributions

Chapter5 contributes to the study of a key question that has remained largely neglected in political science: the formation and change of international bor- ders. Following previous research, I conceptualize border drawing as part of a bargaining process between states, in which competing parties rely on focal principles to find agreement (Schelling, 1980; Carter & Goemans, 2011; Huth et al., 2013). I argue that some principles are more likely than others to generate borders that are well-defined and recognizable on the ground, which helps reduce uncertainty and increases the potential for joint benefits, which in turn should make borders more durable. In line with this reasoning, I show that straight borders are less likely to endure, while borders that follow clear-cut natural features are more likely to remain in place. I also find some evidence that borders that follow historical precedents are more stable, although this finding is less reliable. Interestingly, not all of these effects remain constant over time. The negative effects of border straightness decrease and eventually

135 8. Conclusion disappear with time, suggesting that states and people learn to coordinate even on borders that were arguably "flawed" at the time of their establishment. In Chapter6, I present a first global analysis of the consequences of border change, which explores whether and to what extent border instability may persist within regions. I argue that historically unstable regions are likely to experience further instability due to the weak consolidation of current borders, and the existence of border precedents. Previous borders can have a lasting impact on individual behavior and identities, and as such may become likely targets for new territorial claims. A cross-sectional analysis on the level of grid cells yields strong support of my arguments: Regions with unstable borders prior to World War II period have remained more prone to interstate and domestic territorial disputes in the post-1946 period. Furthermore, these regions have also been more likely to experience further instances of border change. These findings are consistent with a previous study by Abramson and Carter (2016), which focuses on Europe. However, they also go beyond previous studies in showing that the relationship between between border instability and territorial disputes holds globally, and does not only apply to the interstate domain but also extends to domestic conflict. Chapter7 adopts a broader regional perspective and considers the potential spill-over effects of border change. I offer the first empirical assessment of a common claim in the literature, according to which individual border changes can destabilize borders on a larger regional scale. In a dyadic analysis of border changes and territorial disputes since 1816, I find partial support for this claim, as it shows that territorial changes between one pair of states increase the risk of new disputes between other, neighboring states. This effect is largely driven by conflictual instances of territorial change as opposed to cooperative exchanges of territory. However, these spillover effects also mostly seem to be limited to the pre-World War II period. In addition to these empirical contributions, two main contributions of this dissertation are the creation of two new datasets. As part of this project, I have supervised the creation of the CShapes 2.0 dataset, in collaboration with Luc Girardin, Seraina Rüegger, Lars-Erik Cederman, Nils Weidmann and Kristian Skrede Gleditsch and with the help of a team of research assistants. The new dataset is the first to map borders since the late 19th century with global coverage and will soon be made publicly available. It allows researchers to compute a range of spatial variables using accurate historical boundaries and will enable new spatially explicit research on a number of issues that were so far difficult to study, such as border change, state formation or the spatial expansion of colonial rule.

136 8. Conclusion

A second data contribution of this dissertation is GeoSDM, which maps the territories claimed by self-determination movements since 1946, and is based on the SDM dataset by Sambanis et al.(2018). Data collection is currently still ongoing but will soon be completed, after which the dataset will be made publicly available. In this dissertation, I have presented a first application of the new data to examine the historical origins of separatist claims. However, the potential usefulness of the dataset extends far beyond this particular application. For example, the new data can be used to study the emergence of separatist movements and examining how geography shapes their political demands, as well as the government’s response to these demands.

8.2 Limitations

Despite its contributions, this dissertation has a number of limitations that are briefly discussed here. A first limitation relates to the theoretical framework outlined in Chapter3. To explain how borders are drawn, I have sketched out a simple bargaining model in which states compete over the distribution of territory but aim to coordinate on a given outcome. As I have noted, this model cannot account for all instances of border drawing. In some cases, states may try to overrun their opponent and seize their entire territory, while in others, borders are the direct outcome of battle rather than a coordinated attempt to define territorial lines. Although I have argued that such cases have been the exception in recent history, they still demonstrate the limitations of my argument. Even more importantly, my argument does not account for the problem of indivisibility, which can drastically limit the opportunities for coordination in territorial bargaining. Indivisibility remains one of the key reasons why states are often unable to settle on any given boundary. It should be noted that by focusing on the stability of borders after they have been established, the analyses in Chapter5 are able to partially sidestep the issue of indivisibility. Still, to develop a better understanding of where and how borders are drawn, it is important to account for indivisibility and other bargaining problems that have been discussed elsewhere in the literature. A second limitation is that I have relied entirely on a "black-box" approach to studying causal relationships, at the expense of testing possible causal mech- anisms. For example, my theoretical framework has outlined a number of plausible causal mechanisms that link historical border instability to conflict but the validity of these mechanisms has not been tested empirically. In particular, it would be useful to know whether the lasting effects of borders are mainly due to their impact on group identities, institutions, economic exchange, or all of

137 8. Conclusion the above. To examine this, we would require a more fine-grained quantitative analysis or could alternatively rely on qualitative evidence.

A third issue is endogeneity, which remains a concern throughout all of my empirical analyses. For example, I have noted that the institutional "design" of borders and their durability over time could both be the result unobserved patterns in state behavior that give rise to certain types of borders, while also affecting the risk of future border changes. Endogeneity poses a threat espe- cially when it comes to unpacking the relationship between border change and conflict, which are often part of one and the same process. Despite my efforts to address these inferential concerns, I cannot entirely rule out the possibility that some of my findings are driven by omitted variables.

A fourth limitation relates to the units of analysis used in the empirical chapters. The focus on border segments in Chapter5 enables a fine-grained study of where border changes occur, but this approach can only offer a partial understanding of why borders change. Borders are part of the state as a whole, and a more comprehensive analysis may require us to pay closer attention to state properties and interactions between states. For example, the distribution of power between states or differences in regime types may play an important role in the process of border drawing, which should be examined further. Chapter6 focuses on grid cells as the unit of analysis, which allows me to address certain methodological issues related to unit change. However, this approach creates its own difficulties, as grid cells cannot meaningfully represent the relevant actors, thereby raising questions about agency. To overcome this, additional analyses on the level of states, dyads and ethnic groups would be particularly useful. Lastly, Chapter7 relies on dyads to study the spillover effects of border change, but I have noted that dyads create their own methodological difficulties that stem from mis-specifying and breaking up complex dependence structures in international politics. As a possible alternative, network-based methods could potentially offer a more comprehensive approach to studying interdependencies between states.

Fifth and finally, the analyses in Chapters6 and7 have focused entirely on territorial disputes as the outcome of interest. Territorial disputes are the first step in a process that often, but not always leads to violent forms of conflict, and therefore serve as a natural starting point to study the consequences of border change. However, one key question that has remained unanswered relates to the escalation of disputes: Under what conditions does historical border insta- bility result in violent conflict? To help answer this question, a configurational analysis using geocoded conflict data may be particularly useful.

138 8. Conclusion

8.3 Directions for future research

The institutional literature on borders is still relatively young, and hence there remain many opportunities for further research. Some tasks for future research flow directly from the limitations discussed in the previous section. For example, a more detailed analysis of causal mechanisms and the process of escalation are key to better understand the implications of border change. In addition, this section outlines three promising tasks for future research that build on the preceding chapters. Drawing on the theory and analysis presented in Chapter5, future research could further examine the formation of international borders. Where do borders come from? How many international borders follow clear focal principles and how many do not? What determines the size and shape of states? The general logic of focal principles could offer an improved understanding of how and where borders are drawn, which would make a significant contribution to the broader literature on state formation and state size (e.g., Friedman, 1977; Tilly, 1985; Alesina & Spolaore, 2003; Lake & O’Mahony, 2004). Furthermore, if it is possible to "predict" the location of borders with relative accuracy, this may also help to address inferential problems that arise from the endogeneity of borders. A second main question left largely unaddressed by this dissertation is the relationship between borders and identities. Research on nationalism has generally stated that national identities play a central role in the drawing and redrawing of borders (Gellner, 1983; Hechter, 2000). At the same time, several studies have argued that borders themselves are essential in defining and shaping group identities (Abbott, 1995; Paasi, 1996; White, 2000). Future research could attempt to further disentangle this relationship, for example by studying the lasting effects of previous borders on ethnic and national identities. Lastly, future research should further examine the spillover effects of border change, as there are good theoretical reasons to believe that border instability can become contagious. Chapter7 has already taken a first step into this direction by examining how territorial changes between one pair of states affect the risk of disputes elsewhere. However, the most important implications of border change arguably relate to the risk of secession. A long-standing debate within the literature on secession concerns the risk of secessionist domino effects, as concessions to separatist groups may inspire other aggrieved minorities to make similar demands (Hale, 2000; Fearon, 2004; Walter, 2009; Forsberg, 2013; Bormann & Savun, 2018). While most empirical studies have focused on demonstration effects within countries, cross-border spillover effects constitute another important subject that warrants further study.

139 8. Conclusion

8.4 Policy implications

Keeping in mind the limitations of this study, the findings and theoretical ar- guments presented in the preceding chapters may shed light on current policy debates in three main ways. First, my findings provide further support for the idea that the way in which new borders are drawn matters for international sta- bility. Previous studies have already shown that the alignment of new borders with administrative lines and prominent natural features helps to reduce the risk of interstate conflict (Carter & Goemans, 2011; Goemans & Schultz, 2017). This dissertation furthers this argument by showing that borders that follow recognizable features have proven more durable over time. Considering that many borders remain disputed and separatist movements have been on the rise, it is likely that further border changes will occur in the near future. In order to minimize the destabilizing effects of such changes, it is important to ensure that new borders are well-defined, recognizable and indisputable. This also underlines the importance of the uti possidetis principle, which states that in the case of secession, new borders should follow well-established administrative divides. Where such a solution is not feasible, natural features such as rivers may help to reduce the disruptive impacts of border change. Second, the finding that border instability can persist raises the question of how states can escape a potential vicious cycle of instability. Based on my arguments, we can distinguish between three main mechanisms by which this could be achieved. One option is simply to let time pass, as suggested by the path dependence argument. Absent any major disruptions, local populations may increasingly adapt to the existing political map, which may eventually result in increasing border stability. As a second option, states can also attempt to consolidate control over potentially unstable regions, making it more difficult for rival states to claim and seize these territories or for local populations to secede. States may attempt to do so by increasing their presence, investing in infrastructure or by attempting to secure the loyalty of the local population. The latter may either be achieved by granting special rights to local minority groups or through educational policies or assimilation strategies. However, state- led assimilation attempts raise their own ethical concerns, as they have often involved substantial repression and violence. Aside from capacity-building and nation-building strategies, the potential economic gains of border stability suggest that regional policies that promote economic interdependence could also help to stabilize borders in historically unstable regions. Third and lastly, the preliminary evidence showing that individual border changes can destabilize borders elsewhere underlines the importance of inter-

140 8. Conclusion national rules, norms and cooperation to preserve existing borders. As recent events such as the annexation of Crimea has shown, a general commitment to preserving the territorial order does not rule out individual rule violations. However, to reduce the risk that such violations generate further instability, it is crucial that states continue to cooperate to enforce existing rules, in order to prevent disruptive actions in the future.

