Pragmaa's posts

The Long-Run E

ffects of the Scramble for Africa∗

Stelios Michalopoulos

†, Elias Papaioannou‡

January 2011

Abstract

We examine the economic consequences of the partitioning of Africa among European

powers in the late 19th century; a process historically known as the scramble for Africa.

First, using information on the spatial distribution of African ethnicities before colonization

we establish that border drawing was largely arbitrary. Apart from the land mass and water

area of an ethnicity’s historical homeland, no other geographic, ecological, historical, and

ethnic-speci

borders. Second, employing data on the location of civil con

show that compared to ethnicities that have not been impacted by the border design,

partitioned ethnic groups have su

civil wars. Third, we

density at night- is systematically lower in the historical homeland of partitioned ethnicities.

These results are robust to a rich set of controls at a

and ethnic-family

of the scramble for Africa on comparative regional development.

fic trait predicts which ethnic groups have been partitioned by the nationalflicts after independence weffered significantly more, longer, and more devastatingfind that economic development —as reflected by satellite data on lightfine level and the inclusion of countryfixed-effects. Our regressions thus identify a sizable causal negative effect

Keywords:

Africa, Borders, Ethnicities, Conflict, Development.

JEL classi

fication Numbers: O10, O40, O43, N17, D74, Z10.

∗

Denver for useful comments. All errors are our sole responsibility.

We thank Gregorios Siourounis, Heiwai Tang and participants at the Econometric Society Meetings in

†

michalopoulos@tufts.edu. The author was a Deutsche Bank Member at the Institute for Advanced Study,

Princeton and gratefully acknowledges their hospitality.

Tufts University, Braker Hall, 8 Upper Campus Rd., Medford, MA 02155. E-mail: stelios.

‡

Department, Littauer 324, 1805 Cambridge Street, Cambridge, MA 02138, USA. E-mail:

elias.papaioannou@dartmouth.edu.

Dartmouth College and CEPR and Harvard University. Address: Harvard University, Economics

0

1 Introduction

The predominant explanations on the deep roots of contemporary African underdevelopment

are centered around the in

(2001, 2002, 2005)), but also in the centuries before colonization when close to

slaves were exported from Africa (Nunn (2008)). Yet in the period between the ending of the

slave trades and the commencement of the colonial period, another major event took place in

the European capitals that according to the African historiography had malicious long-lasting

consequences. The "scramble for Africa" starts with the Berlin Conference of

is completed by the turn of the 20th century. In this brief period, Europeans partitioned Africa

into spheres of in

in European capitals at a time when Europeans had barely settled in Africa and had little

knowledge of the geography and ethnic composition of the areas whose borders were designing.

These borders endured after the African independence in the 1960’s leading to the partitioning

of numerous ethnic groups across the newly created African states.

research in African history (e.g. Asiwaju (1985); Dowden (2008)) argues that the main impact

of Europeans’ in

Partitioning, the argument goes, led to ethnic struggles, patronage politics, and spurred civil

con

examines the impact of ethnic partitioning.

fluence of Europeans during the colonial period (Acemoglu et al.20 million1884 − 1885 andfluence, protectorates, colonies, and free-trade-areas. The borders designed1 A considerable body offluence in Africa was not colonization per se, but the improper border design.flict, leading to poverty and underdevelopment. Yet there is very little work that formally2

This paper is a

Africa on civil con

the current national borders on George Peter Murdock’s Ethnolinguistic Atlas (1959) that

portrays the spatial distribution of ethnicities before colonization (Figure

out of a total of

falls into more than one contemporary state (Figure

that colonial borders were arbitrarily drawn, we start our empirical analysis by establishing

formally their randomness. We estimate probabilistic models trying to identify signi

factors that predict whether a group was partitioned. With the sole exceptions of the size of

first-step to empirically assess the long-run effects of the scramble forflict and economic development. We identify partitioned groups projecting1). We find that834 ethnicities, for 231 ethnic groups at least 10% of their historical homeland1).3 While there is little disagreementficant

1

et al. (2002) estimates that the population of partitioned ethnic groups is on average more than

population. Likewise Alesina et al. (2011) estimate that in many African countries the percent of the population

that belongs to a partitioned group exceeds

Burundi (

Asiwaju (1985) identifies 177 partitioned ethnic groups that span all African borders. Moreover, Englebert40% of the total80% (e.g. Guinea-Bissau (80%); Guinea (884%); Eritrea (83%);974%); Malawi (89%); Senegal (91%); Rwanda (100%); Zimbabwe (99%)).

2

relationship to our work we discuss below.

Notable exceptions are the cross-country studies of Alesina et al. (2011) and Englebert et al. (2002) whose

3

have been a

to have been partitioned across the national borders.

When we consider all split ethnic groups irrespective of the degree of partitioning, we find that 358 groupsffected by the border design. When we use a more restrictive threshold of 20%, we find 164 ethnicities

1

the historical homeland and water bodies, we are unable to detect any signi

between partitioned and non-partitioned ethnicities with respect to other geographical features,

the disease environment, natural resources and the magnitude of slave raids. We further show

that there are no major di

precolonial ethnic-speci

ficant differencesfferences between split and non-partitioned groups across manyfic institutional, cultural, and economic features (Murdock (1967)).

Ü

Ethnic Homelands

and National Borders

National Boundaries

Non-Border Groups

Border Groups

Partitioned Groups >10%

Figure

1 Figure 1

We then employ the scramble for Africa as a quasi-natural experiment to assess the

impact of partitioning on development and civil con

main channel of in

all civil wars in Africa in the post-independence period (

is concentrated in the historical homelands of partitioned ethnicities. We also

areas populated by ethnic groups only modestly a

experience more con

signi

and statistical signi

so as to control for national and broad ethnic characteristics respectively. Our most

conservative estimates suggest that civil con

duration, is approximately

examine the e

income at our ethnic-level of analysis, we follow Henderson et al. (2009) and proxy regional

development with satellite image data on light density at night. The cross-ethnicity estimates

suggest that development in the historical homeland of partitioned ethnic groups is lower by

2

almost a half (

flict, as this has been theorized to be thefluence. Using regional data on the incidence, duration, and total casualties of1970−2005), we find that civil conflictfind that borderffected by the artificial border design alsoflict, though this effect is of smaller magnitude and not always statisticallyficant. The positive effect of partitioning on all aspects of civil war retains its economicficance when we condition on country fixed-effects and ethnic-family fixedeffects,flict intensity, as reflected in war casualties and35% higher in areas where partitioned ethnicities reside. We thenffect of partitioning on regional development. Due to data unavailability on−46%), compared to non-partitioned ethnic areas.

Historical Background

The scramble for Africa started in the late 19th century and was fully completed by the turn

of the 20th century.

that Otto von Bismarck organized in Berlin from November

Berlin conference mainly discussed the boundaries of Central Africa (the Congo Free State),

it came to symbolize the partitioning, because it laid down the principles that would be used

among Europeans to divide the continent.

among Europeans over the African territories (as the memories of the European wars of the

18th-19th century were still alive). In light of this, Europeans divided territories and drew

borders in maps, without taking into account local geographic conditions and/or the ethnic

composition. African leaders were not invited and had no say. Europeans were in such a

rush that they didn’t wait for the new information arriving from explorers, geographers, and

missionaries.

4 The event that stands for the partitioning of Africa is the conference1884 till February 1885. While the5 The key consideration was to prevent conflict6

The anecdotal evidence suggests that the scramble for the continent was arbitrary. As the

British prime minister at the time Lord Salisbury put it in a famous sally, "

in drawing lines upon maps where no white man’s feet have ever tord; we have been giving away

mountains and rivers and lakes to each other, only hindered by the small impediment that we

never knew exactly where the mountains and rivers and lakes were

among African scholars on the arti

that the "

of the marking of African boundaries

we have been engaged." There is little disagreementficial border design. For example, Asiwaju (1985) arguesstudy of European archives supports the accidental rather than a conspiratorial theory."7

4

Western Africa and sign bilateral agreements assigning spheres of in

the Franco-Spanish partition of Morocco and the annexation of Libya by Italy in

The scramble for Africa begins in 1860s-1870s when the French and the British start exploring systematicallyfluence. The scramble was completed with1912.

5

according to which a power claiming the coast had also a right to its interior. Yet, the applicability of this

principle became problematic, as it was not clear what exactly constitutes the hinterland. For example, at some

point France demanded Nigeria claiming that it was the hinterland of Algeria. Second, the principle of e

possession required that Europeans need to base their claim on treaties with local tribe leaders. Yet, it was hard

to assign zones of in

a piece of paper with a lot of Negro crosses at the bottom

doctrine required that European powers exert signi

the insistence of the British this principle was soon diminished to apply mostly in the coastline.

The three major principles that emerged from the Berlin Conference were: First, the hinterland doctrine,ffectivefluence based on such treaties, because as Bismarck pointed out "it was too easy to come by" (Wesseling (1996)). Third, the effective occupationficant control of the territory they were claiming. Yet, with

6

history of Africa, was essentially a European a

were, if they mattered at all, completely marginal to the basic economic, strategic, and political interests of the

negotiating European powers

For example, Asiwaju (1985) notes that "the Berlin conference, despite its importance for the subsequentffair: there was no African representation, and African concerns".

7

might not be able to describe them accurately, the French preferred to allocate territory along some natural feature

like a watershed. Yet, the problem was that the Europeans had a rather imperfect idea of where the water streams

exactly where. A prominent example is the Anglo-German agreement on the Nigeria-Cameroon boundary that

Likewise Hargreaves (1985) writes "rather than attempting to follow the boundaries of states whose rulers

3

There are many reasons explaining the arbitrary border design: First, at the time Europeans

had limited knowledge of local geographic conditions and were unwilling to wait. Second,

Europeans were not drawing borders of prospective states or in many cases even colonies. For

example, the Democratic Republic of Congo that corresponds to Congo Free State is so large

simply because it was meant to be a free trade area rather than King Leopold’s property or

Belgian colony, not to say an independent state. Third, there was a constant imperialist back

and forth with European powers swapping pieces of land with no idea what they were worth of.

An illustrative case is the annexation of Katanga by the Congo Free State which turned out to

be the richest and most important province. Leopold demanded and eventually got Katanga

in exchange for the Niari-Kwilu area that the French insisted of having themselves (Wesseling

(1996)). Fourth, while in most cases the treaties indicated that the exact boundaries would

be set and demarcated by special commissions, demarcation was poor and the commissions

did not alter much. Fifth, Europeans were not willing to sacri

to war for any part of Africa.

local administrators to redraw the border because it did not coincide with a physical boundary

or because an ethnic group was split. Sixth, there was an implicit agreement between Europeans

that ethnicities could freely move across colonial borders. Asiwaju (1985) cites the Ketu

king, saying that "

the English and the French, not the Yoruba

The other major event in African history, the wave of independence, was also rapid.

The independence of Northern African countries in the 1950’s was followed by Ghana’s and

Guinea’ independence in

had gained independence. While at the time, many proposed changing the colonial borders,

African leaders and leaving Europeans avoided this issue. The leaders of African independence

believed that nation building would sideline ethnic divisions; moreover building new states

and national institutions seemed more important than massive border realignment. Likewise

Europeans’ main objective was to maintain their special rights and corporate deals with their

former colonies, and as such, they were reluctant to open the border issue. Thus almost all

African countries accepted the colonial borders when signing the Charter of the Organization

of African Union in

mentioned that nations would allow their citizens moving across the border, so as to mitigate

fice their commitment to not go8 In many cases London and Paris turned down requests fromwe regard the boundary (between Benin-Dahomey and Nigeria) separating."1957 and in 1958, respectively. By the end of 1966, 40 countries1964.9 Moreover, the treaty for the formation of African Union explicitly

was supposed to be Rio del Rey. The latter proved to be an estuary receiving several small streams.

"

8

the eyes of British imperialists such as Cecil Rhodes and Harry Johnston. When they approached the British

government on the subject, it stuck to its guns. Anderson let them know that Leopold’s map had been recognized

in 1885 and that his territory unmistakably comprised the mining region of Katanga. What was done, was done

For example Wesseling (1996) writes "in later years, Katanga was to become a most desirable possession in."