141 8. Conclusion

142 Appendix

143

Appendix: Defining the Outlines (Chapter5)

A.1 Descriptive statistics

Table A1: Summary statistics: Chapter5

Statistic Min Max Mean Median St. Dev. Fractal Dimension 1.022 1.205 1.103 1.094 0.019 Watershed 0 1 0.083 0 0.199 River 0 1 0.178 0 0.340 Lake 0 1 0.019 0 0.119 Terrain ruggedness, log 0.000 7.519 4.568 4.541 1.275 Reestablished 0 1 0.058 0 0.234 Colonial 0 1 0.357 0 0.479 Population density, log 0.000 6.203 1.593 1.410 1.353 Agric. Suit. 0.000 5.881 2.204 1.980 1.498 Pre-1886 0 1 0.421 0 0.494 log Length (km) 1.618 4.606 4.322 4.605 0.633

Table A2: Summary statistics: Chapter5 (standardized variables)

Statistic Min Max Mean Median St. Dev. Fractal Dimension −4.247 5.954 0.250 −0.216 1.035 Watershed −0.383 4.163 −0.006 −0.383 0.905 River −0.504 2.373 0.008 −0.504 0.979 Lake −0.157 7.834 −0.002 −0.157 0.947 Terrain ruggedness, log −3.402 2.310 0.068 0.048 0.969 Reestablished 0 1 0.058 0 0.234 Colonial 0 1 0.357 0 0.479 Population density, log −1.156 3.423 0.020 −0.115 0.999 Agric. Suit. −1.447 2.432 0.007 −0.141 0.988 Pre-1886 0 1 0.421 0 0.494 log Length (km) −2.521 1.025 0.688 1.024 0.751

145 ln.len.km.sc

pre86 0

ag.suit.sc −0.1 −0.1

ln.pop.d.sc 0.5 −0.1 −0.2

colonial −0.2 0.1 −0.3 0 1.0

0.5 restored 0.1 0.1 0.1 0 −0.2 0.0

−0.5 ln.elev.sd.sc 0 −0.2 0.1 −0.1 0.2 0 −1.0

lake.share.sc−0.1 0 0 −0.1 −0.1 0 0

rv.share.sc 0 −0.2 0.1 0 0.1 0.1 0.1 −0.1

ws.share.sc−0.1 −0.1 0.1 0 0 0 0 0 0

fd.sc 0.1 0 0 0.1 0 −0.2 0.2 0.1 0 0.2

Figure A1: Correlation plot: Main explanatory variables

146 Figure A2: Average border straightness and alignment over time

Figure A3: Distribution of segment lengths across alternative datasets

147 Table A3: Grambsch-Thernau proportitionality test (Model 4 in Table 5.1)

rho chisq p Fractal Dimension 0.063 7.419 0.006 Watershed -0.036 2.284 0.131 River -0.057 6.500 0.011 Lake -0.032 1.671 0.196 Terrain ruggedness, log 0.046 7.385 0.007 Reestablished -0.255 84.723 0 Colonial 0.049 32.598 0 Population density, log -0.047 12.670 0.0004 Agric. Suit. -0.007 0.112 0.738 Pre-1886 0.235 351.848 0 log Length (km) -0.067 8.075 0.004 GLOBAL 497.112 0 Notes: Significant p-values indicate violations of the pro- portional hazards assumption

Table A4: Grambsch-Thernau proportitionality test (Model 4 in Table 5.2)

rho chisq p Fractal Dimension 0.004 0.043 0.835 Watershed -0.0004 0.0005 0.982 River -0.045 6.563 0.010 Lake -0.020 0.618 0.432 Terrain ruggedness, log -0.080 18.468 0.00002 Reestablished -0.139 212.203 0 Colonial 0.135 156.412 0 Population density, log -0.093 22.049 0.00000 Agric. Suit. 0.062 6.573 0.010 Pre-1886 0.131 147.977 0 log Length (km) 0.035 1.950 0.163 GLOBAL 430.662 0 Notes: Significant p-values indicate violations of the pro- portional hazards assumption

148 A.2 Robustness

Table A5: OLS models. Dependent variable: Border change

Model 1 Model 2 Model 3 Model 4 Fractal Dimension −0.0009∗∗ −0.0006∗∗ −0.0009∗∗ −0.0006∗∗ (0.0003)(0.0002)(0.0003)(0.0002) Watershed −0.0008∗∗∗ −0.0006∗∗∗ −0.0008∗∗∗ −0.0006∗∗∗ (0.0002)(0.0002)(0.0002)(0.0002) River −0.0010∗∗∗ −0.0008∗∗ −0.0011∗∗∗ −0.0009∗∗ (0.0003)(0.0003)(0.0003)(0.0003) Lake −0.0004∗∗ −0.0001 −0.0004∗∗ −0.0001 (0.0002)(0.0001)(0.0002)(0.0001) Mountains −0.0005 −0.0003 −0.0006 −0.0005 (0.0004)(0.0005)(0.0004)(0.0005) Reestablished −0.0043∗∗∗ −0.0065∗∗ −0.0011 −0.0034∗ (0.0012)(0.0020)(0.0015)(0.0020) Colonial 0.0008 0.0008 −0.0002 0.0003 (0.0010)(0.0020)(0.0010)(0.0020) log Pop Dens. 0.0008 −0.0001 0.0010∗ −0.0000 (0.0005)(0.0006)(0.0005)(0.0006) Agric. Suit. 0.0003 0.0000 0.0003 0.0000 (0.0004)(0.0005)(0.0004)(0.0005) Pre-1886 0.0014 0.0016 0.0000 0.0001 (0.0010)(0.0015)(0.0012)(0.0020) log Length (km) −0.0004 −0.0001 −0.0005 −0.0001 (0.0003)(0.0003)(0.0003)(0.0003) t 0.0002 0.0004∗∗ 0.0000 0.0001 (0.0001)(0.0001)(0.0001)(0.0001) t2 −0.0000∗∗ −0.0000∗∗∗ −0.0000 −0.0000 (0.0000)(0.0000)(0.0000)(0.0000) t3 0.0000∗∗ 0.0000∗∗∗ 0.0000 0.0000 (0.0000)(0.0000)(0.0000)(0.0000) Dyad FE no yes no yes Year FE no no yes yes Num. obs. 291877 291877 291877 291877 Adj. R2 (full model) 0.0029 0.0147 0.0139 0.0258 Adj. R2 (proj model) 0.0029 0.0009 0.0006 −0.0005 Notes: OLS Models. Unit of analysis: Border segments (max length: 100 km). All continuous variables have been standardized to facilitate interpretation. Robust standard errors are clustered on the dyad level. Significance levels: ∗p<0.1, ∗∗p<0.05, ∗∗∗p<0.01

Max. segment length (Km) ● 25 ● 50 ● 100 ● 200 ● 400

Fractal Dimension Watershed River Reestablished

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

−0.4 −0.3 −0.2 −0.1 0.0 −0.4 −0.3 −0.2 −0.1 0.0 −0.5 −0.4 −0.3 −0.2 −0.1 0.0 −2 −1 0 Coefficient and 95% / 90% Confidence Intervals

Figure A4: Results from model 2 in Table 5.1 across alternative datasets with varying segment lengths

149 150 Appendix: Vicious cycles? (Chapter6)

B.1 Descriptive statistics

lon

lat

col.rule.past

ln.riv.len

ln.watershed.len

ln.elev.sd 1.0

oil.fld.ct 0.5

0.0 min.dep.ct −0.5

ag.suit.max −1.0

greg.elf

ln.p.dens

avg.bdens

ln.b.time.pre.ww2

gw.any.chg.ct.pre.ww2

Figure B1: Correlation plot: Main explanatory variables

151 Table B1: Summary statistics: Chapter 6

Statistic Min Max Mean Median St. Dev. Past border chg. 0 5 0.512 0 0.868 Hist. border variability 0 61 3.015 0 7.028 Years without b. chg, log 0.000 4.007 3.709 4.007 0.612 Hist. border dens. 0 15 1.349 0 2.448 Population density, log 0.000 5.838 1.234 0.704 1.373 ELF (Greg) 0.000 0.844 0.242 0.164 0.247 Agr. suitability 0 6 4.183 5 1.791 Mineral deposits 0 65 1.106 0 2.915 Oil fields 0 9 0.686 0 1.266 Terrain ruggedness, log 0.199 7.627 4.952 4.925 1.043 Watershed length, log 0.000 7.608 2.849 2.649 2.878 River length, log 0.000 7.200 3.796 5.295 2.645 Colonial past 0 1 0.376 0 0.485 Latitude −54.616 81.705 23.076 26.975 30.528 Longitude −179.921 179.418 20.295 27.975 81.888

152 B.2 Robustness

153 Table B2: Main results, alternative measures of border instability

(1) (2) (3) (4) (5) (6) Interstate Interstate Domestic Domestic Border Chg. Border Chg. (MITC) (MITC) (SDM) (SDM) (Post-46) (Post-46) Hist. border variability 0.028∗∗ 0.030∗∗∗ 0.041∗∗∗ (0.012)(0.009)(0.011) Years without b. chg, log −0.423∗∗ −0.413∗∗∗ −0.398 154 (0.180)(0.129)(0.278) Controls yes yes yes yes yes yes Region FE yes yes yes yes yes yes Log Likelihood −451.566 −447.045 −755.029 −751.247 −1006.384 −1009.880 Observations 2776 2776 2776 2776 2776 2776 Notes: Logit models. Robust standard errors clustered on the country level in 1946. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01. Table B3: Main results, regional sub-sample: Africa

(1) (2) (3) Interstate Domestic Border Chg. (MITC) (SDM) (Post-46) Past border chg. 0.365∗∗ 0.403∗∗∗ 0.468∗ (0.166)(0.155)(0.250) Intercept −4.327∗∗∗ −2.866∗∗ −7.668∗∗∗ (1.458)(1.255)(1.512) Region FE no no no Log Likelihood −163.404 −240.613 −236.784 Observations 608 608 608 Notes: Logit models. Robust standard errors clustered on the country level in 1946. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

Table B4: Main results, regional sub-sample: World except Europe

(1) (2) (3) Interstate Domestic Border Chg. (MITC) (SDM) (Post-46) Past border chg. 0.501∗∗∗ 0.496∗∗∗ 0.744∗∗ (0.172)(0.154)(0.329) Intercept −5.431∗∗∗ −5.310∗∗∗ −4.278∗∗ (1.203)(1.011)(1.688) Region FE no no no Log Likelihood −291.460 −516.403 −747.978 Observations 2168 2168 2168 Notes: Logit models. Robust standard errors clustered on the country level in 1946. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

Table B5: Main results, controlling for conflict persistence

(1) (2) (3) Interstate Domestic Border Chg. (MITC) (SDM) (Post-46) Past border chg. 0.393∗∗∗ 0.324∗∗∗ 0.610∗∗∗ (0.116)(0.119)(0.196) Log War Years (pre-1945) −0.199 0.069 0.281 (0.137)(0.099)(0.269) Intercept −5.466∗∗∗ −6.782∗∗∗ −6.524∗∗∗ (1.189)(0.999)(1.010) Controls yes yes yes Region FE yes yes yes Log Likelihood −445.242 −752.190 −982.174 Observations 2776 2776 2776 Notes: Logit models. Robust standard errors clustered on the country level in 1946. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

155 Figure B2: Covariate balance after matching

Table B6: Main results after matching

(1) (2) (3) Interstate Domestic Border Chg. (MITC) (SDM) (Post-46) Past border chg. 0.350∗∗∗ 0.306∗∗∗ 0.601∗∗∗ (0.129)(0.114)(0.171) Intercept −4.929∗∗∗ −6.236∗∗∗ −5.304∗∗∗ (1.047)(1.144)(1.251) Region FE yes yes yes Log Likelihood −363.330 −484.879 −596.725 Observations 1405 1405 1405 Notes: Logit models. Unit of analysis: Grid cells (50’000 Sq Km). Regions consist of: Western Hemisphere, Europe, Africa, Asia, Australia and Oceania. Robust standard errors clustered on the country level in 1946. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

156 Figure B3: Connectivity matrix used in spatial filtering

Table B7: Main results after spatial filtering

(1) (2) (3) Interstate Domestic Border Chg. (MITC) (SDM) (Post-46) Past border chg. 0.056∗∗ 0.095∗∗ 0.137∗∗∗ (0.024)(0.040)(0.043) Hist. border dens. 0.022∗∗∗ 0.013 0.014∗ (0.006)(0.008)(0.008) Population density, log −0.002 0.068∗∗∗ −0.004 (0.014)(0.024)(0.024) ELF (Greg) 0.034 0.247∗∗∗ −0.079 (0.035)(0.070)(0.054) Agr. suitability −0.005 −0.008 −0.014 (0.007)(0.010)(0.015) Mineral deposits 0.002 0.004 0.008 (0.003)(0.007)(0.006) Oil fields 0.013 0.016 0.012 (0.012)(0.014)(0.010) Terrain ruggedness, log 0.020 0.046∗∗ −0.015 (0.020)(0.022)(0.019) Watershed length, log −0.003 0.009∗∗ 0.000 (0.004)(0.004)(0.006) River length, log 0.005 −0.005 0.002 (0.005)(0.006)(0.006) Colonial past 0.188∗∗ 0.263∗∗ 0.416∗∗∗ (0.092)(0.106)(0.129) Intercept −0.153 −0.367∗∗∗ 0.106 (0.128)(0.125)(0.111) Moran Eigenvectors yes yes yes Region FE no no no Log Likelihood −1019.716 −1791.171 −1444.378 Observations 2776 2776 2776 Notes: Linear Probability Models. Robust standard errors clustered on the country level in 1946. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

157 B.3 Creating voronoi cells

1 2

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ●● ●● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ●● ● ●● ●● ● ●● ●●● ● ● ● ● ● ● ● ● ●●● ●● ● ●● ●● ● ●● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ●● ● ● ●● ● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ●● ● ● ●● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●●● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●●● ●●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●●● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ●● ● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ●●●● ● ● ● ● ● ● ●● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ●●●● ● ● ● ● ● ● ●● ●●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●●●● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●●●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●●● ● ●● ● ● ● ●●● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●●● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ●●● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ●● ●● ● ●● ● ● ●● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●●● ● ●●● ● ● ● ● ●● ● ● ●●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ●● ●● ●● ● ● ●● ● ● ● ●● ● ●● ●● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ●●● ● ● ● ● ●● ●● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●●● ● ●● ● ● ●●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ●

3 4

● ● ● ●

● ● ● ● ● ●

● ●

● ● ● ● ● ● ● ● ● ●

● ●

● ● ● ● ● ● ● ●

● ●

● ●

● ●

● ● ● ●

● ●

● ●

● ●

● ● ● ●

● ● ● ● ● ●

● ●

Figure B4: Dividing large areas into equal-area Voronoi polygons. (1): Cover the area by a set of N random points. (2): Using a K-means clustering algorithm, divide the points into K clusters. (3:) Locate the centroid of each point cluster. (4): Use Voronoi tesselation on all centroids to divide the region into K chunks, which approximately fall within local UTM projection zones.

158 5 6

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ●●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ●● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ●● ● ●● ●● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ●●● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●●●● ● ● ● ● ● ●● ●● ●● ● ● ●● ● ● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ●● ● ● ● ● ●●● ● ● ● ●● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● 7 ● ● ● ● ● ● ● ●8 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●●● ● ●●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ●●● ● ● ●●●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ●● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ●● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●● ● ●●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Figure B5: Dividing large areas into equal-area Voronoi polygons (continued). (5): Re-project each chunk using the most adequate UTM projection and calcu- late its total area size in square kilometers, S. (6): Within each chunk, divide points into K clusters, where S/K represents the desired size of each polygon. (7): Extract centroids from clusters, and create a single grid of centroids that covers the entire area. (8): Use Voronoi tesselation to obtain equal-area cells.

159 Figure B6: Distribution of cell sizes (1000 Sq Km) across alternative grid specifi- cations.

160 B.4 Data: GeoSDM

To examine the relationship between border change and domestic territorial disputes, Chapter6 uses data from GeoSDM, 1 a new dataset that maps the territorial claims of 466 self-determination movements (SDM) across the globe between 1946 and 2012, based on the SDM dataset by Sambanis et al.(2018). For each movement, SDM tracks its activity over time and provides a host of additional variables for a random subsample of 106 groups. SDM is the most comprehensive data source on separatist movements to date, and is designed to overcome the selection bias built into other datasets such as the Minori- ties at Risk (MAR) and the Center of International Development and Conflict Management (CIDCDM) data. As Sambanis et al.(2018) have demonstrated, the over-representation of highly mobilized SDMs in developing countries in existing datasets is likely to produce biased inferences about the causes and dynamics of separatism. SDM defines self-determination as a “group-defined right to self-rule within the boundaries of a territory”. As becomes clear from this definition, self-determination claims are inextricably linked to a specific territory. The goal of GeoSDM is to map the territorial claims of each movement. Demands for self-determination can range from moderate calls for regional autonomy within the state to the more radical goal of secession. More specifically, SDM distinguishes between four types of dominant claims, which are coded on a yearly basis: Regional autonomy, sub-state secession, independence and irredentism. While the first two claims aim to improve a group’s status within the existing boundaries of the state, the latter two explicitly call for border adjustments. In many instances, SDMs are internally divided, with some representatives making more radical claims than others. In these cases, SDM records the claims that enjoy the broad- est support within the movement in a particular year. Although the original dataset only codes dominant claims for a random sample of 106 groups, the authors and I have extended this claim coding to all 470 groups. Figure B7 shows the distribution of SDMs and dominant claims over time. Although GeoSDM ultimately aims to map all SDM claims, the current version only covers a subsample of 280 groups. This sample consists of all 213 groups that have made secessionist or irredentist demands at any point in time, but also includes 67 remaining groups from the random sample that have made more moderate claims. However, the analyses in Chapter6 focus exclusively

1I would like to thank Micha Germann, Andreas Schädel and Nicholas Sambanis for sharing all the supplementary information to their dataset, which has been extremely valuable. I also thank Lukas Dick, Irina Siminichina, Benjamin Füglister and Katherine Woolbright for their research assistance.

161 Figure B7: SDM: Dominant claims over time on the subset of secessionist and irredentist movements, as these are the most relevant to my argument. Fortunately, identifying SDM territorial claims has proved to be relatively straight-forward because the SDM dataset comes with a comprehensive doc- umentation that offers extensive background information on each movement and relies on a number of in-depth sources that often describe territorial claims in more detail, such as Minahan(1996, 2002) and Hewitt and Cheetham(2000). Moreover, some SDMs that are presently active even maintain websites that provide descriptions and maps of the territories groups claim as homelands. Another key resource is Roth(2015), which provides a comprehensive overview of separatist claims in recent history, along with detailed maps. For the most part, I relied on these sources to gather information on the territories claimed by each SDM. As it turned out, the vast majority of SDM claims map neatly onto current or past administrative boundaries, and can therefore be easily identified. 2 For example, the territory claimed by the Irish Republican Army consists of the Province of Northern Ireland, as shown on numerous maps (Minahan, 2002). In a smaller subset of cases, groups make claims that do not match existing administrative divisions. For example, Libya’s Toubou have demanded the independence of a loosely defined territory in the South that corresponds to the group’s nomadic settlement patterns, while the Czech Republic’s Moro- vians have pushed for independence for the historic region of Morovia, which stretches across the country’s current administrative divisions.

2This is an interesting finding by itself, as it suggests that most of today’s separatist move- ments seem to comply with the uti possidetis norm, which states that the borders of new states should follow previous (administrative) boundaries (Shaw, 1997).

162 To facilitate data collection in cases where SDM claims do not coincide with existing administrative boundaries, I distinguish between three scenar- ios. First, some movements have made self-determination demands that are best represented by the ethnic group’s settlement territory. This applies to many indigenous movements in Latin America, which revolve around cultural autonomy and land rights for specific groups, whereas the spatial extent of these claims often remains ambiguous. Second, some groups have explicitly called for the restoration of old kingdoms or historical polities, as shown by the Morovian example. Lastly, some claims are delimited at least in part by physical geography as is the case for many island nations, or groups that claim territories defined in part by natural features, such as rivers or mountains. After collecting information on each territorial claim, I proceeded to geocode them, relying on existing GIS data sources as far as possible. For claims that align with current or past administrative units, I relied on the Database of Global Administrative Areas (GADM) (Hijmans, 2012) and on the Federal Administrative Boundaries (FAB) dataset (Deiwiks, 2011). For claims that are based on ethnic settlement territories, I rely on the GeoEPR (Vogt et al., 2014) and GREG (Weidmann, Rod, & Cederman, 2010) datasets, depending on which source provides the best representation of the SDM’s claim. Claims that align with natural features such as islands, rivers and lakes are coded using a combination of the GADM dataset and the Natural Earth GIS data (Natural Earth, 2015). In the remaining cases, I rely on digitized maps to draw polygons, based primarily on Roth(2015), and a number of online sources. While most territorial claims were relatively easy to identify, there were also some ambiguities. One main issue is that some territorial claims appear to be poorly defined. For example, descriptions and maps of the territory claimed by Kurdish separatists in Turkey are surprisingly vague and contradictory. A second, related problem is that actors within the same movement have made competing claims at the same time. This is also true in the Kurdish case, where proposals for an independent Kurdish state have varied substantially, with some claims limited to territories within Eastern or South-Eastern Turkey and others extending far across the border to encompass a "greater Kurdistan" (Roth, 2015). In this case, I have coded the "dominant" territorial claim within each movement, following SDM dataset’s general coding rules.3 Finally, one remaining challenge is that some claims have evolved over time (Mylonas & Shelef, 2014, 2017). Changes in a movement’s territorial claim can either be the result of border change, political concessions or changes in an SDM’s

3The dominant claim is either the claim expressed by the leading organization within the movement or the majority of organizations.

163 leadership or stated objectives. For example, the Chechen movement in Russia initially demanded increased autonomy for the Checheno-Ingush ASSR from 1989 onwards. After a leadership change in 1991, the movement unilaterally declared independence of Chechnya and limited its claims to this particular region. In the current version of GeoSDM, such time-varying claims are flagged but not coded. Instead, I have coded each SDM’s initial claim as it was made at the start of the dispute. In the Chechen case, this means that I have coded the Checheno-Ingush ASSR throughout the entire period. Figure B8 maps all secessionist and irredentist claims made between 1946 and 2012.