9

objections on their boundary that splits the Ewe.

Only Somalia and Morocco did not accept formally the colonial borders. Ghana and Togo raised also

4

the e

ffect of ethnic partitioning.

Channels

The African historiography has put forward many explanations on how the partitioning of

ethnicities and the creation of arti

First, in many instances partitioning has generated irredentist demands, as ethnicities

that are minority groups in a country want to unify with their peers across the border. For

example, Somali tribes were split between three di

got a slice. As a result, nowadays besides Somalia a large portion of Somalis occupy Northern

Kenya, the Ogaden region in Ethiopia, as well as Eritrea and Djibouti. Three long-lasting wars

in our sample have (partly at least) been driven by the desire of Somalis in Ethiopia, Djibouti,

and Kenya to become part of Somalia.

Second, partitioned ethnicities may

illustrative example is the recurring civil con

where the partitioned ethnic groups Diola and the Malinke reside. As Gambia e

Senegal into a Northern and Southern part, the Southern province of Casamance is disconnected

from the central government in Dakar and has demanded independence.

ficial states has contributed to underdevelopment.fferent European colonies, while Ethiopia alsofight to gain independence or obtain autonomy.10 Anflict in the Casamance region in Southern Senegal,ffectively splits11

Third, partitioned ethnicities have reacted to their marginalization by participating in

coups and rebellions to overthrow or capture the government. For example, the Ewe in Togo

helped Flt.-Lt. Jerry Rawlings (half Ewe) in his coup in

government in Ghana. This escalated ethnic tensions between the Ewe, the Ashanti and the

Akan in Ghana leading to civil warfare in the subsequent years.

Fourth, African borders are poorly demarcated and not well delineated due to the imprecise

colonial treaties. This has resulted in border disputes, especially when such poorly

demarcated borders cause the partitioning of ethnic groups.

Burkina Faso over the Agacher Strip, where the Bobo reside, illustrates the problems caused by

poor demarcation. The escalation of minor con

in a fully blown war in

1979 and 1981 to overthrow the12 The conflict between Mali andflicts that started after independence resulted1985.13 Imprecise colonial treaties seem to have contributed to conflict

10

examples of secessions that have resulted in de facto autonomous and independently governed areas include the

Western Sahara and the Somaliland (former British Somaliland).

Wimmer et al. (2009) estimate that around 20% of all civil wars in Africa have a secessionist demand. Other

11

by using the much lengthier overland route, The partition was undertaken (between the French and the British)

without any consideration for cultural ties, economic viability or regional coherence.

Renner (1985) writes "Senegal itself became truncated, and could only be linked by traversing Gambia or"

12

Court of Justice since

worldwide are in Africa

For example Englebert et al. (2002) write "of all the territorial disputes brought before the International1960, 57% were African, while only 33% ( 104 out of 315) of all bilateral boundaries."

13

the

Eventually this dispute was settled in the International Court of Justice in the end of 1986. The court split3 000  of disputed territory almost equally between the two countries.

5

in Somalia (Higham (1985)), while the ambiguity of the tripartite treaty between Britain, Italy

and Ethiopia of

Fifth, Africa is characterized by patronage politics where dominant ethnic groups discriminate

against minority groups (see i Miguel (2007) for a theoretical exposition and Wimmer

et al. (2009) for empirical evidence). In many cases the central government tries to su

partitioned ethnicities, for example by seizing property and imposing higher taxation in the

activities of speci

either to support their peers or to prevent migration and refugee

Alur-land exempli

Protectorate of Uganda during the late phase of the scramble for Africa (

the regime of Mobutu Sese Seko initiated the subjugation of many minority groups in Congo,

a large portion of the Alur in Congo moved to their historical homeland in Uganda. This in

turn generated opposition from the dominant Buganda group leading to civil con

Sixth, due to their ethnic contacts across the border, partitioned ethnicities may engage

in smuggling and other criminal activities. For example, in his analysis of the Anglo-French

partitioning of the Sultanate of the Mandara in the Nigeria-Cameroon boundary, Barkindo

(1985) writes that "

the border. The fact that the border divided people of the same family and settlements made

it di

allowed the Hausa to arbitrage price caps and other distortionary policies in Niger and Nigeria.

Seventh, border arti

heterogeneous ethnic groups were forced to be part of the same usually large country. Many

African scholars emphasize that civil con

hard for the state to broadcast political power and prevent secessionist movements among

diverse ethnicities (e.g. Herbst (2000)). Indeed most long-lasting civil wars have taken place

in the largest African countries, namely the Democratic Republic of Congo, Chad, Niger and

Angola with Sudan being the most illustrative example. The ethnically, religiously, and racially

distinct tribes of the North (that are part of the Nilo-Saharan families) and the South (that

belong to the Afro-Asiatic family) resulted in a three-decade long civil war and an ongoing

referendum for the independence of Southern Sudan.

Eighth, partitioning may lead to armed warfare by interacting with natural resources.

If the historical homeland of a partitioned ethnic group is rich in natural resources then the

bene

oppressive. For example, armed con

of Angola by a narrow strip of territory belonging to the Democratic Republic of the Congo

6

is driven by the interaction between the arti

partitioning of the Bakongo people.

1902 has also played a role in the Eritrea-Ethiopia war.ffocatefic groups (Bates (1981)). As a result, the neighboring country intervenesflows. The conflict in thefies the case. The Alur were split between the Belgian Congo and the British1910 − 1914). Whenflict.the most serious problem was the increase in crime and disputes acrossfficult to check crime and control smuggling." Collins (1985) also describes how smugglingficiality (though not partitioning itself) spurred conflict, becauseflict is more pervasive in large countries where it isfit of secession increases; moreover in this case the central government is more likely to beflict in the Cabinda enclave that is separated from the restficial border design, the vast oil fields, and the

Related Literature

Our paper contributes to two main strands of the literature. First, our work relates to studies

that aim to uncover the deep roots of African -and more broadly global- development. This

literature has mainly focused on the impact of colonization mainly via early institutions (see

for a review). Glaeser et al. (2004) and Easterly and Levine (2009) also stress the key role of

colonization for long-run development, but emphasize the human capital channel. In contrast

to this body of work, Gennaioli and Rainer (2006, 2007), Michalopoulos and Papaioannou

(2010) and Nunn (2008) focus on the pre-colonial period stressing the role of pre-colonial ethnic

institutions and the slave trades between the 15th-19th century respectively (see also Nunn and

Puga (2011) and Nunn andWantchekon (2011) on the latter). Of relevance to our work are also

studies linking ethnic fragmentation/polarization with development (e.g. Easterly and Levine

(1997); Alesina et al. (2003); Montalvo and Reynal-Querol (2005a); Alesina and Zhuravskaya

(2011); see Alesina and Ferrara (2005) for a review).

Our paper contributes to this body of research, by emphasizing a somewhat neglected

aspect of colonization; the drawing of political boundaries in the end of the 19th century that

in the eve of African independence partitioned numerous ethnicities across the newly created

African states. Thus our work is mostly related to Alesina et al. (2011), who show that "arti

states" de

the population resides in more than one country, perform economically worse compared to

countries with more organic (squiggly) borders. We focus on Africa, as the random design

of colonial borders that endured after African independence allows us to identify the causal

e

to under-development, we uncover the detrimental role of the border design in fomenting

civil con

disease environment, natural resources, and other factors that a vast literature emphasizes as

key determinants of economic development. Furthermore, we estimate country-

speci

the country-level (such as institutional quality, foreign aid, national policies, etc.).

condition on ethnic-family

in pre-colonial institutions, cultural traits, and economic well-being.

ficialfined as those with straight borders and those where a significant part offfect of partitioning. Moreover, besides reporting reduced-form estimates linking partitioningflict. Our regional focus allows us to control at a very fine level for geography, thefixed effectsfications to control for factors affecting economic development and civil war likelihood at14 We alsofixed-effects and thus account for broad cross-ethnicity differences

14

and historical features employing regional variation (e.g. Banerjee and Iyer (2005); Iyer (2010); Dell (2010);

Huillery (2009); Acemoglu et al. (2008); Naritomi et al. (2009); Berger (2009); Arbesu (2011); Michalopoulos

and Papaioannou (2010)).

Our within-country analysis is thus similar in spirit to recent works that assess the effect of institutions

7

Second, our work contributes to the literature on the origins of civil con

and Hoe

and Sambanis (2005) for case study evidence on Africa). The literature has examined the

role of many country characteristics, such as income, natural resources, colonization and ethnic

composition on several aspects of civil con

fragmentation and civil war is weak, recent studies document interesting cross-country correlations

linking various aspects of the societal structure with armed con

Reynal-Querol (2005b) and Esteban et al. (2010) show a strong negative correlation between

ethnolinguistic polarization and con

con

ruling coalition consists of many ethnic groups. Matuszeski and Schneider (2006) document

that the likelihood, duration, and intensity of civil wars is much higher in countries where

ethnicities are clustered in speci

that of Englebert et al. (2002), who show a positive cross-country correlation between measures

of su

and civil warfare.

The correlations found in studies exploiting cross-country variation are informative; yet

due to a host of concerns, they cannot be causally interpreted (see Blattman andMiguel (2010)).

The main reason is that the process of border drawing is ultimately related to the process of

state formation and is thus associated invariably with both voluntary and forced movements of

people. Our study accounts for some of the shortcomings of cross-country works. First, it establishes

that African borders are random by showing that there are no systematic di

in numerous geographic, economic, institutional, and cultural characteristics between partitioned

and non-partitioned groups. Second, the use of information on the spatial distribution

of ethnicities in the end of 19th century well before the current national boundaries came into

e

us to condition on country

positioned to consider local factors that may a

flict (see Collierffler (2007), Kalyvas (2007) and Blattman and Miguel (2010) for reviews and Collierflict.15 While the correlation between ethnicflict. Montalvo andflict. Wimmer et al. (2009) find that the likelihood of ethnicflict increases when a large share of the population is excluded from power and when thefic areas within a country. The most closely related study isffocation and dismemberment and political violence, secession attempts, border disputes,fferencesffect alleviates concerns related to the movement of people. Third, our micro approach allowsfixed-factors and ethnic-family factors.16 Finally, we are also wellffect civil conflict.17

15

evidence.

See among others Collier and Hoeffler (1998), Collier et al. (2004), and Fearon and Laitin (2003) for crosscountry

16

quite sensitive to the employed set of conditioning variables. Since there are at most

and potentially numerous variables a

single variable utilizing solely the cross-country dimension. Moreover, because the key correlates of civil con

are themselves highly correlated (e.g. poverty, corruption, and fragmentation go in tandem) and contain nonnegligible

measurement error the estimates are plagued with multi-colinearity (see Levine and Renelt (1992),

Martin et al. (2004), and Ciccone and Jarocinski (2010) for a discussion of these issues in the similar context of

cross-country growth regressions.). The regional approach gives us more degrees of freedom and multi-colinearity

concerns are mitigated because the main explanatory variables at this level of aggregation are not collinear.

As the sensitivity analysis of Herge and Sambanis (2006)) shows the estimates of the cross-country works are180 country observationsffecting civil conflict, it is eminently difficult to isolate the effect of aflict

17

Buhaug and Rod (2006) study the local determinants of civil war in Africa using as a unit of analysis 100

8

Structure

In the next section we present the regional data on civil con

We also discuss some issues on estimation and inference. In Section

of the African border design. Section

e

and duration). Section

development. Section

flict and satellite light density.3 we establish the arbitrariness4 reports the results from our analysis on theffect of partitioning on various aspects of civil conflict (number of war incidents, casualties,5 gives our results on the effect of partitioning on regional economic6 concludes.