Figure B8: GeoSDM: Preview of territorial claims (groups coded so far)

164 Appendix: Dangerous precedents (Chapter7)

C.1 Descriptive statistics

Table C1: Summary statistics. W indicates spatially lagged variables.

Statistic Min Max Mean Median St. Dev. W: TC (neighborhood) 0.000 1.000 0.356 0.000 0.479 W: TC (common member dyads) 0.000 1.000 0.122 0.000 0.327 W: TC conflict (neighborhood) 0.000 1.000 0.225 0.000 0.417 W: TC coop. (neighborhood) 0.000 1.000 0.220 0.000 0.414 W: TC conflict (common member dyads) 0.000 1.000 0.074 0.000 0.262 W: TC coop. (common member dyads) 0.000 1.000 0.063 0.000 0.244 W: Disputes (neighborhood) 0.000 1.000 0.927 1.000 0.260 W: Disputes (common member dyads) 0.000 1.000 0.410 0.000 0.492 TC (dyad) 0 1 0.036 0 0.187 Ongoing dispute (dyad) 0.000 1.000 0.135 0.000 0.342 Joint democracy 0.000 1.000 0.150 0.000 0.357 Rel. capabilities −0.382 0.382 0.000 0.000 0.057 Past MIDs 0.000 13.000 0.527 0.000 1.532 Dyad age (log) 0.000 5.220 3.231 3.367 1.127 Neighbors (count) 0.000 36.000 9.320 9.000 4.246 World wars 0 1 0.040 0 0.196 TC (world) 0 29 4.370 3 4.290 Systemic instability 0.000 0.033 0.005 0.003 0.005

Figure C1: Territorial changes by region

165 incidence.lag

ww 0

pastmids 0.1 0.1

cinc.diff 0 0 0

1.0 jointdem 0 0 0 −0.1 0.5

0.0 ll.incidence.5.cmd.100−0.1 0.2 0.1 0 0.2 −0.5

−1.0 ll.incidence.5.nbd.1000.4 −0.1 0 0.1 0.1 0.1

l.any.tc.5 0.1 0.1 0 0 0.1 0 0.1

ll.any.tc.5.cmd.1000.2 0.1 0.2 0 0.1 0.1 0 0

ll.any.tc.5.nbd.1000.3 0.1 0.3 0.1 0.1 0 0.1 0 0

Figure C2: Correlation plot of the main explanatory variables used in first analysis

incidence.lag

ww 0

pastmids 0.1 0.1

cinc.diff 0 0 0

jointdem 0 0 0 −0.1

ll.incidence.5.cmd.100−0.1 0.2 0.1 0 0.2 1.0 0.5 ll.incidence.5.nbd.1000.4 −0.1 0 0.1 0.1 0.1 0.0

−0.5 l.coop.5 0.1 0 0 0 0.1 0 0.1 −1.0 l.confl.5 0.1 0.1 0 0 0 0.1 0 0.1

ll.coop.5.cmd.1000 0.1 0.1 0.1 0 0.1 0.1 0 0

ll.confl.5.cmd.1000.3 0.2 0.1 0.1 0.2 0 0.1 0.1 0 0

ll.coop.5.nbd.1000.1 0.1 0 0.1 0.3 0.1 0.1 0 0 0 0

ll.confl.5.nbd.1000.4 0.3 0.1 0.2 0.1 0.2 0.1 0.1 0 0.1 0 0

Figure C3: Correlation plot of the main explanatory variables used in second analysis

166 C.2 Results not shown in main text

Table C2: Logit models: Territorial dispute onset, 1816-1945

Model 1 Model 2 Model 3 Model 4 W: TC (neighborhood) 0.62∗∗∗ 0.66∗∗ 0.55∗∗ 0.46∗ (0.24)(0.26)(0.25)(0.24) W: TC (common member dyads) 0.56∗∗ 0.48∗ 0.49∗ 0.41 (0.25)(0.26)(0.25)(0.26) W: Disputes (neighborhood) −0.12 −0.06 −0.09 −0.22 (0.62)(0.77)(0.76)(0.72) W: Disputes (common member dyads) 0.28 0.31 0.29 0.41∗ (0.24)(0.23)(0.22)(0.22) TC (dyad) 1.42∗∗∗ 1.32∗∗∗ 1.27∗∗∗ 1.25∗∗∗ (0.30)(0.32)(0.30)(0.31) Ongoing dispute (dyad) 0.05 −0.13 −0.31 −0.20 (0.40)(0.47)(0.41)(0.41) TC (world) 0.06∗∗∗ 0.06∗∗∗ (0.02)(0.02) Intercept −5.75∗∗∗ −6.13∗∗∗ −6.72∗∗∗ −7.29∗∗∗ (0.69)(0.82)(0.84)(0.90) t, t2, t3 yes yes yes yes Controls yes yes yes yes Region FE no no no yes Log Likelihood −572.96 −532.43 −508.05 −502.25 Observations 17299 16806 16806 16806 Notes: Unit of analysis: directed dyad-years (contiguous states only). Dyads consist of a potential challenger and a target state in a territorial dispute. W denotes spatial lags on variables among common-member and neighboring dyads. TC is short for territorial changes. Main explanatory variables in bold. Robust standard errors clustered on the (undirected) dyad level. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

167 Table C3: Logit models: Territorial dispute onset, 1816-1945

Model 1 Model 2 Model 3 Model 4 W: TC conflict (neighborhood) 0.93∗∗∗ 0.95∗∗∗ 0.80∗∗∗ 0.60∗∗ (0.24)(0.26)(0.26)(0.25) W: TC coop. (neighborhood) 0.06 0.05 0.02 0.08 (0.30)(0.31)(0.28)(0.28) W: TC conflict (common member dyads) 0.35 0.24 0.25 0.10 (0.27)(0.27)(0.26)(0.28) W: TC coop. (common member dyads) −0.21 −0.11 0.01 0.15 (0.38)(0.39)(0.35)(0.36) W: Disputes (neighborhood) −0.17 −0.03 −0.07 −0.20 (0.63)(0.77)(0.76)(0.72) W: Disputes (common member dyads) 0.38∗ 0.40∗ 0.35∗ 0.43∗∗ (0.23)(0.22)(0.21)(0.22) TC Conflict (dyad) 1.43∗∗∗ 1.27∗∗∗ 1.24∗∗∗ 1.18∗∗∗ (0.35)(0.39)(0.36)(0.35) TC Coop. (dyad) 1.02∗∗∗ 1.15∗∗∗ 1.14∗∗∗ 1.26∗∗∗ (0.39)(0.37)(0.38)(0.39) Ongoing dispute (dyad) 0.20 −0.05 −0.26 −0.17 (0.39)(0.45)(0.41)(0.41) TC (world) 0.05∗∗∗ 0.06∗∗∗ (0.02)(0.02) Intercept −5.79∗∗∗ −6.17∗∗∗ −6.69∗∗∗ −7.18∗∗∗ (0.70)(0.82)(0.84)(0.91) t, t2, t3 yes yes yes yes Controls yes yes yes yes Region FE no no no yes Log Likelihood −567.38 −528.28 −505.55 −501.31 Observations 17299 16806 16806 16806 Notes: Unit of analysis: directed dyad-years (contiguous states only). Dyads consist of a potential challenger and a target state in a territorial dispute. W denotes spatial lags on variables among common-member and neighboring dyads. TC is short for territorial changes. Main explanatory variables in bold. Robust standard errors clustered on the (undirected) dyad level. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

168 Table C4: Logit models: Territorial dispute onset, 1946-2001

Model 1 Model 2 Model 3 Model 4 W: TC (neighborhood) 0.58∗∗∗ 0.60∗∗∗ 0.53∗∗∗ 0.43∗∗∗ (0.15)(0.16)(0.16)(0.16) W: TC (common member dyads) 0.58∗∗∗ 0.52∗∗∗ 0.47∗∗∗ 0.46∗∗∗ (0.17)(0.17)(0.18)(0.18) W: Disputes (neighborhood) −0.22 −0.26 −0.29 −0.37 (0.33)(0.35)(0.35)(0.34) W: Disputes (common member dyads) 0.11 0.10 0.08 0.08 (0.15)(0.15)(0.14)(0.14) TC (dyad) 1.44∗∗∗ 1.40∗∗∗ 1.29∗∗∗ 1.32∗∗∗ (0.24)(0.25)(0.26)(0.25) Ongoing dispute (dyad) −0.31 −0.28 −0.34 −0.36 (0.26)(0.31)(0.29)(0.29) TC (world) 0.06∗∗∗ 0.06∗∗∗ (0.01)(0.01) Intercept −5.23∗∗∗ −4.88∗∗∗ −5.34∗∗∗ −6.03∗∗∗ (0.40)(0.44)(0.44)(0.47) t, t2, t3 yes yes yes yes Controls yes yes yes yes Region FE no no no yes Log Likelihood −1217.51 −1172.39 −1145.19 −1134.17 Observations 44681 43834 43834 43834 Notes: Unit of analysis: directed dyad-years (contiguous states only). Dyads consist of a potential challenger and a target state in a territorial dispute. W denotes spatial lags on variables among common-member and neighboring dyads. TC is short for territorial changes. Main explanatory variables in bold. Robust standard errors clustered on the (undirected) dyad level. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

169 Table C5: Logit models: Territorial dispute onset, 1946-2001

Model 1 Model 2 Model 3 Model 4 W: TC conflict (neighborhood) 0.11 0.11 0.03 0.01 (0.21)(0.21)(0.20)(0.20) W: TC coop. (neighborhood) 0.55∗∗∗ 0.61∗∗∗ 0.66∗∗∗ 0.52∗∗ (0.21)(0.21)(0.21)(0.23) W: TC conflict (common member dyads) 0.76∗∗∗ 0.76∗∗∗ 0.68∗∗ 0.76∗∗∗ (0.28)(0.26)(0.27)(0.27) W: TC coop. (common member dyads) 0.15 0.25 0.28 0.18 (0.33)(0.34)(0.34)(0.35) W: Disputes (neighborhood) −0.31 −0.40 −0.46 −0.72∗∗ (0.37)(0.35)(0.34)(0.32) W: Disputes (common member dyads) −0.04 −0.07 −0.08 −0.22 (0.19)(0.18)(0.18)(0.19) TC Conflict (dyad) 1.36∗∗∗ 1.45∗∗∗ 1.30∗∗∗ 1.41∗∗∗ (0.36)(0.36)(0.34)(0.34) TC Coop. (dyad) 0.92∗ 1.14∗∗ 1.16∗∗ 1.07∗∗ (0.53)(0.53)(0.52)(0.51) Ongoing dispute (dyad) −0.56∗ −0.16 −0.13 −0.34 (0.33)(0.39)(0.37)(0.38) TC (world) 0.05∗∗ 0.06∗∗ (0.03)(0.03) Intercept −4.73∗∗∗ −3.93∗∗∗ −4.16∗∗∗ −4.66∗∗∗ (0.44)(0.50)(0.52)(0.56) t, t2, t3 yes yes yes yes Controls yes yes yes yes Region FE no no no yes Log Likelihood −640.17 −631.82 −626.41 −619.20 Observations 27382 27028 27028 27028 Notes: Unit of analysis: directed dyad-years (contiguous states only). Dyads consist of a potential challenger and a target state in a territorial dispute. W denotes spatial lags on variables among common-member and neighboring dyads. TC is short for territorial changes. Main explanatory variables in bold. Robust standard errors clustered on the (undirected) dyad level. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