2 Data and Empirical Methodology

2.1 Civil War Data

The main data source for the occurrence and duration of civil wars comes from the Uppsala

Con

Con

cover the period

flict Data Program (UCDP)/International Peace Research Institute, Oslo (PRIO) Armedflict Dataset, Version 4-2006, initially assembled by Petter et al. (2002) and extended to1946−2005.18 We limit our analysis to armed conflicts that started after 1970

when the majority of African states had gained independence.

war we focus on con

group(s) without intervention from other states (intrastate) and internationalized intrastate

con

intervention from other states (secondary parties). According to the UCDP/PRIO dataset

since

internal (e.g. the civil war of 1990 in Rwanda when Congo intervened and the civil con

the late 1990s in Guinea-Bissau where Guinea and Senegal intervened), with the overwhelming

majority (

We use the dataset of Raleigh et al. (2006) to obtain information on the spatial extent of

each civil war. This dataset assigns to each con

kilometers. The coordinates represent general estimates of where battles have occurred whereas

the radius indicates the largest geographic extent of the con

19 Following the literature on civilflicts between the government of a state and one or more internal oppositionflicts between the government of a state and one or more internal opposition group(s) with1970 in Africa there have been in total 49 civil wars, 7 are classified as internationalizedflict in42) classified as internal armed conflicts.flict a centroid with a corresponding radius inflict zone.20 The location of several

kilometer by 100 kilometer grids. While they do not control for country characteristics or

fixed-effects, they

fi

diamond mines and petroleum

nd that conflicts are more likely to occur far from the capital, near national borders, and close to regions withfields.

18

the use of armed force between two parties, of which at least one is the government of a state, results in at least

25 battle-related deaths

is the party controlling the capital of a state.

Armed Conflict is defined as “a contested incompatibility that concerns government and/or territory where.” A minimum of 25 battle-related deaths per year and per dyad is required. Government

19

represents when the con

are similar if we include the 1960’s or limit our attention in the period

In particular, we are considering conflicts that are classified with a start date as early as 1970 where the dateflict for the first time reached 25-battle-related deaths in a calendar year. The results1980 − 2005.

20

given point in time, the actual con

There are limitations with respect to the georeferencing conflict. For example, the authors note that at aflict zone might be more constrained than the maximum that is recorded.

9

con

having a

flicts has changed over time. For example, the long-lasting Liberian civil war is coded asffected 3 different conflict zones.21 Overall, the 49 African civil wars between 1970 and

2005

We also construct tribe-speci

Gleditsch (2005). This dataset reports for each con

of the number of civilians and combatants killed in the course of combat.

identi

measures of civil war.

played out in 77 conflict zones.fic measure of combat deaths with data from Lacina andflict-year a low, a high and a best estimate22 Using the eventfier we merge the three datasets, so as to examine the effect of partitioning on various

2.1.1 Civil Con

flict Incidence

Based on Raleigh et al. (2006) we generate a map depicting each con

with a civil war from

ethnolinguistic map (Figure

have been a

of civil war incidence at the ethnicity level. The

wars.

the index capturing the number of civil wars is a coarse measure of civil war intensity, because

it does not take into account that the zone of a con

calculate for each tribal area the number of con

distribution of civil war incidents at our ethnic level of analysis.

flict zone associated1970 to 2005. Then we project the constructed map on top of Murdock’s1). This allows us to identify in a systematic way which ethnicitiesffected by civil conflict during this period. We construct two alternative indexesfirst measure captures the number of civil23 If the civil war changed location over time, we combine all the conflict zones. Hence,flict may migrate over time. Thus, we alsoflict zones. Figures 2-2 portray the spatial

Furthermore, the authors de

contours of international boundaries, mountains, rivers, etc. To mitigate such issues in the empirical analysis

several indexes of war intensity are constructed by taking into account the size of the tribal area a

armed con

fine a circular zone of conflict whereas the actual shape is more likely to follow theffected byflict.

21

Namely in 1980 the conflict zone centroid had the following coordinates:  = 632 and  =

−

108 and a radius of 50 kilometers. From 1989 to 1995 the conflict zone’s centroid moved to  = 6 and



centred on

= −10 with a radius of 300 kilometers. The final part of the conflict between 2000 and 2003 was = 75 and  = −105 with a radius of 150 km.

22

the sensitivity of our results to the low and high estimate. For two incidents the best estimate is unavailable. In

this case we replace it with the average of the high and low estimates.

We use the best estimate death measure as our benchmark number for battle fatalities; yet we also explore

23

Ethnologue’s language family tree); yet

not considered in our analysis. Also, we eliminate the Guanche, a small group in the Madeira islands that is

currently part of Portugal.

Murdock’s map includes 843 tribal areas (the mapped ethnicities correspond roughly to levels 7 − 8 of the8 areas are classified as uninhabited upon colonization and are therefore

10

Ü

Number of Civil Wars

Between 1970-2005

National Boundaries

0

1

2

3

4

5

Ü

Number of Conflict Zones

Between 1970-2005

National Boundaries

0

1

2

3

4

5

6

7

8

Figure

2 Figure 2

2.1.2 Civil Con

flict Casualties

We construct ethnic-speci

we calculate for each con

related deaths across years. (For example in the case of the civil war in Sierra Leone (event ID

fic estimates of civil war casualties with the following procedure. First,flict zone the total number of casualties by summing up the battle

187

the overall area that a con

area of

been a

historical homeland. Thus,

Mende’s territory). Fourth, assuming that total casualties are distributed uniformly across the

total area of the respective con

total number of casualties with the fraction of each ethnicity’s territory that falls within the

con

(

for each con

Liberia (event ID

) the total battle fatalities between 1991 and 2000 sum up to 12 997). Second, we calculateflict zone extends to. (The Sierra Leone civil war affected a total54 287 2). Third, we estimate for each ethnic group how much of its homeland hasffected. (The Sierra Leone civil war involved 18 770 2 out of 22 946 of the Mende’s35% of the civil war in Sierra Leone has taken place within theflict zone, we derive ethnic-specific casualties by multiplying theflict zone. (This implies that 35% of the 12 997 battle deaths of the Sierra Leone civil war4 497 casualties) have taken place in Mende’s homeland). Fifth, we repeat this calculationflict zone. For example, Mende’s homeland takes up 29% of zone 1 of civil war in146), that has caused a total of 4 058 battle deaths; this translates into 1 175

casualties to the Mende’s historical homeland. Similarly, con

war (event ID

11

fatalities of the Mende to

by con

have a casualty rate of

plots civil con

flict zone 2 of the Liberian civil146) adds an additional 1011 casualties bringing the total number of civil war6 680 individuals.24 Since larger tribes are more likely to be affectedflict, we normalize battle deaths by the area of each group. The Mende, for example,291 casualties per thousand of square kilometers. Figure 3 (below)flict casualties for the 834 ethnic areas in our sample.

2.1.3 Con

flict Duration

Analogously we obtain tribe-speci

each ethnic area the fraction of land that has been a

the Mende, for example,

(event ID

fic estimates on civil war duration. First, we calculate forffected by a civil war. (In the case of81% of its homeland has been affected by the Sierra Leone civil war187) whereas 78% and 85% of its territory has been affected by zone 1 and zone 2

of the Liberian civil war (event ID

con

for

Third, we sum the e

duration for each ethnic group. (Doing so shows that Mende were under civil war for

146) respectively). Second, we weigh the duration of eachflict with the fraction of the tribal area involved. (Since the civil war in Sierra Leone lasted10 years (1991-2000), the effective duration for Mende’s homeland is 10 ∗ 081 = 81 years).ffective duration across all conflicts, so as to derive an estimate of civil war1704

years). Figure

3 graphs the duration of civil conflict across ethnic groups.

Ü

Number of Casualties

Between 1970-2005

National Boundaries

0

1 - 700

701 - 2910

2911 - 5701

5702 - 9252

9253 - 16736

16737 - 29742

Ü

Duration of Civil Conflict

Between 1970-2005

National Boundaries

0

1 - 5

6 - 10

11 - 15

16 - 19

20 - 24

25 - 29

30 - 34

35 - 47

Figure

3 Figure 3

2.2 Development Data

Given data unavailability on income or other development proxies for a large number of African

countries at our ethnic level of analysis, we follow Henderson et al. (2009) and use satellite

24

See Appendix Figures 1 and 1 portraying the conflict incidents on Mende’s homeland.

12

data on light density at night to measure economic activity. Data come from the Defense

Meteorological Satellite Program’s Operational Linescan System (DMSP-OLS) that reports

images of the earth at night captured from 20:00 to 21:30 local time. The measure is a sixbit

(

kilometer), which is averaged across overlapping raw input pixels on all evenings in a year.

To construct light density at the desired level of aggregation we average the digital number of

luminosity across pixels that fall within the respective boundaries of an ethnic group (results

are similar if we use the median value of light intensity). Using these data we construct average

light density per square kilometer between

Chen and Nordhaus (2010) argue that luminosity data can be quite useful for regional

analyses in war prone countries and economies with data statistics of poor quality. Henderson

et al. (2009) further show that satellite light density captures well abrupt changes in economic

activity at a local scale in Africa (as for example during the Rwandan genocide or when large

deposits of rubies and sapphires were accidently discovered in Madagascar). In the same vein,

Michalopoulos and Papaioannou (2010) show a strong correlation between light density at

night and GDP per capita across African countries as well as a signi

negative correlation between light density and infant mortality. Figure

in

0 − 63) digital number calculated for every 30-second area pixel (approximately 1 square2007 and 2008.ficant within country4 reports light density2007 − 2008 for the 834 ethnic areas.

Ü

Light Density Across

Tribal Areas 2007-2008

National Boundaries

0.000 - 0.001

0.002 - 0.005

0.006 - 0.015

0.016 - 0.043

0.044 - 0.127

0.128 - 0.366

0.367 - 1.058

1.059 - 3.054

3.055 - 8.810

8.811 - 25.414

Figure

4

13

2.3 Preliminary Descriptive Evidence

Table

the incidence of civil war. Out of the

non-negligible amount of

experienced

with the highest incidence of civil war are the Afar and the Esa which during

1 presents descriptive statistics of the civil war measures. There is large variation in834 tribes, 343 ethnicities suffered one civil war, but a199 ethnicities experienced two distinct armed conflicts; 54 ethnicities3 civil wars, while 12 tribes were affected by four or even five conflicts. The groups1970 − 2005

experienced

Afar being partitioned between Ethiopia, Eritrea and Djibouti and the Esa located in the

border of Ethiopia and Somalia. The number of con

5 civil wars. Both groups have been greatly impacted by the border design withflict zones also varies considerably (from 0

to

con

kilometers is an indicator of how destructive civil con

is

8). Not surprisingly, there is a strong correlation, 085, between civil war occurrences andflict zones per ethnic area (Table 1). The number of casualties per thousand of squareflict has been. The average casualty rate40 fatalities. The average rate however, masks a great deal of heterogeneity. In particular,

50%

like the Wanga and the Sabei partitioned between Kenya and Uganda, have been involved only

in one civil war between

war. Con

(corresponding to the long-lasting civil war in Sudan (

have experienced incessant warfare with the case of the Alur group (partitioned between Zaire

and Uganda) registering the highest duration of warfare across African tribes.

correlation between the various measures of civil war (reported in Table

measure captures distinct aspects of civil war.

Before presenting our econometric framework we compare the means (results are the

same with the medians) of the various measures of civil wars and development across di

groups of ethnicities. In Table

tribes, border tribes with more than

partitioned ethnicities, de

single country (using alternative thresholds for de

results). Both border and partitioned tribes are systematically di

groups. Border and partitioned groups have experienced more civil wars, have su

higher casualty rates, and have seen their homelands under civil con

Furthermore, economic performance is on average

of tribal areas experienced less than 3 casualties per thousand kilometers. Ethnic groups1970 and 2005 however they have lost thousands of people during thisflict duration varies from 1 year (which is the case for 27 conflict zones) to 22 years1983-2004)). There are tribal areas that25 The modest1) suggests that eachfferent1 we classify ethnicities in 3 groups, namely non-border90% of their homeland belonging to a single country, andfined as those with less than 90% of their territory found within afining a partitioned group delivers similarfferent compared to nonborderfferedflict for longer periods.60% lower.

25

involved in overlapping con

Note that because of how the duration variable is constructed ethnic groups whose homeland has beenflict zones may have a duration of civil war larger than 35 years. This is the case for

7

ethnic groups.

14

2.4 Econometric Speci

fication

We estimate the long-run e

empirical speci

ffect of the scramble for Africa running variants of the followingfication







The dependent variable,

historical homeland of ethnic group

= 0 +   +  + 0Φ +  ( ) +  +  +  (1), reflects civil conflict and economic development in the. Ethnicities are part of larger ethnic families (clusters) 

and countries

by the political boundaries. Thus the coe

identify partitioned ethnicities in our benchmark estimates we require that at least

historical homeland of each ethnic group falls into more than one countries.

indicator that equals one when the historical homeland of an ethnic group is adjacent to the

national border, but more than

include the

error in identifying partitioned ethnic groups based solely on Murdock’s map using the

.26   is an indicator that takes the value 1 when a tribal area is partitionedfficient  captures the direct effect of partitioning. To10% of the27  is an90% of the land area of this group falls in the same country. We variable for two reasons. First, to account for potential measurement10%

threshold. Second, to capture potential spillover e

reside (

of spillovers below. Appendix Table

ffects from areas where partitioned ethnicities  = 1) to neighboring border regions ( = 1). We return to the issue1 gives the list of all partitioned ethnic groups. Vector



a malaria stability index and land’s suitability for agriculture; natural resources as proxied

by the presence of diamond mines and petroleum

institutional structure and economic development. The Data Appendix gives detailed variable

de

To capture unobservable characteristics and spatial e

for smooth functions of geographic location by introducing a cubic polynomial (

in the latitude and longitude of the centroid of each ethnic group.

country

nation-wide factors related to post-independence policies, contemporary institutions, ethnolinguistic

diversity and development that a vast literature on civil warfare identi

0 includes geographical controls, like land area and elevation; ecological features, such asfields; pre-colonial ethnic traits related tofinitions and sources.ffects in many specifications we control ( ))28 In many specification we includefixed-effects () and ethnic family fixed-effects ( ). Country constants capturefies as significant

26

Partitioned ethnicities are assigned to the country where the centroid of the historical homeland lies.

27

are similar.

We also experiment with other thresholds to identify partitioned ethnicities (5%, 20%, and 30%). The results

28

Letting  denote latitude and  denote longitude the polynomial becomes  +  + 2 + 2 +  + 3 + 3 +



concerns of over

adding a cubic polynomial in distance to sea coast,

2+2. As Dell (2010) notices this parameterization is transparent and quite flexible. As there might be somefitting, we report in all tables results without the polynomial. We also estimated specificationsfinding similar results (not reported).

15

correlates of armed con

and other hard-to-observe and measure ethnic-speci

flict. The ethnic family fixed-effects capture broad cultural, institutional,fic factors (Murdock assigns the 834

groups into

A notable fraction of the observations on civil warfare (number of civil wars, number

of war zones, battle casualties and duration) and our proxy for regional development (light

density) takes on the value of zero. Moreover the civil war measures and the satellite light

density variable are highly skewed as we have many observations close to zero and a few

extreme observations in the right tail of the distribution. (See Table

non-linear ML speci

(2002)). The non-linear estimators are appealing, because they do not require log-linearizing

the dependent variable and thus preserve the higher moments of the distribution, (see Silva

and Tenreyro (2006) and Silva et al. (2010)). Yet we also report log-linear LS speci

taking the log of one plus the dependent variable.

In all speci



errors at the country-level and at the ethnic-family level using the multi-way method of Miller et

al. (2006). This correction also accounts for arbitrary residual correlation within each country

and each ethnic family. We also estimated standard errors using Conley’s GMM method to

account for spatial dependence of an unknown form,

96 ethnolinguistic clusters/families).1). Hence, we estimatefications with a Poisson and/or a Negative Binomial process (Wooldridgeficationsfications we account for spatially correlated residuals () clustering standardfinding similar results.

3 Are African Borders Arti

ficial?

While the anecdotal and historical evidence point out that Europeans did not take into account

local conditions and ethnic characteristics during the scramble for Africa, it is necessary

to investigate whether there are systemic di

border groups, and non-partitioned ethnicities.

institutional, and other tribe-speci

partitioned by the national borders, estimating probabilistic models of the following form:

fferences between partitioned ethnic groups, other29 We test whether there are geographic, economic,fic factors that predict whether an ethnic group is

 

The dependent variable (

of an ethnic group has been partitioned into more than one contemporary states. We

explore the robustness of our results using a broader index that takes the value one when the

 =  + 0Ψ + 0Θ +  (2) ) equals one when at least 10% of the historical homeland

29

and non-split ethnic groups is necessary, because in some cases Europeans did try taking into account

local conditions (as for example when German West Africa was split into Urundi and Rwanda). Likewise in

two cases (Cameroon-Nigeria; Ghana-Togo) there were referenda on the redrawing of borders at independence.

Moreover we had the secession of Eritrea from Ethiopia (in

(in

Examining formally whether there are systematic differences on observable characteristics between partitioned1993) and the unification of Tanganyika and Zanzibar1964).

16

historical homeland of an ethnicity falls into more than one country irrespective of the size of

the partitioning (

and economic features at the ethnicity level;

institutional, cultural, and early development traits, extracted from Murdock (1967). Tables

 =   +).  is a vector of geographic, ecological, is a vector of ethnicity-specific pre-colonial2

and

the other regressors and double-clustered at the country and the ethnic- family level standard

errors. (The results are similar when we estimate linear probability or logit models).

In Table

in the full sample (

is the

partitioning index,

large territories in the pre-colonial period are more likely to be partitioned. This

line with the historical evidence that colonizers drew arbitrary lines across imprecise maps.

The estimates further show that ethnicities residing in areas with large water bodies (lakes

and rivers) were more likely to

in line with the historical narrative that Europeans tried to use some natural barriers while

delineating spheres of in

and malaria prevalence are not signi

include region

across the continent. The results remain unchanged.

In columns (5)-(6) we explore whether colonial powers took into account pre-existing

development and the slave trades that preceded the scramble for Africa when designing the

borders. Following Nunn and Watchenkon (2010) we proxy the pre-slave trade level of economic

development using an indicator variable that equals one when a city with population

exceeding

We also include a measure of ethnic-speci

data from Nunn (2008). Neither the city in

exported during the slave trades a

the historical homeland of each ethnic group and the amount of its water streams continue

to be the only signi

include indicators identifying ethnic areas with diamond mines and petroleum

the initial phase of colonization Europeans were mostly interested in agricultural goods and

minerals, adding these two indicators allows us to investigate whether partitioned ethnicities

di

one of the most robust correlates of civil warfare across countries is oil and other underlying

17

hydrocarbon deposits, as well as diamonds (e.g. Ross (2006); Herge and Sambanis (2006)).

Again there is no systematic di

groups. In columns (7)-(8) we add log population density around African independence (in

3 present the results. The tables report probit marginal effects estimated at the mean of2 we examine the role of geographical, ecological, and demographic features834 observations). In odd-numbered specifications the dependent variable  index, while in even-numbered columns the dependent variable is the broad. The results in columns (1)-(2) show that ethnic groups spanningfinding is infind themselves split by national boundaries. This result isfluence. Elevation, distance from the sea, land’s agricultural quality,ficant predictors of partitioning. In columns (3)-(4) wefixed effects to account for the somewhat different patterns of border design20 000 people in 1400  was present in the historical homeland of an ethnicity.fic slave trades between the 16th-18th century using1400 indicator nor the log number of slavesffect which ethnicities have been partitioned. The size officant correlates of partitioning. The specifications in columns (5)-(6) alsofields. While inffer from non-partitioned ethnic groups in terms of natural resources. This is important asfferences in natural resources among split and non-split ethnic

1960

estimate on log population density in

in economic performance and urbanization between partitioned and non-partitioned ethnicities.

Hence, any contemporary divergence in economic performance between these two groups

cannot be attributed to di

In Table

cultural traits a

Murdock (1967) does not provide information for all the ethnicities in his Ethnographic

Atlas (1959). In all speci

measures, size and land area under water, that were found to be signi

In columns (1)-(2) we investigate whether Europeans took into the account the degree

of political centralization of the African ethnicities when designing the borders. Following

Gennaioli and Rainer (2006, 2007), we proxy political centralization with an indicator variable

that equals zero when Murdock assigns an ethnicity either as "

(e.g. Xam or the Ibo); and becomes

chiefdom

a negative coe

(3)-(6) we examine whether there were systematic di

protection between ethnicities that were partitioned and those that were not. In (3)-(4) we use

Murdock’s class strati

society is egalitarian "

in Kenya). The variable equals one when there are "

aristocracy is present, or there is complex class di

the Shilluk in Sudan).

property rights institutions with a dummy variable that equals one when the society has some

form of inheritance rules for property (e.g. the Ewe in Togo and Ghana or the Soga in Uganda)

and zero otherwise (e.g. the Fang in Gabon or the Namshi in Cameroon). In all permutations

neither proxy of pre-colonial property rights institutions enters with statistically signi

coe

institutional development was higher in areas with intensive use of agriculture (Fenske (2010)

provides empirical evidence supportive to this conjecture); thus in columns (7)-(8) we augment

). In a Malthusian regime where richer areas are more densely populated, the insignificant1960 implies that there were no systematic differencesfferences in economic conditions in the eve of African independence.3 we examine whether ethnic-specific pre-colonial institutional, economic, andffect which ethnic groups have been partitioned. The sample size drops becausefications we include the region fixed-effects and the two geographicalficant predictors of partitioning.stateless" or "a petty chiefdom"1 when the ethnicity is part of either a "large paramount" or a "large state" (e.g. Thonga and Zulu). The political centralization enters withfficient, but the estimate is statistically indistinguishable from zero. In columnsfferences in the degree of property rightsfication index and define a dummy variable that equals zero when thewithout significant class distinctions" (e.g. Fang in Nigeria or theKikuyuwealth or elite distinctions, a hereditaryfferentiation" (e.g. the Yoruba in Nigeria or30 In columns (5)-(6) we follow Fenske (2010) and proxy the presence officantfficient. African scholars (e.g. Hopkins (1973); Austin (2008)) argue that economic and

30

original

This transformation follows Gennaioli and Rainer (2006, 2007). The results are similar if in turn we use the0 − 4 class stratification index.

18

the speci

the ethnicity level, failing again to detect a signi

the pre-colonial institutional, cultural, and economic characteristics in the speci

we are not able to

The results reported in tables

design of African borders. Overall Europeans did not take into consideration local geography,

ethnic-speci

of dozens of potentially relevant variables, only land size and the presence of water streams

correlate with partitioning. Moreover, the explanatory power of the models is poor. Mc

Fadden’s pseudo-

that of the full speci

the linear probability models (not reported) is always below

perform quite poorly in predicting which ethnicities have been partitioned. For example, the

full speci

fication with a 0 − 10 index measuring the importance of agricultural production atficant effect. In columns (9)-(10) we include allfication. Againfind any significant variables affecting the likelihood of partitioning.2 and 3 strongly support the notion of the arbitraryfic institutional, cultural and economic features during the scramble for Africa. Out2 (that compares the log likelihood value of the constant-only model withfication) is low across all permutations, at most 010. Likewise the 2 of012. The probit specificationsfications in columns (7) and (8) of Table 2 predict correctly ((0Ψ)  05) only 30

out of the

when we use the broad split index (

231 partitions with the benchmark index ( ) and 164 out of the 358 partitions).