170 C.3 Robustness

Table C6: OLS models: Territorial dispute onset, 1816-2001

Model 1 Model 2 Model 3 Model 4 W: TC (neighborhood) 0.003∗∗∗ 0.003∗∗∗ (0.001)(0.001) W: TC conflict (neighborhood) 0.003∗∗ 0.002∗∗ (0.001)(0.001) W: TC coop. (neighborhood) 0.002 0.002 (0.001)(0.001) Dyad FE yes yes yes yes Controls no yes no yes t, t2, t3 yes yes yes yes Num. obs. 44681 43834 44681 43834 Adj. R2 (full model) 0.017 0.020 0.014 0.017 Adj. R2 (proj model) −0.014 −0.011 −0.017 −0.013 Notes: Unit of analysis: directed dyad-years (contiguous states only). Dyads consist of a po- tential challenger and a target state in a territorial dispute. W denotes spatial lags on variables among common-member and neighboring dyads. TC is short for territorial changes. Main ex- planatory variables in bold. Robust standard errors clustered on the (undirected) dyad level. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

Table C7: OLS models: Territorial dispute onset, 1816-2001

Model 1 Model 2 Model 3 Model 4 W: TC (neighborhood) 0.002∗∗∗ 0.002∗∗ (0.001)(0.001) W: TC conflict (neighborhood) 0.002∗∗ 0.002∗ (0.001)(0.001) W: TC coop. (neighborhood) 0.002∗ 0.001 (0.001)(0.001) Year FE yes yes yes yes Controls no yes no yes t, t2, t3 yes yes yes yes Num. obs. 44681 43834 44681 43834 Adj. R2 (full model) 0.013 0.013 0.011 0.011 Adj. R2 (proj model) 0.003 0.003 0.001 0.001 Notes: Unit of analysis: directed dyad-years (contiguous states only). Dyads consist of a po- tential challenger and a target state in a territorial dispute. W denotes spatial lags on variables among common-member and neighboring dyads. TC is short for territorial changes. Main ex- planatory variables in bold. Robust standard errors clustered on the (undirected) dyad level. Significance levels: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.

171 C.4 Data: Coding conflictual vs. cooperative changes

Additional data preparation for this chapter involved coding a dummy variable to distinguish "conflictual" from "cooperative" instances of territorial change. This was done for each of the 837 exchanges of territory recorded by the Territo- rial Change dataset (Tir et al., 1998), in part automatically and in part manually. Cooperative changes occur with the explicit bilateral agreement of the gain- ing and losing side, while conflictual changes are unilateral, i.e. imposed against the will of the losing party. According to this definition, conflictual cases are often, but not always violent, as some instances involve the threat of violence, or the use of non-lethal force. My coding of conflictual changes is therefore broader than Goertz and Diehl’s definition of violent territorial change (Goertz & Diehl, 1992), as it also includes cases in which the threat of violence is implicit. However, the coding of cooperative exchanges of territory is largely consistent with a common definition of peaceful territorial change (Kacowicz, 1994). I coded the variable largely based on the information contained within the Territorial Change Dataset (TC) (Tir et al., 1998). The dataset contains a variable procedure, which distinguishes between six different types of territorial change: Conquest, annexation, cession, secession, unification and mandated territory. Cession refers to instances in which territory is ceded as a result of plebiscites, purchases, compensating agreements or as a consequence of hostilities (Ibid.). In addition, the data codes a separate variable for cases in which a state be- comes independent. Some categories fit neatly into the "conflictual-cooperative" scheme, but others do not: All instances of conquest and annexation were assigned to the conflictual category, while all cases where territories came under an international mandate were coded as cooperative. The remaining 602 cases are mixed, and required some further sorting. These cases include all changes coded as cessions, secessions, unifications, and independence. To further nar- row down this list, I relied on the TC dataset’s conflict variable that flags those changes that were the result of militarized conflict, which are inherently con- flictual. Lastly, I code all instances of (post-World War II) colonial independence as cooperative, unless they involved violence according to TC. This left me with 453 cases that required hand-coding, for which I relied on the Encyclopaedia of International Boundaries (Biger, 1995), the documen- tation of the CShapes 2.0 dataset, and other relevant sources such such as the Encyclopaedia Britannica. In most cases, the relevant information could easily be found. For example, many cessions were the result of purchases, leases, or resulted from bilateral agreements or international arbitration to settle disputes. Likewise, some secessions or unification were the result of referenda or bilat-

172 eral agreements, as for example the dissolution of Czechoslovakia in 1993 or German Reunification in 1990. These cases were all coded as cooperative. In contrast, some transfers where the result of unilateral actions or involved the threat of force, such as Senegal’s secession from Mali in 1960, which followed a military stand-off, or Russia’s seizure of the Amur province from China, which occurred under the threat of force. In 15 remaining cases, no information about the circumstances of the exchange could be found. These cases were dropped from the analysis.

Conflictual Cooperative Conquest Annexation Mandated territory Cession (Conf.) Cession (Coop.) Secession (Conf.) Secession (Coop) Unification (Conf.) Unification (Coop.) Independence (Conf.) Independence (Coop.)

Table C8: Coding conflictual vs. cooperative territorial changes

173 174 References

Abadie, A., Diamond, A., & Hainmueller, J. (2010). Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program. Journal of the American statistical Association, 105(409), 493– 505.

Abbott, A. (1995). Sequence Analysis: New Methods for Old Ideas. Annual Review of Sociology, 21(1), 93–113. doi: 10.1146/annurev.so.21.080195.000521

Abramson, S. (2017). The economic origins of the territorial state. International Organization, 17(11), 97–130.

Abramson, S., & Carter, D. (2018). Systemic Instability and the Emergence of Border Disputes. Working Paper.

Abramson, S., & Carter, D. B. (2016). The Historical Origins of Territorial Disputes. American Political Science Review, 111(4), 675–698.

Agnew, J. (1994, mar). The territorial trap: The geographical assumptions of international relations theory. Review of International Political Economy, 1(1), 53–80.

Agnew, J. (2008). Borders on the mind: re-framing border thinking. Ethnics and Global Politics 1.4, 1(4), 175–191.

Alesina, A., Easterly, W., & Matuszeski, J. (2011). Artificial States. Journal of the European Economic Association, 9(2), 246–277.

Alesina, A., & Spolaore, E. (2003). The Size of Nations. Cambridge, Mas- sachusetts: The MIT Press.

Altonji, J. G., Elder, T. E., & Taber, C. (2005). Selection on Observed and Unobserved Variables: Assessing the Effectiveness of Catholic Schools. Journal of Political Economy, 113(1), 151–84.

Ambrosio, T. (2001). Irredentism. Ethnic Conflict and International Politics. Westport: Greenwood Publishing Group.

Ambrosio, T. (2016). The rhetoric of irredentism: The Russian Federation’s perception management campaign and the annexation of Crimea. Small Wars & Insurgencies, 27(3), 467–490.

Anderson, B. (2006). Imagined communities: Reflections on the origin and spread of nationalism. London: Verso Books.

175 REFERENCES

Anderson, M. (1996). Frontiers. Territory and State Formation in the Modern World. Cambridge: Polity Press.

Andreas, P. (2003). Redrawing the line: Borders and security in the twenty-first century. International security, 28(2), 78–111.

Anselin, L. (2001). Spatial Econometrics. In B. Baltagi (Ed.), A companion to theoretical econometrics. (pp. 310–330). Blackwell Publishing.

Ardrey, R. (1969). The territorial imperativea personal inquiry into the animal origins of property and nations (No. 591.5 A7).

Atzili, B. (2012). Good Fences, Bad Neighbors: Border Fixity and International Conflict. Chicago: Chicago University Press.

Bartuseviˇcius,H., & Gleditsch, K. S. (2019). A Two-Stage Approach to Civil Conflict: Contested Incompatibilities and Armed Violence. International Orga- nization, 73(1), 225–248.

Baud, M., & Van Schendel, W. (1997). Toward a comparative history of borderlands. Journal of World History, 211–242.

Beck, N., Gleditsch, K., & Beardsley, K. (2006). Space Is More than Geography: Using Spatial Econometrics in the Study of Political Economy. International Studies Quarterly, 50(1), 27–44.

Beck, N., Katz, J., & Tucker, R. (1998). Taking time seriously: Time-series-cross- section analysis with a binary dependent variable. Americal Journal of Political Science, 1260-1288(42).

Bennett, R. W., & Zitomersky, J. (1982). The Delimitation of International Diplomatic Systems: 1816–1970. The Correlates of War Project’s Systems Re- considered. In J. Zitomersky (Ed.), On making use of history (pp. 67–129). Lund: Esselt Studium.

Besley, T., & Reynal-Querol, M. (2014). The Legacy of Historical Conflict: Evidence from Africa. American Political Science Review, 108(2), 319–336.

Biger, G. (1995). The Encyclopedia of International Boundaries. New York: Facts on File.

Borger, J. (2019, apr). Pompeo flounders on why annexation is good for the Golan but not for Crimea. Retrieved from https://www.theguardian.com/us-news/ 2019/apr/11/golan-heights

Bormann, N., & Savun, B. (2018). Reputation, concessions, and territorial civil war: Do ethnic dominoes fall, or don’t they? Journal of Peace Research.

Bouda, M., Caplan, J. S., & Saiers, J. E. (2016). Box-Counting Dimension Revisited: Presenting an Efficient Method of Minimizing Quantization Error and an Assessment of the Self-Similarity of Structural Root Systems. Frontiers in plant science, 7, 149. doi: 10.3389/fpls.2016.00149

176 REFERENCES

Boulding, K. (1964). Conflict and Defense. New York: Harper and Row.

Box-Steffensmeier, J. M., & Jones, B. S. (2004). Event history modeling: A guide for social scientists. Cambridge, Massachusetts: Cambridge University Press.

Box-Steffensmeier, J. M., Reiter, D., & Zorn, C. (2003). Nonproportional Hazards and Event History Analysis in International Relations. Journal of Conflict Resolution, 47(1), 33–53.

Branch, J. (2013). The Cartographic State: Maps, Territory, and the Origins of Sovereignty. Cambridge, Massachusetts: Cambridge University Press.

Branch, J. (2016). Geographic information systems (GIS) in international relations. International Organization, 70(4), 845–869.

Bremer, S. A. (1992). Dangerous Dyads: Conditions Affecting the Likelihood of Interstate War, 1816–1965. Journal of Conflict Resolution, 36(2), 309–41.

Brigham, A. P. (1919). Principles in the Determination of Boundaries. Geo- graphical Review, 7, 212–219.

Broek, J. O. M. (1943). The problem of ‘natural frontiers’. In Frontiers of the future (pp. 3–20).

Brownlie, I., & Burns, I. R. (1979). African Boundaries: A Legal and Diplomatic Encyclopaedia. London: C. Hurst & Co. Publishers.

Brueckner, J. K. (2003). Strategic interaction among governments: An overview of empirical studies. International regional science review, 26(2), 175–188.

Buhaug, H., & Gleditsch, K. S. (2008). Contagion or confusion? Why conflicts cluster in space. International Studies Quarterly, 52(2), 215–233.

Burke-White, W. B. (2014). Crimea and the international legal order. Survival, 56(4), 65–80.

Carnegie, A., & Carson, A. (2018). The Spotlight’s Harsh Glare: Rethinking Publicity and International Order. International Organization, 1–31.

Carter, D. B. (2016). History as a Double-Edged Sword: Historcal Boundaries and Territorial Claims. Politics, Philosophy & Economics.

Carter, D. B., & Goemans, H. (2014). The temporal dynamics of new interna- tional borders. Conflict Management and Peace Science, 31(3), 285–302.

Carter, D. B., & Goemans, H. (2018). International Trade and coordination: tracing border effects. World Politics, 70(1), 1–52.

Carter, D. B., & Goemans, H. E. (2011). The making of the territorial order: New borders and the emergence of interstate conflict. International Organization, 65(2), 275–309.

Carter, D. B., & Poast, P. (2017). Why Do States Build Walls? Political Economy, Security, and Border Stability. Journal of Conflict Resolution, 61(2), 239–270.

177 REFERENCES

Carter, D. B., & Signorino, C. S. (2010, jun). Back to the Future: Modeling Time Dependence in Binary Data. Political Analysis, 18(3), 271–292. Cederman, L.-E. (1997). Emergent actors in world politics: how states and nations develop and dissolve. Princeton, NJ: Princeton University Press. Cederman, L.-E., & Gleditsch, K. S. (2009). Introduction to Special Issue on “Disaggregating Civil War”. Journal of Conflict Resolution, 53(4), 487–495. Cederman, L.-E., Gleditsch, K. S., Salehyan, I., & Wucherpfennig, J. (2013). Transborder ethnic kin and civil war. International Organization, 67(2), 389–410. Cederman, L.-E., Rüegger, S., & Schvitz, G. (2019). Redemption through Rebellion: Linking Border Change to Ethnonationalist Conflict. Working Paper. Chapman, T., & Roeder, P. (2007). Partition as a Solution to Wars of National- ism: The Importance of Institutions. American Political Science Review, 101(4), 667–691. Claude, I. L. (1989). The balance of power revisited. Review of International Studies, 15(2), 77–85. Coleman, M. (2004). U.S. statecraft and the U.S.-Mexico border as secu- rity/economy nexus. Political Geography, 24(2), 185–209. Cook, S. J. (2014). Rare Events in International Relations: Modeling Heterogeneity and Interdependence with Sparse Data (Dissertati ed.). Cranmer, S., & Desmarais, B. (2016). A Critique of Dyadic Design. International Studies Quarterly, 60(2), 355–362. Cunningham, D. E., & Lemke, D. (2013). Combining Civil and Interstate Wars. International Organization, 67(03), 609–627. Curzon, L. o. K. (1907). Frontiers. The Romanes Lecture. Oxford: The Clarendon Press. Danielson, J., & Gesh, D. (2011). Global multi-resolution terrain elevation data 2010 (GMTED2010): U.S. Geological Survey Open-File Report 2011–1073. Open-File Report, 2010(1073), iV, 26. Darden, K. (2013). Resisting occupation: Mass schooling and the creation of durable national loyalties. Book Manuscript, 825–850. Day, A. J. (1982). Border and Territorial Disputes. Harlow: Longman Group Ltd. Deiwiks, C. (2011). Ethnofederalism: a slippery slope towards secessionist conflict? Zürich: Dissertation, ETH Zürich. Depetris-Chauvin, E. (2015). State history and contemporary conflict: Evidence from sub-saharan africa. Working Paper. Diamond, A., & J. Sekhon. (2013). Genetic Matching for Estimating Causal Effects: A General Multivariate Matching Method for Achieving Balance in Observational Studies. The Review of Economics and Statistics, 932–945.

178 REFERENCES

Diehl, P. F. (1985). Contiguity and military escalation in major power rivalries, 1816-1980. The Journal of Politics, 47(4), 1203–1211.

Dincecco, M., Fenske, J., & Onorato, M. (2019). Is Africa Different? Historical Conflict and State Development. Economic History of Developing Regions, 0(0), 1–42.

Dray, S Legendre, P., & Peres-Neto, P. (2006). Spatial modelling: a compre- hensive framework for principal coordinate analysis of neighbour matrices (PCNM). Ecological Modelling, 196, 434–493.

Eager, C. D. (2017). standardize: Tools for standardizing variables for re- gression in R. Retrieved 2019-06-02, from https://cran.r-project.org/ package=standardize

Elkins, Z., Guzman, A. T., & Simmons, B. (2006). Competing for capital: The diffusion of bilateral investment treaties, 1960–2000. International organization, 60(4), 811–846.

Englebert, P., Tarango, S., & Carter, M. (2002). Dismemberment and Suffocation: A Contribution to the Debate on African Boundaries. Comparative Political Studies, 35(10), 1093–1118. doi: 10.1177/001041402237944

Fall, J. (2010). Artificial states? On the enduring geographical myth of natural borders. Political Geography, 29, 140–147.

FAO/IIASA. (2011). Global Agro-Ecological Zones (GAEZ Version 3.0). Rome, Italy and Laxenberg, Austria: FAO and IIASA.

Fazal, T. M. (2011). State death: The politics and geography of conquest, occupation, and annexation. Princeton University Press.

Fazal, T. M., & Griffiths, R. D. (2014). Membership has its privileges: The changing benefits of statehood. International Studies Review, 16(1), 79–106. doi: 10.1111/misr.12099

Fearon, J. D. (1995, may). Rationalist explanations for war. International Or- ganization, 49(03), 379–414. Retrieved from http://www.journals.cambridge .org/abstract{_}S0020818300033324 doi: 10.1017/S0020818300033324

Fearon, J. D. (2004). Separatist Wars, Partition, and World Order. Security Studies, 13(4), 394–415. doi: 10.1080/09636410490945965

Fearon, J. D., & Laitin, D. (2014). Does contemporary armed conflict have “deep historical roots”? Working Paper.

Fischer, E. (1949). On Boundaries. World Politics, 1, 196–222.

Forsberg, E. (2013). Do Ethnic Dominoes Fall? Evaluating Domino Effects of Granting Territorial Concessions to Separatist Groups. International Studies Quarterly, 57(2), 329–340.

179 REFERENCES

Franzese, R., & Hays, J. (2007). Spatial Econometric Models of Cross-Sectional Interdependence in Political Science Panel and Time-Series-Cross-Section Data. Political Analysis, 14, 140–164.

Friedman, D. (1977). A Theory of the Size and Shape of Nations. Journal of political economy, 85(1), 59–77.

Gandrud, C. (2015). simPH: An R Package for Illustrating Estimates from Cox Proportional Hazard Models Including for Interactive and Nonlinear E ects. JSS Journal of Statistical Software, 65(3), 1–19.

Gavrilis, G. (2008). The Dynamics of Interstate Boundaries. Cambridge, Mas- sachusetts: Cambridge University Press.

Gellner, E. (1983). Nations and Nationalism. Ithaca: Cornell University Press. Retrieved from papers2://publication/uuid/DA2D5611-A284-4848 -989E-086A8A9C36E4

Germann, M., & Sambanis, N. (2019). Political Exclusion, Lost Autonomy, and Escalating Conflict over Self-Determination. Working Paper.

Gilpin, R. (1981). War and Change in International Politics. Cambridge, Mas- sachusetts: Cambridge University Press.

Gleditsch, K. (2004). A revised list of wars between and within independent states, 1816-2002. International Interactions, 30(3), 231–262.

Gleditsch, K. S. (2002). Expanded Trade and GDP Data. Journal of Conflict Resolution, 46(5), 712–724.

Gleditsch, K. S., Salehyan, I., & Schultz, K. (2008). Fighting at Home, Fighting Abroad: How Civil Wars Lead to International Disputes. Journal of Conflict Resolution, 52(4), 479–506. doi: 10.1177/0022002707313305

Gleditsch, K. S., & Ward, M. D. (1999). A revised list of independent states since the congress of vienna. International Interactions, 25(4), 393–413.

Gleditsch, K. S., & Ward, M. D. (2006). Diffusion and the International Context of Democratization. International Organization, 60(04), 911–933. doi: 10.1017/S0020818306060309

Global Runoff Data Centre. (2007). Major River Basins of the World. Retrieved 2019-05-29, from https://www.bafg.de/GRDC/EN/02{_}srvcs/22{_}gslrs/ 221{_}MRB/riverbasins.html?nn=201570

Goddard, S. E. (2006). Uncommon Ground: Indivisible Territory and the Politics of Legitimacy. International Organization, 60(January 2006), 35–68. doi: 10.1017/S0020818306060024

Goemans, H. E. (2006). Bounded communities: territoriality, territorial attach- ment, and conflict. In M. Kahler & B. Walter (Eds.), Territoriality and conflict in an era of globalization (pp. 25–61). Cambridge, Massachusetts: Cambridge University Press.

180 REFERENCES

Goemans, H. E., & Schultz, K. A. (2017). The Politics of Territorial Claims: A Geospatial Approach Applied to Africa. International Organization, 71(1), 31–64.

Goertz, G., & Diehl, P. (1992). Territorial Changes and International Conflict. London: Routledge.

Goertz, G., Diehl, P., & Balas, A. (2016). The Puzzle of Peace. Oxford: Oxford University Press.

Goettlich, K. (2018, mar). The rise of linear borders in world politics. European Journal of International Relations, 25(1), 203–228. Retrieved from https:// doi.org/10.1177/1354066118760991 doi: 10.1177/1354066118760991

Grambsch, P., & Thernau, T. (1994). Proportional hazards tests and diagnostics based on weighted residuals. Biometrika, 81, 515–526.

Griffiths, R. D. (2016). Age of Secession. Cambridge University Press.

Griffiths, R. D., & Butcher, C. (2013). Introducing the International System(s) Dataset (ISD), 1816–2011. International Interactions, 39, 748–768.

Gunitsky, S. (2014). From Shocks to Waves: Hegemonic Transitions and Democratization in the Twentieth Century. International Organization, 68(3), 561–597.

Hale, H. E. (2000). The Parade of Sovereignties: Testing Theories of Secession in the Soviet Setting. British Journal of Political Science, 30(1), 31–56. doi: 10.1017/S0007123400000028

Hartshorne, R. (1933). Geographic and Political Boundaries in Upper Silesia. Annals, Association of American Geographers, 23, 195–228.

Hartshorne, R. (1936). Suggestions on the Terminology of Political Boundaries. Association of American Geographers, 26, 56–57.

Hartshorne, R. (1937). The Polish Corridor. Journal of Geography, 36, 161–176.

Hassner, R. E. (2003). “To halve and to hold”: Conflicts over sacred space and the problem of indivisibility. Security Studies, 12(4), 1–33.