4 Partitioning and Civil Con

flict

4.1 Civil Con

flict Incidence

In Table

report Poisson ML estimates where the dependent variable equals the number of civil con

(Panel

Speci

independence, size and area under water (the only geographic variables that are correlated

with partitioning) partitioned ethnicities are signi

the cubic polynomial in latitude and longitude (column (2)) has little e

Column (3) includes a rich set of controls, re

elevation, distance to the sea coast), current and past urbanization (dummy variable

that takes on the value one when a current capital city belongs to the historical homeland of

an ethnic group and an analogous indicator that equals one if a major city was in the historical

homeland in

Accounting for these factors seems a priori important, because the cross-country literature

documents signi

warfare. For example Fearon and Laitin (2003)

con

19

regional studies (e.g. Buhaug and Rød (2006); Bellows and Miguel (2009)) show that con

is higher in areas with diamond mines and petroleum

little impact on the coe

4 we examine the effect of partitioning on the incidence on civil conflict. Both panelsflicts) and the number of conflict zones (Panel ).fication (1) shows that, conditional on region fixed-effects, population density atficantly more prone to civil conflict. Addingffect on the estimate.flecting local geography (land suitability for agriculture,1400 AD) and natural resources (indicators of a diamond mine or an oil deposit).ficant correlations between many of these variables and various aspects of civilfind that there is a higher likelihood of civilflict in mountainous countries. Likewise both cross-country works (e.g. Ross (2006)) andflictfields. Yet adding these controls hasfficient on the partitioning index.31 In column (4) we add ethnic family

fi

on the partitioning variable falls somewhat, though it retains signi

level. In column (5) we add country

dummy is signi

an increase of approximately

xed-effects to account for hard-to-measure differences across African groups. The coefficientficance at the 99% confidencefixed-effects.32 The estimate on the partitioningficant at the 99%. The coefficient implies that partitioned ethnicities experience025 log points in the number of conflicts. This translates into a

28%

ethnicities reside.

higher increase in civil conflict activity (exp(025) − 1 = 028) in areas where partitioned33

The estimates further show that there is a higher degree of civil war occurrence in

border areas populated by ethnic groups that were a

smaller degree. Yet the economic e

much smaller than the coe

insigni

ffected by the border design, but to affect implied by the coefficient of the  index isfficient of  . Moreover, the coefficient on  becomesficant when we condition on the rich set of controls and/or fixed-effects.

4.2 War Casualties

In Table

5 we study the effect of the artificial border design on the number of casualties. Panel



of casualties standardized by land area. Due to overdispersion in the dependent variable,

speci

To illustrate the robustness of our results, Panel

the dependent variable is the log of one plus the number of casualties per thousand of square

kilometers.

The results in column (1) show that casualties from armed con

higher in areas where partitioned ethnic groups reside; casualties are also higher in other border

regions. The estimates of the negative binomial speci

reports negative binomial (NB) ML estimates, where the dependent variable is the numberfication tests suggest that the negative binomial model is preferable to the Poisson model. reports analogous LS estimates, whereflict have been muchfication imply that casualties are by 835%

31

that there is a higher intensity of civil warfare in the capitals. The natural resource measures enter with positive

estimates that in most (though not all) speci

con

In all specifications the capital city dummy variable enters with a positive and significant estimate indicatingfications are significant. There is also some evidence that civilflict is higher in more mountainous regions.

32

since in these countries we have just one ethnic group.

When we add country fixed-effects we lose variation from Rwanda, Swaziland, Burundi and the Comoros,

33

(

as large as the coe

on

considerably and becomes insigni

prefer the Poisson speci

when we introduce ethnic-family and country

Ordered probit estimation yields similar results. The coefficient on the   index is highly significantt-stat higher than 4 in all permutations). In all ordered probit specifications the coefficient on   is twicefficient on the  index. As our Poisson results in columns (1)-(4) the estimate is statistically significant at standard confidence levels in the simple specifications, but dropsficant when we condition on the rich set of controls (as in column (4)). Wefications to avoid the incidental variables problem of the ordered probit specificationfixed-effects.

20

(

of partitioned ethnicities and of the other border ethnic groups. The estimates in (1) do not

take into account the potential spillovers in civil con

reside (the "treatment" group) to areas where non-border ethnicities reside ("control" group).

Spillovers may emerge for numerous reasons. First, the battleground between a partitioned

ethnic group and another tribe or the central government might take place outside the historical

homeland of the split ethnicity. Second, in many cases the con

ethnicity leads to displacement and refugee

con

in the same ethnolinguistic family as the partitioned ethnic group might get involved in the

con

Southern Sudanese troops in their long

and religiously distinct Northern Sudan.

Fourth, we have introduced some spatial correlation with the way the dependent variable

is generated. If spillovers are present, the estimates in Table

because of non-accounted externalities to ethnic areas where non-partitioned ethnic groups

reside (see Miguel and Kremer (2004)). In other words if there are war spillovers to areas where

non-split ethnic group reside, then civil war activity in the control group of ethnicities will be

higher as compared to the case where partitioning did not have any e

thus include as controls the total number of casualties per square kilometer in each country and

each ethnic-family, netting out the number of casualties of each ethnic group’s own homeland;

this allows us to account for spillovers in civil con

families. The results show that there are sizable spillovers in both dimensions. For example the

estimates in Panel

where an ethnicity resides is associated with at

of this ethnic group. Likewise a

the same ethnic cluster is associated with

Appendix Table

that model explicitly spatial correlation corroborate the

are more prone to con

In columns (4) and (5) we include ethnic-family and country

while in column (6) we include both. The ML-NB coe

remains stable (

and the rich set of controls, the estimate on the

LS speci

21

exp(06072)−1 = 0835) and by 74% (exp(0555)−1 = 074) higher in the historical homelandflict from areas were partitioned ethnicitiesflict in the area of the partitionedflows to nearby areas, which in turn may spurflict (e.g. Salehyan and Gleditsch (2006) and Blattman and Miguel (2010)). Third, tribesflict to support their peers. For example many Ugandan tribes assisted the ethnically similarfight against the government troops of the ethnically5 column (1) are lower boundsffect. In columns (2)-(3) weflict within countries and within ethnic column (3) imply that a 10% increase in civil war casualties in the country25% increase in casualties in the homeland10% increase in the number of casualties in nearby areas of18% increase in casualties for each ethnic group. In we report estimates from a general spatial lag model. These specificationsfinding that partitioned ethnic groupsflict compared to the other border and non-border ethnicities.fixed-effects respectively,fficient on the partitioning indicator061). When we control for ethnic-family factors, countrywide characteristics, index drops considerably and in thefication becomes indistinguishable from zero. In the NB-ML specification with country

fi

but the size of the coe

between the two coe

we report double

casualties with similar results. The coe

Moreover, the estimate on

xed-effects and ethnic family fixed-effects, the estimate on  retains statistical significance,fficient is half of that for the partitioning index. (The differencefficients is statistically significant at the 90% level.) In columns (7) and (8)fixed-effects estimates using the low estimate and the high estimate of conflictfficient on   is positive and highly significant.  is two times larger than the  coefficient.

4.3 Con

flict Duration

In Table

negative binomial ML estimates, while Panel

variable is the log of one plus the war duration. The simple speci

suggest that the duration of civil con

ethnicities reside. The NB-ML estimate implies that civil con

in the historical homeland of partitioned ethnic groups, as compared to regions where nonpartitioned

groups reside (

that captures the degree of civil war duration for ethnic groups that have been a

lesser extent by the arti

(

with the average duration of civil con

within each ethnic family (netting out the duration in each ethnicity’s own location). The LS

estimates suggest that a

and ethnic-family is associated with approximately

In columns (4)-(5) we add a vector of ethnic-family and country

The coe

Column (6) reports the most restrictive speci

resources, ecology, ethnic-family unobservables and country

that civil wars last

where non-partitioned, non-border groups reside (

further show that compared to non-partitioned non-border ethnic groups, civil con

is higher by

design reside (

of Fearon (2004) and Fearon and Laitin (2010), showing that wars involving land con

a peripheral ethnic minority and state-supported migrants of a dominant ethnic group

are on average quite long-lived (see also Wimmer,

22

6 we examine the effect of partitioning on the duration of civil conflict. Panel  gives reports LS estimates where the dependentfications in column (1)flict is significantly higher in regions where partitionedflicts last on average 38% longerexp(032) − 1 = 0377). The coefficient on the  variableffected to aficial border design is positive, but smaller and (marginally) insignificantp-value: 015). In columns (2)-(3) we control for spatial spillovers augmenting the specificationflict in all other groups within each country and all groups10% increase in the duration of armed civil conflict in the same country5% increase in local civil war duration.fixed-effects respectively.fficient on the   indicator retains its economic and statistical significance.fication, where we control for geography, naturalfixed-factors. The results suggest35% more in areas of partitioned ethnic groups, as compared to regionsexp(030)−1 = 035). The NB-ML estimatesflict duration185% in border areas where ethnicities modestly affected by the artificial borderexp(0173) −1 = 0188). Our estimates are in line with the cross-country resultsflict betweenet al. (2009)).

5 Partitioning and Development

The results in Tables

homeland of ethnic groups that have been partitioned by national borders. The results also

show that border ethnicities less impacted by the border drawing do su

con

regional development is lower in areas where partitioned ethnic groups reside? This is the

question we tackle in Table

Table

2 − 6 reveal that civil war intensity is much higher in the historicalffer more from civilflict compared to non—partitioned groups, though to a lesser extent. Does this imply that7.7 reports NB-ML estimates (in Panel ) and log-linear LS estimates (in Panel



(1) show that regional development is signi

been a

adding the RD-type cubic polynomial in latitude and longitude. The coe

the

(3) we control for natural resources, geographical features, ecological characteristics, historical

and current urbanization. While many of these variables enter with signi

the estimates on the partitioning indicator variables are not a

evidence produced in Table

endowments between partitioned ethnicities and non-partitioned ones.

Cross-ethnicity di

country-level factors, such as income, institutional quality, foreign aid, education, country size,

etc. Moreover Michalopoulos and Papaioannou (2010) show that ethnic-speci

and social features correlate signi

augment the speci

the e

rich set of local controls, allows us to isolate the direct e

The estimate on

speci

yet the coe

development is approximately

to non-border ethnic groups (

extreme observations. Thus in columns (6) and (7) we report speci

winsorize the top

in the NB-ML speci

also enters with a negative and signi

), where we use luminosity to proxy for regional development. The estimates in columnficantly lower in areas of ethnic groups that haveffected by the artificial border design. In column (2) we control for geographic locationfficients on both  and the  variables are negative and statistically significant. In columnficant coefficients,ffected. This is in line with the2 that there are no systematic differences with respect to geographicfferences in regional development (and electrification) are also driven byfic institutionalficantly with luminosity. Thus in columns (4) and (5) wefication with ethnic-family fixed-effects and country fixed-effects. Examiningffect of partitioning within countries and within ethnic families, while conditioning on affect of partitioning on development.  in the NB-ML specification drops in absolute value compared to thefication without the ethnic-family and the country fixed-effects (from −0835 to −0486);fficient retains its significance at the 99% confidence level. It implies that regional60% lower in areas where partitioned ethnic groups reside, comparedexp(0485)−1 = 062). The light density data contain somefications where we trim and1% of the distribution, respectively.34 The coefficient on the   dummyfications retains significance at the 99% level. The  dummyficant estimate, suggesting that regional development is

34

The results are similar if we trim or winsorize the data at the top 2% or top 5% of the lights distribution.

23

signi

ficantly lower in all ethnic areas were marginally affected by the artificial border design.

6 Further Robustness Checks

We have performed many sensitivity checks to investigate the robustness of our results reported

in the supplementary Appendix Tables

In Appendix Table

development with partitioning are strong when we estimate spatial lag models that account for

spatial interdependencies in the dependent variable and the error term.

In Panel

Northern African countries (Algeria, Egypt, Morocco, Tunisia, Libya). This is a necessary

robustness check because Europeans had long contacts with Northern Africa and were familiar

with local ethnic and geographic conditions. The results are similar to the ones presented

in Tables

in Southern Africa (Malawi, Mozambique, Namibia, South Africa, and Zimbabwe). This is

important because the apartheid governments of South Africa intervened in most civil wars in

the neighboring countries. Again all our results remain unchanged.

In Appendix Table

with a continuous measure of ethnic partitioning that captures the probability that a square

kilometer of an ethnic homeland falls to di

the most fractionalized ethnicity is the Malinke (

countries. In all permutations tribal land fractionalization enters with a signi

implying that ethnic partitioning leads to more devastating and prolonged civil con

a lower level of development.

, , and . we show that our results associating civil conflict and regional of Appendix Table  we report NB-ML estimates excluding ethnicities in4-7. In Panel  of Appendix Table  we exclude from the estimation countries we report specifications associating civil war and developmentfferent countries. According to this continuous index077), a group partitioned between 6 differentficant estimateflict and to35

7 Conclusion

We examine the economic consequences of a neglected aspect of colonization, the arti

drawing of African political boundaries among European powers in the end of the 19th century

which led to the partitioning of several ethnicities across African states.

In the

boundaries. Using regional data from Murdock (1959) on the spatial distribution of ethnicities

ficialfirst part of our paper we formally establish the random nature of African political

35

and political science we reviewed suggest that what matters for civil con

ethnicity has been partitioned or not rather than the degree of the split. Second, from a theoretical point of view

a priori there is no clear guidance as to whether propensity to con

degree of partitioning. Finally, to the extent that Murdock’s tribal map certainly contains measurement error,

small di

We prefer the binary index of ethnic partitioning for two reasons. First, all studies in African historiographyflict and development is whether anflict should monotonically increase with thefferences in this continuous measure are more likely to be affected by it.

24

at the time of colonization, we estimate probabilistic models associating ethnic partitioning

with various geographical, ecological, and ethnic-speci

sole exceptions of the size of the historical homeland and the magnitude of water bodies,

there are no other signi

These results support African historiography on the accidental and arti

colonial/national borders.

Second, we examine the e

(incidence, casualties, duration), as this has been hypothesized to be the major e

scramble for Africa. In contrast to most works on the long-run e

con

areas across Africa. Our rich dataset allows us to control at a very

resources, geography and early development. We also include country and ethnic-family

to account for national factors and broad cultural characteristics respectively. We

that partitioned ethnicities compared to tribes that have not been directly a

improper border design, have experienced more civil war incidents that lasted longer and were

more devastating in terms of casualties.

Third, using satellite light density data at night to proxy for development at the ethnicity

level, we show that partitioning is negatively correlated with luminosity. While one could

always argue that another ethnic-speci

between partitioning and development, the largely arbitrary design of contemporary African

borders together with the rich set of conditioning factors at a

of country and ethnic-family

relationship. The uncovered contemporary divergence in comparative economic performance

between partitioned and non-partitioned groups becomes more dramatic when viewed in light of

the

in the eve of colonization and at the time of African independence. Our work suggests that the

scramble for Africa by partitioning ethnicities in di

unrest shaping their economic trajectory.

25

fic precolonial characteristics. With theficant differences between partitioned and non-partitioned ethnicities.ficial drawing of theffect of ethnic partitioning on various aspects of civil conflictffect of theffects of colonization on civilflict (and development), our analysis is based on detailed regional data spanning 834 ethnicfine level for naturalfixedeffectsfindffected by thefic feature is driving the significant negative correlationfine regional level and the inclusionfixed-effects suggests that these correlations imply a causalfinding that partitioned and non-partitioned ethnic groups were economically similar bothfferent countries spurred civil conflict and

8 Data Appendix

Light Density at Night

density observations across pixels that fall within the historical homeland of each of the

: Light Density is calculated at a ethnicity level averaging light834

ethnic groups in Murdock’s Atlas. To smooth weather variation we use the average of the

values in

2007 and 2008.

Source: Available at http://www.ngdc.noaa.gov/dmsp/global_composites_v2.html.

Population Density

UNESCO (1987). Available at: http://na.unep.net/datasets/datalist.php

: Log of population density per sq. km. in 1960 plus one. Source:.

Civil Con

flict Casualties (Best.High/Low Estimate): See text

Number of Civil Con

flicts: See text

Number of Civil War Con

flict Zones: See text

Civil Con

flict Duration: See text

Land Area

: Log surface area of the historical homeland of each ethnic group in 1000

of sq. km.

Source: Global Mapping International, Colorado Springs, Colorado, USA.

Water Area

group covered by rivers or lakes in sq. km.

features" dataset from Global Mapping International, Colorado Springs, Colorado, USA.

: Log of one plus the total area of the historical homeland of each ethnicSource: Constructed using the "Inland water area

Elevation

(NOAA) and U.S. National Geophysical Data Center, TerrainBase, release 1.0 (CDROM),

Boulder, Colorado. http://www.sage.wisc.edu/atlas/data.php?incdataset=Topography

: Average elevation in km. Source: National Oceanic and Atmospheric Administration

Land Suitability for Agriculture

area of each ethnic-country observation. The index is the product of two components re

the climatic and soil suitability for cultivation.

Atlas of the Biosphere. Available at http://www.sage.wisc.edu/iamdata/grid_data_sel.php.

: Average land quality for cultivation within theflectingSource: Michalopoulos (2008); Original Source:

Malaria Stability Index:

mosquitoes indigenous to a region, their human biting rate, their daily survival rate, and their

incubation period. The index has been constructed for

globally.

The index takes into account the prevalence and type of05 degree by 05 degree grid-cellsSource: Kiszewski et al. (2004)

Capital City

historical homeland of each ethnic group area and zero otherwise.

: Indicator variable that equals one when the capital city falls into the

Sea Distance

ethnic group from the nearest coastline, measured in

International, Colorado Springs, Colorado, USA. Series name: Global Ministry Mapping

System. Series issue: Version 3.0

: The geodesic distance of the centroid of the historical homeland of each1000 of km’s. Source: Global Mapping

Petroleum:

26

the historical homeland of an ethnic group and zero otherwise.

v.1.1 contains information on all known on-shore oil and gas deposits throughout the world.

http://www.prio.no/CSCW/Datasets/Geographical-and-Resource/Petroleum-Dataset/Petroleum-

Dataset-v11/

Indicator variable that takes on the value of one if an oil field is found inSource: The Petroleum Dataset

Diamond:

in the historical homeland of an ethnic group and zero otherwise.

Resources.

Indicator variable that takes on the value of one if a diamond mine is foundSource: Map of Diamondwww.prio.no/CSCW/Datasets/Geographical-and-Resource/Diamond-Resources/

City in 1400:

larger than

Indicator variable that takes on the value of one if a city with a population20 000 in 1400 was in the historical homeland of an ethnic group and zero otherwise.

Source: Chandler (1987).

Split

ethnic group is partitioned into di

(1959) ethnic map of Africa with the Digital Chart of the World (DCW) shape

contains the polygons delineating the international boundaries in 1992. Appendix Table 1 reports

partitioned ethnicities.

: Indicator variable that equals 1 if at least 10% of the historical homeland of anfferent countries. Source: Calculated intersecting Murdock’sfile. The latter

Border:

Indicator variable that equals one when a miniscule or small part ( 0% and



intersecting Murdock’s (1959) ethnic map of Africa with the Digital Chart of the World

(DCW) shape

1992. Appendix Table 2 reports partitioned ethnicities.

10%) of the historical homeland of an ethnic group falls into different countries. Source: Calculatedfile. The latter contains the polygons delineating the international boundaries in

Continuous Measure of Partitioning

with

fractionalization is calculated as:

: If an ethnic homeland belongs to  countries denoting the fraction of the tribal homeland belonging to country  then tribal land1 −

P



=1 2



Latitude:

Source: Constructed using ArcGis Software.

Longitude:

Source: Constructed using ArcGis Software.

Regional Indicators:

Africa, Central Africa, Eastern Africa, and Southern Africa.

There are five regional indicator variables, North Africa, WesternSource: Nunn (2008).

Slavery:

Atlantic and Indian Ocean slave trades. Following Nunn (2008) in the regressions we use the log

of one plus the number of slaves standardized by the land area of each ethnic group’s historical

homeland.

Number of persons of each ethnic group that were shipped during the trans-Source: Nunn (2008).

Political Centralization:

Hierarchy beyond Local Community

levels (political complexity) in each society above the local level

index takes the value

27

The binary index is constructed using Murdock’s Jurisdictional0−4 index that indicates the number of jurisdictional. The political centralization0 if the Jurisdictional Hierarchy beyond Local Community variable equals

0

takes on the value

or 1 (when the society is classified as either stateless or forming a small chiefdom). The index1 if the Jurisdictional Hierarchy beyond Local Community variable equals

2

state). This aggregation follows Gennaioli and Rainer (2006, 2007).

, 3, and 4 (when the society is classified as being part of large paramount chiefdom or a largeSource: Murdock (1967).

A revised version of Murdock’s Atlas has been made available by J. Patrick Gray at:

http://eclectic.ss.uci.edu/~drwhite/worldcul/EthnographicAtlasWCRevisedByWorldCultures.sav.

Class Strati

excluding purely political and religious statuses

class distinctions among freemen, ignoring variations in individual repute achieved through skill,

valor, piety, or wisdom.

elite distinctions, the presence of hereditary aristocracy, or complex class di

index is based on Murdock’s

and Rainer (2006, 2007).

fication: This binary index reflects "the degree of class differentiation,". A zero score indicates "absence of significant" A score of 1 indicates some degree of stratification based on wealth orfferentiation. The0 −4 class stratification index. This aggregation follows GennaioliSource: Murdock (1967).

Property Rights.

exist and zero otherwise. This aggregation follows Fenske (2010).

A binary index that equals one when inheritance rights for landSource: Murdock (1967).

Ü

Number of Civil Wars

Between 1970-2005

in Mende Homeland

National Boundaries

Part Affected by Civil War 187

Part Affected by Civil War 146

Part Affected by Civil Wars 146, 187

Ü

Number of Conflict Zones

Between 1970-2005 in

Mende Homeland

National Boundaries

Affected by Civil War 187

Affected by Civil War 187 and Zone 1 of Civil War 146

Affected by Zones 1, 2 of Civil War 146

Affected by Civil War 187 and Zone 2 of Civil War 146

Affected by Civil War 187 and Zones 1, 2 of Civil War 146

Appendix Figure

1 Appendix Figure 1

28

References

Acemoglu, Daron, Maria Angelica Bautista, Pablo Querubin, and James A. Robinson

“Economic and Political Inequality in Development: The Case of Cundinamarca,

Colombia,” in Elhanan Helpman, ed.,

University Press, 2008.

,Institutions and Economic Performance, Harvard

, Simon Johnson, and James A. Robinson

Development: An Empirical Investigation,”

1369—1401.

, “The Colonial Origins of ComparativeAmerican Economic Review, 2001, 91 (5),

, , and

The Modern World Income Distribution,”

1231—1294.

, “Reversal Of Fortune: Geography And Institutions In The Making OfQuarterly Journal of Economics, 2002, 107 (4),

, , and

of Economic Growth

, “Institutions as a Fundamental Cause of Long-Run Growth,” Handbook, 2005, 91, 386—472.

Alesina, Alberto and Ekaterina Zhuravskaya

in a Cross-Section of Countries,”

, “Segregation and the Quality of GovernmentAmerican Economic Review, forthcoming, 2011.

and Eliana La Ferrara

762—800.

, “Fractionalization,” Journal of Economic Literature, 2005, 43,

, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain

Wacziarg

, “Fractionalization,” Journal of Economic Growth, 2003, 8, 155—194.

, William Easterly, and Janina Matuszeski

European Economic Association, forthcoming.

, “Artificial States,” 2011. Journal of the

Arbesu, Horacio Alejandro Larreguy

from a Natural Experiment in Colonial Nigeri,” 2011. mimeo MIT.

, “Colonialism and Economic Development: Evidence

Asiwaju, A.

Press, 1985, pp. 1—18.

, “The Conceptual Framework,” in “Partitioned Africans,” New York: St. Martin

Banerjee, Abhijit and Lakshmi Iyer

The Legacy of Colonial Land Tenure Systems in India,”

, “History, Institutions, and Economic Performance:American Economic Review, 2005,

95

29

, 1190—1213.

Barkindo, Bawuro M.

Asiwaju, ed.,

194.

, “The Mandara Astride the Nigeria-Cameroon Boundary,” in A.I.Partitioned Africans, New York, NM: St. Martin’s Press„ 1985, pp. 155—

Bates, Robert H.

Press, 1981.