Hathaway, O., & Shapiro, S. (2017). The Internationalists: How a Radical Plan to Outlaw War Remade the World. New York: Simon & Schuster.

Hechter, M. (2000). Containing Nationalism. Oxford: Oxford University Press.

Hensel, P. (2000). Territory: theory and evidence on geography and conflict. In J. A. Vasquez (Ed.), What do we know about war? (pp. 57–84). Lanham: Rowman & Littlefield.

Hensel, P., Allison, M. E., & Khanani, a. (2009). Territorial Integrity Treaties and Armed Conflict over Territory. Conflict Management and Peace Science, 26(October 2006), 120–143. doi: 10.1177/0738894208101126

181 REFERENCES

Hensel, P., & Mitchell, S. (2005). Issue indivisibility and territorial claims. GeoJournal, 64(4), 275–285. doi: 10.1007/s10708-005-5803-3

Hensel, P., & Mitchell, S. (2017). From Territorial Claims to Identity Claims: The Issue Correlates of War (ICOW) Project. Conflict Management and Peace Science, 34.

Herbst, J. (1989). The creation and matintenance of national boundaries in Africa. International Organization, 43(4), 673–692.

Herbst, J. (2000). States and Power in Africa: Comparative Lessons in Authority and Control. Princeton, NJ: Princeton University Press.

Herz, J. H. (1957). Rise and demise of the territorial state. World Politics, 9(4), 473–493.

Hewitt, C., & Cheetham, T. (2000). Encyclopedia of Modern Separatist Movements. Santa Barbara: CLIO.

Hijmans, R. (2012). GADM database of global administrative areas (version 2.0). Retrieved from https://gadm.org/index.html

Hoff, P. D., & Ward, M. D. (2004). Modeling dependencies in international relations networks. Political Analysis, 12(2), 160–175.

Holdich, T. (1899). The use of practical geography illustrated by recent frontier operations. The Geographical Journal, 13(5), 465–477.

Holsti, K. J. (1991). Peace and War: Armed Conflicts and International Order 1648-1989. Cambridge, Massachusetts: Cambridge University Press.

Huth, P. K. (1988). Extended deterrence and the outbreak of war. American Political Science Review, 82(2), 423–433.

Huth, P. K. (1996). Standing Your Ground: Territorial Disputes and International Conflict. Ann Arbor: University of Michigan Press.

Huth, P. K., & Allee, T. L. (2002). The democratic peace and territorial conflict in the twentieth century. Cambridge: Cambridge University Press.

Huth, P. K., Croco, S. E., & Appel, B. J. (2013). Bringing law to the table: Legal claims, focal points, and the settlement of territorial disputes since 1945. American Journal of Political Science, 57(1), 90–103.

Johnson, D. D. P., & Toft, M. D. (2014). Grounds for war: The evolution of territorial conflict. International Security, 38(3), 7–38.

Jones, S. B. (1945). Boundary Making: A Handbook for Statesmen, Treaty Editors and Boundary Commissioners. Washington, DC.: Carnegie Endowment for International Peace.

Kacowicz, A. M. (1994). The problem of peaceful territorial change. Interna- tional Studies Quarterly, 38(2), 219–254.

182 REFERENCES

Kaufmann, C. (1996). Possible and Impossible Solutions to Ethnic Civil Wars. International Security, 72(5), 136–175.

Kier, E., & Mercer, J. (1996). Setting precedents in anarchy: military interven- tion and weapons of mass destruction. International Security, 20(4), 77–106.

King, G., Keohane, R., & Verba, S. (1994). Designing Social Inquiry: Scientific Inference in Qualitative Research. Princeton, NJ: Princeton University Press.

King, G., Tomz, M., & Wittenberg, J. (2000). Making the Most of Statistical Analyses: Improving Interpretation and Presentation. American Journal of Political Science, 44(2), 341–355.

King, G., & Zeng, L. (2001). Logistic regression in rare events data. Political Analysis, 9(2), 137–163.

Klein Goldewijk, K., Beusen, A., De Vos, M., & Van Drecht, G. (2011). The HYDE 3.1 Spatially Explicit Database of Human Induced Land Use Change Over the Past 12,000 Years. Global Ecology and Biogeography, 20(1), 73–86.

Kocs, S. A. (1995). Territorial disputes and interstate war, 1945-1987. The Journal of Politics, 57(1), 159–175.

Kratochwil, F. (1986). Of systems, boundaries, and territoriality: An inquiry into the formation of the state system. World Politics, 39(1), 27–52.

Kristof, L. (1959). The Nature of Frontiers and Boundaries. Annals of the Association of American Geographers, 49(3), 269–282.

Lake, D. A. (2003). International relations theory and internal conflict: insights from the interstices. International Studies Review, 5(4), 81–89.

Lake, D. A., & O’Mahony, A. (2004). The incredible shrinking state: Explaining change in the territorial size of countries. Journal of Conflict Resolution, 48(5), 699–722.

Lehner, B., & Döll, P. (2004). Global Lakes and Wetlands Database GLWD. GLWD Docu mentation.

Lehner, B., Verdin, K., & Jarvis, A. (2006). HydroSHEDS Technical Documentation. Washington, DC.: World Wildlife Fund US.

Lemke, D., & Reed, W. (2001). The relevance of politically relevant dyads. Journal of Conflict Resolution, 45(1), 126–144.

Levitsky, S., & Murillo, M. V. (2009). Variation in Institutional Strength. Annual Review of Political Science, 1(2), 115–133.

Lex Berman, M. (2001). China Historical Geographic Information System. Re- trieved 2019-05-17, from http://chgis.fas.harvard.edu/pages/intro/

Lex Berman, M. (2017). Geographic Data Resources for Japan. Retrieved 2019-06- 19, from https://dataverse.harvard.edu/dataverse/japan{_}hgis?

183 REFERENCES

Licht, A. (2011). Change Comes with Time: Substantive Interpretation of Nonproportional Hazards in Event History Analysis. Political Analysis, 19(2), 227–243.

Lost Autonomy, Nationalism and Separatism. (2014). Comparative Political Studies, 48(1), 3–34. Retrieved from http://cps.sagepub.com/cgi/doi/10 .1177/0010414013516927 doi: 10.1177/0010414013516927

Lujala, P., Rød, J., & Thieme, N. (2007). Fighting over Oil: Introducing A New Dataset. Conflict Management and Peace Science, 24(3).

Lyde, L. W. (1915). Some Frontiers of Tomorrow: An Aspiration for Europe. London: A. & C. Black.

Maoz, Z., & Russett, B. (1993). Normative and Structural Causes of Democratic Peace. The American Political Science Review, 87(3), 624–638.

Marshall, M. G., Gurr, T. R., & Jaggers, K. (2014). Polity IV project: Political regime characteristics and transitions, 1800–2013. Center for Systemic Peace.

Mearsheimer, J. J. (1994). The false promise of international institutions. International security, 19(3), 5–49.

Michalopoulos, S., & Papaioannou, E. (2016). The Long-Run Effects of The Scramble for Africa. American Economic Review, 106(7), 1802–1848.

Minahan, J. (1996). Nations without States: A Historical Dictionary of Contempo- rary National Movements. London: Greenwood Press.

Minahan, J. (2002). Encyclopedia of the Stateless Nations: Ethnic and National Groups Around the World A-Z. ABC-CLIO.

Minghi, J. (1963). Boundary studies in political geography. Annals of the Association of American Geographers, 53(3), 407–428.

Minhas, S., Hoff, P., & Ward, M. (2016). A new approach to analyzing coevolv- ing longitudinal networks in international relations. Journal of Peace Research, 53(3), 491–505.

Moodie, A. E. (1943). The Italo-Yugoslav Boundary. Geographic Journal, 101, 49–65.

Muir. (1975). Modern Political Geography. London: MacMillan Press.

Murakami, D., & Griffith, D. (2018). Eigenvector Spatial Filtering for Large Data Sets: Fixed and Random Effects Approaches. Geographical Analysis, 0(0), 1–27.

Murphy, A. B. (1990). Historical Justifications for Territorial Claims. Annals of the Association of American Geographers, 80(4), 531–548.

Mylonas, H., & Shelef, N. G. (2014). Which land is our land? Domestic politics and change in the territorial claims of stateless nationalist movements. Security Studies, 23(4), 754–786.

184 REFERENCES

Mylonas, H., & Shelef, N. G. (2017). Methodological challenges in the study of stateless nationalist territorial claims. Territory, Politics, Governance, 5(2), 145–157.

Natural Earth. (2015). Natural Earth Data: Rivers and Lake Centerlines. Retrieved 2019-05-29, from http://www.naturalearthdata.com/downloads/ 10m-physical-vectors/10m-rivers-lake-centerlines/

Neumayer, E., & Plümper, T. (2010a). Making spatial analysis operational: Commands for generating spatial-effect variables in monadic and dyadic data. The Stata Journal, 10(4), 585–605.

Neumayer, E., & Plümper, T. (2010b). Spatial Effects in Dyadic Data. Interna- tional Organization, 64(1), 145–166.

Newman, D. (1999). Real spaces, symbolic spaces: Interrelated notions of territory in the Arab-Israeli conflict. A road map to war: Territorial dimensions of international conflict, 3–34.

Newman, D. (2002). From ‘moribund backwater’to ‘thriving into the next century’. The razor’s edge: International boundaries and political geography. London: Kluwer Law.

Newman, D. (2006). Borders and Bordering. Towards an Interdisciplinary Dialogue. European Journal of Social Theory, 9(2), 171–186.

Newman, D., & Paasi, A. (1998). Fences and neighbours in the postmodern world: boundary narratives in political geography. Progress in human geography, 22(2), 186–207.

Nicol, H. (2005). ‘Resiliency or change? The contemporary Canada-US border’. Geopolitics, 10(4), 767–790.

North, D. (1990). Institutions, Institutional Change and Economic Growth. Cam- bridge: Cambridge University Press.

Nuessli, C. (2010). Euratlas Historical Atlas and Gazetteer of Europe. Retrieved 2019-05-17, from https://www.euratlas.com/

O’Brien, R. (1992). Global financial integration: the end of geography. Royal Institute of International Affairs.

Ohmae, K. (1991). The borderless world: Power and strategy in the interlinked economy. (Tech. Rep.).

O’Leary, B. (2001). Right-sizing the state: the politics of moving borders. In B. O’Leary (Ed.), . Oxford: Oxford University Press.

Openshaw. (1984). The Modifiable Areal Unit Problem (Geo Books, Ed.). Nor- wich.

Osiander, A. (2001). Sovereignty, international relations, and the Westphalian myth. International organization, 55(2), 251–287.

185 REFERENCES

Paasi, A. (1996). Territories, boundaries, and consciousness: The changing geogra- phies of the Finnish-Russian boundary (Vol. 1). Wiley.

Paasi, A. (2016). A border theory: an unattainable dream or a realistic aim for border scholars? In The routledge research companion to border studies (pp. 33–54). London: Routledge.

Peterson, J. E. (2020). Historical Dictionary of Saudi Arabia. Rowman & Littlefield Publishers.

Pierson, P. (2000). Increasing Returns, Path Dependence and the Study of Politics. American Political Science Review, 94(2), 251–267.