, in “States and Markets in Africa,” Berkeley, CA: University of California

Bellows, John and Edward Miguel

, “War and Local Collective Action in Sierra Leone,”

Journal of Public Economics

, 2009, 93, 1144—1157.

Berger, Daneil

mimeo, New York University.

, “Taxes, Institutions and Governance: Evidence from Colonial Nigeria,” 2009.

Blattman, Chris and Edward Miguel

, “CivilWar,” Journal of Economic Literature, 2010,

48

(1), 3—57.

Buhaug, Halvard and Jan Ketil Rod

Geography

, “Local Determinants of African CivilWars,” Political, 2006, 25 (1), 315—335.

Chandler, Tertius

NY: Edwin Mellon Press, 1987.

, Four Thousand Years of Urban Growth: An Historical Census, Lewiston,

Chen, Xi and William D. Nordhaus

Economic Statistics,” 2010. Working Paper, NBER, No. 16317.

, “The Value of Luminosity Data as a Proxy for

Ciccone, Antonio and Marek Jarocinski

Tell?,”

, “Determinants of Economic Growth: Will DataAmerican Economic Journal Macroeconomics, 2010, 2, 222—246.

Collier, Paul and Anke Hoe

Papers

ffler, “On Economic Causes of Civil War,” Oxford Economic, 1998, 50 (4), 563—573.

and

Economics, Volume 2, Defense in a Globalized World,

North Holland, 2007, pp. 711—740.

, “Civil War,” in Todd Sandler and Keith Hartley, eds., Handbook of Defense, Amsterdam and Oxford: Elsevier,

and Nicholas Sambanis

DC: World Bank, 2005.

, Understanding Civil War. Volume 1: Africa, Washington,

, Anke Hoe

Peace Research

30

ffler, and Mans Soderbom, “On the Duration of Civil War,” Journal of, 2004, 41 (3), 253—273.

Collins, David

across the Nigeria-Niger Boundary up to the 1970s,” in A.I. Asiwaju, ed.,

Africans

, “Partitioned Culture Areas and Smuggling: The Hausa and Groundnut TradePartitioned, New York, NM: St. Martin’s Press„ 1985, pp. 155—194.

Conley, Timothy G.

Econometrics

, “GMM Estimation with Cross Sectional Dependence,” Journal of, 1999, 92, 1—45.

Dell, Melissa

, “The Persistent Effects of Peru’s Mining Mita,” Econometrica, 2010, 78.

Dowden, Richard

Ltd, 2008.

, Africa: Altered States, Ordinary Miracles, London, UK: Portobello Books

Easterly, William and Ross Levine

, “Africa’s Growth Tragedy: Policies and Ethnic Divisions,”

Quarterly Journal of Economics

, 1997, 112 (4), 1203—1250.

and

mimeo Brown University.

, “The European Origins of Economic Development,” 2009. Working Paper,

Englebert, Pierre, Stacy Tarango, and Matthew Carter

A Contribution to the Debate on African Boundaries,”

Studies

, “Dismemberment and Suffocation:Comparative Political, 2002, 35 (10), 1093—1118.

Esteban, Joan, Laura Mayoral, and Debraj Ray

Investigation,” 2010. mimeo NYU Department of Economics.

, “Ethnicity and Conflict: An Empirical

Ethnologue

, Languages of the World, SIL International; Fifteenth edition 2005.

Fearon, James

Peace Research

, “Why Do Some Civil Wars Last So Much Longer Than Others?,” Journal of, 2004, 41, 275—302.

and David Laitin

Review

, “Ethnicity, Insurgency and Civil War,” American Political Science, 2003, 97, 75—90.

and

forthcoming.

, “Sons of the Soil, Migrants, and Civil War,” 2010. World Development,

Fenske, James

University.

, “Does Land Abundance Explain African Institutions,” 2010. mimeo Yale

Gennaioli, Nicola and Ilia Rainer

in Africa,” in Mark Gradstein and Kai Konrad, eds.,

Development

31

, “Precolonial Centralization and Institutional QualityInstitutions and Norms in Economic, MIT Press, 2006.

and

Economic Growth

, “The Modern Impact of Precolonial Centralization in Africa,” Journal of, 2007, 12 (3), 185—234.

Glaeser, Edward L., Rafael La Porta, Florencio Lopez de Silanes, and Andrei

Shleifer

271—303.

, “Do Institutions Cause Growth?,” Journal of Economic Growth, 2004, 9 (3),

Hargreaves, J. D.

Mc Millan, 1985.

, West Africa Partitioned, Vol. II. The Elephants adn the Grass, London:

Henderson, Vernon J., Adam Storeygard, and David N. Weil

Growth from Outer Space,” 2009. NBER Working Paper No. 15199.

, “Measuring Economic

Herbst, Je

ffrey, State and Power in Africa, Princeton, NJ: Princeton University Press, 2000.

Herge, Havard and Nicholas Sambanis

War Onset,”

, “Sensitivity Analysis of Empirical Results on CivilJournal of Conflict Resolution, 2006, 50, 508—535.

Higham, Said S.

, “The Somali Dilema: Nation in Search for State,” in A.I. Asiwaju, ed.,

Partitioned Africans

, New York, NM: St. Martin’s Press„ 1985, pp. 155—194.

Hopkins, A. G.

Press, 1973.

, An Economic History of West Africa, New York, NY: Columbia University

Huillery, Elise

in French West Africa,”

176—215.

, “History Matters: The Long Term Impact of Colonial Public InvestmentsAmerican Economic Journal - Applied Economics, 2009, 1 (2),

i Miguel, Gerard Padro

Fear,”

, “The Control of Politicians in Divided Societies: The Politics ofReview of Economic Studies, 2007, 74 (4), 1259—1274.

Iyer, Lakshmi

, “Direct versus Indirect Colonial Rule in India: Long-term Consequences,”

Review of Economics and Statistics

, 2010, forthcoming.

Kalyvas, Stathis

Science

, “Civil Wars,” in Carles Boix and Susan Stokes, eds., Handbook of Political, New York: Oxford University Press, 2007, pp. 416—434.

Kiszewski, Anthony, Andrew Mellinger, Andrew Spielman, Pia Malaney, Sonia

Ehrlich Sachs, and Je

Transmission,”

498.

32

ffrey Sachs, “A Global Index of the Stability of MalariaAmerican Journal of Tropical Medicine and Hygiene, 2004, 70 (5), 486—

Lacina, Bethany and Nils Petter Gleditsch

New Dataset of Battle Deaths,”

, “Monitoring Trends in Global Combat: AEuropean Journal of Population, 2005, 21 (2-3), 145—116.

Levine, Ross E. and David Renelt

Regressions,”

, “A Sensitivity Analysis of Cross-Country GrowthAmerican Economic Review, 1992, 84 (2), 942—963.

Martin, Xavier Sala I., Gernot Doppelhofer, and Ronald I. Miller

of Long-Run Growth: a Bayesian Averaging of Classical Estimates (BACE) Approach,”

, “Determinants

American Economic Review

, 2004, 94 (4), 813—835.

Matuszeski, Janina and Frank Schneider

Civil Con

, “Patterns of Ethnic Group Segregation andflict,” 2006. Working Paper, mimeo Harvard University.

Michalopoulos, Stelios

Tufts University.

, “The Origins of Ethnic Diversity,” 2008. Working Paper, mimeo

and Elias Papaioannou

Africa,” 2010. CEPR Discussion Paper, CEPR DP No. 6414.

, “Divide and Rule or the Rule of the Divided? Evidence from

Miguel, Edward and Michael Kremer

Health in the Presence of Treatment Externalities,”

, “Worms: Identifying Impacts on Education andEconometrica, 2004, 72, 159—217.

Miller, Douglas, Colin A.Cameron, and Jonah Gelbach

Way Clustering,” 2006. NBER Technical Working Paper No. T0327.

, “Robust Inference with Multi-

Montalvo, José G. and Marta Reynal-Querol

, “Ethnic Diversity and Economic Development,”

Journal of Development Economics

, 2005, 76, 293—323.

and

Review

, “Ethnic Polarization, Potential Conflict and Civil War,” American Economic, 2005, 95, 796—816.

Murdock, George P.

Hill Book Company, 1959.

,

, Africa: Its Peoples and Their Culture History, New York: McGraw-Ethnographic Atlas, Pittsburgh, PA: University of Pittsburgh Press, 1967.

Naritomi, Joana, Rodrigo R. Soares, and Juliano J. Assunção

and Colonial Heritage within Brazil,” 2009. Working Paper, IZA DP No. 4276.

, “Institutional Development

Nunn, Nathan

Economics

33

, “The Long Term Effects of Africa’s Slave Trades,” Quarterly Journal of, 2008, 123 (1), 139—176.

and Diego Puga

Economics and Statistics, forthcoming

, “Ruggedness: The Blessing of Bad Geography in Africa,” Review of, 2011.

and Leonard Wantchekon

, “The Slave Trade and the Origins of Mistrust in Africa,”

American Economic Review, forthcoming

, 2011.

Petter, Gleditsch Nils, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg,

and Havard Strand

Research

, “Armed Conflict 1946-2001: A New Datase,” Journal of Peace, 2002, 39 (5), 615—637.

Raleigh, Clionadh, David Cunningham, Lars Wilhelmsen, and Nils Petter Gleditsch

“Con

,flict Sites: 1946-2005. v2.0,” 2006. Centre for the Study of Civil War, PRIO.

Renner, F.A.

Asiwaju, ed.,

, “Ethnic Affinity: Partition and Political Integration in Senegambia,” in A.I.Partitioned Africans, New York, NM: St. Martin’s Press„ 1985, pp. 71—86.

Ross, Michael L.

Political Science

, “A Closer Look on Oil, Diamonds, and Civil War,” Anuual Review of, 2006.

Salehyan, Idean and Kristian Skrede Gleditsch

, “Refugees and the Spread of Civil War,”

International Organization

, 2006, 60 (02), 335—366.

Silva, J.M.C Santos and Silvana Tenreyro

and Statistics

, “The Log of Gravity,” Review of Economics, 2006, 88 (4), 641—658.

, , and FrankWindmeijer

for Non-Negative Data with Many Zeros,” 2010. CeMMAP working papers CWP20/10.

, “Is itDifferent for Zeros? Discriminating Between Models

Wesseling, H. L.

Praeger Publishers, 1996.

, Divide and Rule: The Partition of Africa, 1880-1914, Westport, Conn:

Wimmer, Andreas, Lars-Erik Cederman, and Brian Min.

Con

Review

, “Ethnic politics and Armedflict. A Configurational Analysis of a New Global Dataset,” American Sociological, 2009, 74, 316—337.

Wooldridge, Je

2002.