Poast, P. (2010). (Mis) using dyadic data to analyze multilateral events. Political Analysis, 18(4), 403–425.

Poast, P. (2016). Dyads Are Dead, Long Live Dyads! The Limits of Dyadic Designs in International Relations Research. International Studies Quarterly, 60(2), 369–374.

Pope, D. G., Pope, J. C., & Sydnor, J. R. (2015). Focal points and bargaining in housing markets. Games and Economic Behavior, 93, 89–107.

Pounds, N. J. (1951). The origin of the idea of natural frontiers in France. Annals of the Association of American Geographers, 41(2), 146–157.

Pounds, N. J. (1954). France and “Les Limites Naturelles” from the Seventeenth to the Twentieth Centuries. Annals of the Association of American Geographers, 44(1), 51–62.

Prescott, V. (1978). Boundaries and frontiers. London: Croom Helm.

Prescott, V., & Triggs, G. D. (2008). International frontiers and boundaries: Law, politics and geography. Brill.

Rankin, K. J., & Schofield, R. N. (2004). The troubled historiography of classical boundary terminology. In Revised version of a paper presented at the 30th congress of the international geographical union, glasgow, scotland, 15-20 august 2004. University College Dublin. Institute for British-Irish Studies.

Ratzel, F. (1897). Politische Geographie. München: Oldenbourg.

Reed, F. (2008). Centennia Historical Atlas. Retrieved 2019-05-17, from http:// www.historicalatlas.com

Richardson, L. F. (1960). Statistics of Deadly Quarrels (Q. Wright & C. Lienau, Eds.). Pacific Grove: The Boxwood Press.

Roeder, P. G. (2007). Where Nation-States Come From. Princeton, NJ: Princeton University Press.

Rosenbaum, P. R. (2010). Design of Observational Studies. New York: Springer.

186 REFERENCES

Roth, C. F. (2015). Let’s Split! A Complete Guide to Separatist Movements and Aspirant Nations, from Abkhazia to Zanzibar. Sacramento, CA: Litwin Books.

Ruggie, J. G. (1993). Territoriality and beyond: problematizing modernity in international relations. International organization, 47(1), 139–174.

Russett, B., Singer, D., & Small, M. (1968). National Political Units in the Twentieth Century: A Standardized List. American Political Science Association, 62(3), 932–951.

Sack, R. D. (1986). Human Territoriality. Cambridge: Cambridge University Press.

Sahlins, P. (1989). Boundaries: the making of France and Spain in the Pyrenees. Berkeley: University of California Press.

Saideman, S. M. (1997). Explaining the International Relations of Secession- ist Conflicts: Vulnerability Versus Ethnic Ties (Vol. 51) (No. 4). doi: 10.1162/ 002081897550500

Sambanis, N., Germann, M., & Schädel, A. (2018). SDM: A New Data Set on Self-determination Movements with an Application to the Reputational Theory of Conflict. Journal of Conflict Resolution, 62(3), 656–686.

Schelling, T. C. (1980). The Strategy of Conflict. Cambridge, Massachussets: Harvard University Press.

Schofield, R. (2016). Territorial foundations of the Gulf states. Routledge.

Schultz, K. A. (2014). What’s in a Claim? De Jure Versus de Facto Borders in Interstate Territorial Disputes. Journal of Conflict Resolution, 58(6), 1059–1084.

Schultz, K. A. (2015a). Borders, Conflict, and Trade. Annual Review of Political Science, 18, 125–145.

Schultz, K. A. (2015b). Mapping Interstate Territorial Conflict: A New Data Set and Applications. Journal of Conflict Resolution, 61(7), 1565–1590.

Schultz, K. A., & Goemans, H. E. (2019). Aims, claims, and the bargaining model of war. International Theory, 1–31.

Schutte, S., & Donnay, K. (2014). Matched Wake Analysis: Finding Causal Relationships in Spatiotemporal Event Data. Political Geography, 41, 1–10.

Schvitz, G., Rüegger, S., Girardin, L., Cederman, L.-E., Weidmann, N., & Gleditsch, K. (2019). Mapping The International System, 1886-2017: The Cshapes 2.0 Dataset. Working Paper. Retrieved from https://papers.ssrn .com/sol3/papers.cfm?abstract{_}id=3610680

Shaw, M. N. (1997). The heritage of states: The principle of uti possidetis juris today. The British year book of international law, 67(1), 75.

187 REFERENCES

Shelef, N. G. (2015). Unequal Ground: Homelands and Conflict. International Organization(June), 1–31. Retrieved from http://www.journals.cambridge .org/abstract{_}S0020818315000193 doi: 10.1017/S0020818315000193

Signorino, C. S. (1999). Strategic Interaction and the Statistical Analysis of International Conflict. American Political Science Review, 93(2), 279–297.

Simmons, B. (2005). Rules over Real Estate. Journal of Conflict Resolution, 49(1), 823–848.

Simmons, B. (2019). Border Rules. International Studies Review, 21(2), 256–283.

Singer, J. D. (1987). Reconstructing the Correlates of War Dataset on Material Capabilities of States. International Interactions, 14, 115–132.

Siverson, R., & Starr, H. (1991). The diffusion of the war: a study in opportunity and willingness. University of Michigan Press.

Smith, W. D. (1980). Friedrich Ratzel and the Origins of Lebensraum. German Studies Review, 3(1), 51–68.

Sovey, A., & Green, D. (2011). Instrumental Variables Estimation in Political Science: A Readers’ Guide. Americal Journal of Political Science, 55(1), 188–200.

Starr, H. (1978). “Opportunity” and “willingness’ as ordering concepts in the study of war. International Interactions, 4(4), 363–387.

Starr, H. (2013). On Geopolitics: Space, Place and International Relations. Boulder: Paradigm.

Stinnett, D., Tir, J., Diehl, P. F., Schafer, P., & Gochman, C. (2002). The correlates of war (cow) project direct contiguity data, version 3.0. Conflict Management and Peace Science, 19(2), 59–67.

Tiefelsdorf, M., & Griffith, D. (2007). Semiparametric Filtering of Spatial Autocorrelation: The Eigenvector Approach. Environment and Planning, 39(5), 1193–1221.

Tilly, C. (1985). War Making and State Making as Organized Crime. Princeton, NJ: Princeton University Press.

Tilly, C. (1990). Coercion, Capital, and European States, AD 990-1992. Cambridge, Massachusetts: Blackwell Publishing.

Tir, J. (2003). Averting Armed International Conflicts Through State-to-State Territorial Transfers. The Journal of Politics, 65(4), 1235–1257.

Tir, J. (2006). Redrawing the map to promote peace: territorial dispute management via territorial changes. Lexington Books.

Tir, J., Diehl, P., & Goertz, G. (1998). Territorial Changes, 1816-1996: Procedures and Data. Conflict Management and Peace Science, 16(1), 89–97.

188 REFERENCES

Toft, M. D. (2003). The Geography of Ethnic Violence: Identity, Interests and the Indivisibility of Territory. Princeton, NJ: Princeton University Press.

Touval, S. (1972). The boundary politics of independent africa. Harvard University Press.

USGS. (2005). Mineral Resources Data System. Reston, Virginia: U.S. Geological Survey. Retrieved from https://mrdata.usgs.gov/mrds/

Van Houtum, H. (2005). The Geopolitics of Borders and Boundaries. Geopoli- tics,, 10(4), 672–679.

Vasquez, J. A. (1993). The War Puzzle. Cambridge: Cambridge University Press.

Vasquez, J. A. (1995). Why Do Neighbors Fight? Proximity, Interaction or Territoriality. Journal of Peace Research, 32(3), 227–293.

Vasquez, J. A., & Henehan, M. T. (2001). Territorial Disputes and The Probabil- ity of War. Journal of Peace Research, 38(2), 123–138.

Vattel, E. (1758). Le droit du gens ou principes de la loi naturelle appiqués a la conduite, et aux affeires des nations et des souverains. London.

Vega, S. H., & Elhorst, J. P. (2015). THE SLX MODEL. Journal of Regional Science, 55(3), 339–363.

Vivien de Saint-Martin, L., & Schrader, F. (1937). Atlas Universel de Geogra- phie. In Atlas universel de geographie. Paris: Librairie Hachette.

Vogt, M., Bormann, N. C., Rüegger, S., Cederman, L.-E., Hunziker, P., & Girardin, L. (2014). Integrating Data on Ethnicity, Geography, and Con- flict: The Ethnic Power Relations Dataset Family. Journal of Peace Research, 59(7), 1–30. Retrieved from http://www.icr.ethz.ch/icr/publications/ vogt2015integrating.pdf doi: 10.1177/0022002715591215

Wakefield, J. (2003). Sensitivity analyses for ecological regression. Biometrics, 59, 9–7.

Walker, P. (2011, jun). The World’s Most Dangerous Borders: Thirteen Places you do not Want to be Struck at. Foreign Affairs. Retrieved from https:// foreignpolicy.com/2011/06/24/the-worlds-most-dangerous-borders/

Walter, B. F. (2009). Reputation and Civil War: Why Separatist Conflicts are So Violent. Cambridge: Cambridge University Press.

Ward, M. D., & Gleditsch, K. S. (2007). An introduction to spatial regression models in the social sciences. Manuscript at http://www. faculty. washington. edu/mdw . . . , 1–77.

Weede, E. (1973). Nation-Environment Relations as Determinants of Hostilities Among Nations. Peace Science Society (International) Papers, 20, 67–90.

189 REFERENCES

Weidmann, N. B., Kuse, D., & Gleditsch, K. S. (2010). The Geography of the International System: The CShapes Dataset. International Interactions, 36(1), 86–106. doi: 10.1080/03050620903554614

Weidmann, N. B., Rod, J. K., & Cederman, L.-E. (2010). Representing ethnic groups in space: A new dataset. Journal of Peace Research, 47(4), 491–499. doi: 10.1177/0022343310368352

White, G. W. (2000). Nationalism and Territory. Constructing Group Identity in Southeastern Europe. Lanham: Rowman & Littlefield.

White, G. W. (2004). Nation, State and Territory. Origins, Evolutions and Relation- ships. Lanham: Rowman & Littlefield.

Wimmer, A., & Min, B. (2006). From Empire to Nation-State: Explaining Wars in the Modern World, 1816-2001. American Sociological Review, 71(6), 867–897. doi: 10.1177/000312240607100601

Wolf, N. (2005). Path dependent border effects: the case of Poland’s reunifica- tion (1918–1939). Explorations in Economic History, 42(3), 414–438.

Wooldridge, J. M. (2015). Introductory econometrics: A modern approach. Nelson Education.

Wyckoff, T. (1980). Standardized List of National Political Units in the 20th Century: The Russett-Singer-Small-List of 1968 Updated. International Journal of Social Science, 32(4), 833–846.

Zacher, M. W. (2001). The Territorial Integrity Norm: International Boundaries and the Use of Force (Vol. 55) (No. 2). doi: 10.1162/00208180151140568

Zartman, I. W. (1966). International Relations in the New Africa. Englewood Cliffs N.J.: Prentice-Hall.

Zhukov, Y. M., & Stewart, B. (2013). Choosing Your Neighbors: Networks of Diffusion in International Relations. International Studies Quarterly, 57, 271–287.

190