34

ffrey M., Econometric Analysis of Cross Section and Panel Data, MIT Press,

Obs. mean st. dev. min p25 median p75 max

Number of Civil Conflicts 834 1.14 0.95 0.00 0.00 1.00 2.00 5.00

Number of Civil War Conflict Zones 834 1.58 1.52 0.00 0.00 1.00 2.00 8.00

Civil Conflict Casualties (Best Estimate) 834 39.62 82.25 0.00 0.00 3.21 40.12 589.53

Civil Conflict Casualties (Low Estimate) 834 33.57 77.59 0.00 0.00 1.90 40.12 561.34

Civil Conflict Casualties (High Estimate) 834 45.23 91.47 0.00 0.00 4.32 40.68 639.37

Civil Conflict Duration 834 7.75 9.39 0.00 0.00 3.86 11.63 46.79

Light Density (Development) 834 0.46 1.83 0.00 0.00 0.03 0.17 25.41

Land Area 834 34.06 58.97 0.24 6.13 14.44 35.94 604.90

Land Area under Water 834 0.86 2.25 0.00 0.01 0.17 0.66 27.66

Latitude 834 4.46 13.11 -33.09 -2.92 6.59 11.08 36.58

Longitude 834 17.85 15.47 -16.41 7.32 18.40 31.22 49.25

Population Density at Independence 834 17.47 25.71 0.00 3.17 9.26 20.47 321.53

Mean Elevation 834 0.62 0.43 0.00 0.30 0.49 0.93 2.17

Land Suitability for Agriculture 834 0.41 0.24 0.00 0.25 0.42 0.57 0.98

Malaria Stability Index 834 0.75 0.36 0.00 0.57 0.98 1.00 1.00

Average Distance to the Seacoast 834 600.76 435.52 1.49 209.33 552.27 924.67 1697.01

Diamond Mine Indicator 834 0.12 0.33 0.00 0.00 0.00 0.00 1.00

Oil Deposit Indicator 834 0.12 0.40 0.00 0.00 0.00 0.00 4.00

Capital City Indicator 834 0.06 0.23 0.00 0.00 0.00 0.00 1.00

Major City in 1400 AD Inidcator 834 0.04 0.20 0.00 0.00 0.00 0.00 2.00

Slave Exports 834 551.70 3289.21 0.00 0.00 0.00 17.67 41045.08

Pre-colonial Political Centralization 440 0.34 0.48 0.00 0.00 0.00 1.00 1.00

Pre-colonial Class Stratification 396 0.54 0.50 0.00 0.00 1.00 1.00 1.00

Pre-colonial Property Rights Indicator 375 0.93 0.25 0.00 1.00 1.00 1.00 1.00

Pre-colonial Share of Agriculture 490 5.63 1.79 0.00 5.00 6.00 7.00 9.00

Table 1A : Summary Statistics

The table reports descriptive statistics for all variables employed in the empirical analysis. The Data Appendix gives detailed variable

definitions and data sources.

Panel A: Outcome Measures

Panel B: Control Variables

Panel C: Pre-colonial Ethnic-Specific Variables

Civil Conflict Casualties (Best Estimate)

1

Civil Conflict Casualties (Low Estimate)

0.9780* 1

Civil Conflict Casualties (High Estimate)

0.9787* 0.9344* 1

Number of Civil Conflicts

0.3669* 0.3383* 0.3926* 1

Number of Civil War Conflict Zones

0.5851* 0.5560* 0.5900* 0.8484* 1

Civil Conflict Duration

0.5445* 0.5081* 0.5392* 0.6411* 0.7101* 1

Light Density (Development)

‐0.0494 ‐0.0667 ‐0.0475 ‐0.0631 ‐0.0743* ‐0.0819* 1

Table 1B : Correlation Structure Dependent Variables

The table reports the correlation structure among the outcome measures (civil conflict and economic development). * indicates

statistical significance at the 95% confidence level. The Data Appendix gives detailed variable definitions and data sources.

Non-Border Groups

(1)

mean mean difference mean difference

Number of Civil Conflicts 0.968 1.315 0.346 1.407 0.438

(0.038) (0.093) (0.036) (0.067) (0.072)

0.00 0.00

Number of Civil War Conflict Zones 1.2710 1.9921 0.7211 1.9827 0.7117

(0.0575) (0.1697) (0.1792) (0.1064) (0.1210)

0.00 0.00

Civil Conflict Casualties 29.702 57.383 27.682 50.276 20.574

(Best Estimate) (62.115) (10.175) (10.566) (94.004) (6.809)

0.01 0.00

Civil Conflict Casualties 26.006 49.374 23.368 40.458 14.452

(Low Estimate) (58.237) (9.805) (10.162) (88.344) (6.396)

0.03 0.02

Civil Conflict Casualties 32.532 64.107 31.575 61.005 28.474

(High Estimate) (67.244) (10.814) (11.244) (109.771) (7.853)

0.01 0.00

Civil Conflict Duration 7.229 8.705 1.476 8.290 1.061

9.346 (0.864) (0.964) (9.245) (0.744)

0.13 0.15

Light Density (Development) 0.611 0.228 -0.383 0.273 -0.337

(2.319) (0.057) (0.121) (0.800) (0.119)

0.00 0.00

Observations 476 127 231

Table 1C: Differences in Outcomes between Partitioned and Non-Split Ethnic Groups

Border Groups Partitioned Groups

The table reports summary statistics and test of means for the outcome measures (civil conflict and economic development). Column

(1) gives summary statistics for non-border ethnic groups. Column (2) gives summary statistics for ethnic groups whose historical

homeland is adjacent to the national border, but is not significantly partitioned. Column (3) gives summary statistics for partitioned

ethnic groups, where at least 10% of the historical homeland falls into more than one contemporary country. For each category of

ethnicities we report the mean and the standard deviation in parentheses. The number of observations is reported in italics. The table

also reports the difference, the standard error of the difference, and the corresponding p-value of a test of mean equality (assuming

unequal variances) between the non-border groups and the other two categories. The Data Appendix gives detailed variable

definitions and data sources.

(2) (3)

SPLIT BROAD SPLIT BROAD SPLIT BROAD SPLIT BROAD

(1) (2) (3) (4) (5) (6) (7) (8)

Land Area 0.1256** 0.2243*** 0.1149** 0.2462*** 0.0998* 0.2204*** 0.0931* 0.2194***

(0.0600) (0.0654) (0.0526) (0.0584) (0.0502) (0.0597) (0.0560) (0.0639)

Land Area under Water 0.3910*** 0.4878*** 0.3711*** 0.4778*** 0.3908*** 0.4888*** 0.3925*** 0.4891***

(0.0803) (0.1340) (0.0808) (0.1274) (0.0845) (0.1295) (0.0861) (0.1311)

Elevation -0.0610 -0.0044 -0.0608 0.0676 -0.0933 0.0398 -0.0814 0.0418

(0.2443) (0.2462) (0.2557) (0.2736) (0.2582) (0.2812) (0.2522) (0.2727)

Suitability for Agriculture 0.3956 0.3250 0.5867 0.4169 0.6138 0.4271 0.6399 0.4313

(0.3802) (0.3288) (0.3954) (0.3816) (0.3860) (0.3657) (0.3868) (0.3573)

Malaria Stability Index 0.2504 0.2705 -0.1065 0.2044 -0.1225 0.1827 -0.1274 0.1817

(0.2514) (0.2593) (0.3583) (0.3714) (0.3591) (0.3751) (0.3549) (0.3718)

Distance to the Seacoast -0.0001 -0.0001 -0.0001 -0.0001 -0.0001 -0.0001 -0.0001 -0.0001

(0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002) (0.0002)

Diamond 0.1509 0.2523 0.1527 0.2526

(0.1740) (0.1681) (0.1758) (0.1699)

Oil -0.0689 0.0192 -0.0639 0.0202

(0.1679) (0.1476) (0.1685) (0.1491)

Major City in 1400AD -0.0089 -0.3583 -0.0065 -0.3569

(0.2165) (0.2439) (0.2166) (0.2447)

Slave Exports 0.0122 -0.007 0.013 -0.007

(0.0310) (0.0336) (0.0306) (0.0333)

Population Density 1960 -0.0228 -0.0038

(0.0713) (0.0717)

Region Fixed-Effects No No Yes Yes Yes Yes Yes Yes

Log Likelhood -472.817 -527.644 -463.695 -518.69 -463.057 -517.348 -462.961 -517.345

pseudo R-squared 0.039 0.074 0.058 0.090 0.059 0.092 0.059 0.092

Observations 834 834 834 834 834 834 834 834

Table 2 - Are African Borders Artificial?

The table reports probit marginal effects associating whether a group is partitioned with ethnic-specific features. The dependent variable is

a dummy variable that equals one when an ethnicity is partitioned by the national border. In odd numbered columns partitioned ethnicities

are identified as those with at least 10% of the historical homeland belonging to more than one contemporary countries (SPLIT=1). In even

numbered columns we broaden the definition and identify partitioned ethnicities with those impacted by the national borders irrespective of

the degree of partitioning (BROAD=1). The specifications in columns (3)-(8) include a set of region fixed-effects (constants not reported).

The Data Appendix gives detailed variable definitions and data sources. Standard errors reported in parentheses are adjusted for double

clustering at the country-dimension and the ethno-linguistic family dimension. ***, **, and * indicate statistical significance at the 1%, 5%,

and 10% level respectively.

SPLIT BROAD SPLIT BROAD SPLIT BROAD SPLIT BROAD SPLIT BROAD

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Land Area 0.1672** 0.2997*** 0.1788** 0.2465*** 0.1300* 0.2552*** 0.1721** 0.2834*** 0.1230 0.2583**

(0.0609) (0.0805) (0.0691) (0.0900) (0.0714) (0.0889) (0.0598) (0.0722) (0.0890) (0.1062)

Land Area under Water 0.2452** 0.3634*** 0.3274*** 0.4526*** 0.3047*** 0.4934*** 0.2803** 0.4140*** 0.4732*** 0.6136***

(0.0976) (0.1243) (0.0944) (0.1426) (0.1087) (0.1525) (0.0952) (0.1290) (0.1514) (0.1956)

Political Centralization -0.1865 -0.2367 0.1023 -0.1377

(0.1668) (0.1679) (0.2246) (0.2167)

Class Stratification -0.2433 -0.0866 -0.2814 -0.0565

(0.2433) (0.0866) (0.2165) (0.1845)

Property Rights 0.0671 0.2392 -0.0245 0.1224

(0.2675) (0.2746) (0.3685) (0.3065)

Dependence on Agriculture 0.0280 0.0195 0.0354 0.0247

(0.0328) (0.0340) (0.0474) (0.0421)

Region Fixed-Effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Log Likelhood -251.81 -267.26 -216.82 -238.48 -207.39 -225.26 -274.00 -296.36 -165.34 -175.90

pseudo R-squared 0.07 0.12 0.08 0.12 0.07 0.13 0.07 0.12 0.09 0.14

Observations 440 440 396 396 375 375 490 490 295 295

Table 3 - Do Ethnic Characteristics Predict Partitioning?

The table reports probit marginal effects associating whether a group is partitioned with pre-colonial ethnic-specific features, reflecting political centralization (in (1), (2), (9), and

(10)), class stratification (in (3), (4), (9), and (10)), property rights protection (in (5), (6), (9), and (10)), and historical economic development reflected in the dependence on

agriculture (in (7)-(10)). The dependent variable is a dummy variable that equals one when an ethnicity is partitioned by the national border. In odd numbered columns partitioned

ethnicities are identified as those with at least 10% of the historical homeland belonging to more than one contemporary countries (SPLIT=1). In even numbered columns we

broaden the definition and identify partitioned ethnicities with those impacted by the national borders irrespective of the degree of partitioning (BROAD=1). All specifications

include a set of region fixed-effects (constants not reported), log land area, and log land area under water (lakes, rivers, and other streams).

The Data Appendix gives detailed variable definitions and data sources. Standard errors reported in parentheses are adjusted for double clustering at the country-dimension and the

ethno-linguistic family dimension. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level respectively.

(1) (2) (3) (4) (5)

SPLIT - Partitioning (>10%) 0.3316*** 0.3213*** 0.3139*** 0.2750*** 0.2336***

(0.0752) (0.0649) (0.0621) (0.0537) (0.0526)

BORDER - Partitioning (0% - 10%) 0.2255** 0.1826** 0.1368* 0.0914 0.0980

(0.0919) (0.0786) (0.0758) (0.0627) (0.0597)

Log Likelihood -1034.05 -995.85 -965.83 -855.41 -825.33

R-squared 0.222 0.316 0.406 0.670 0.738

SPLIT - Partitioning (>10%) 0.4194*** 0.4212*** 0.4128*** 0.2872*** 0.2573***

(0.0981) (0.0815) (0.0799) (0.0545) (0.0538)

BORDER - Partitioning (0% - 10%) 0.3864*** 0.3720*** 0.3175*** 0.1038* 0.1449***

(0.1317) (0.1070) (0.1023) (0.0616) (0.0538)

Log Likelihood -1269.93 -1218.97 -1164.74 -969.33 -925.11

R-squared 0.269 0.320 0.435 0.752 0.808

Region Fixed-Effects Yes Yes Yes Yes No

Geography Yes Yes Yes Yes Yes

RD Polynomial No Yes Yes Yes Yes

Additional Controls No No Yes Yes Yes

Ethnic Family Fixed-Effects No No No Yes Yes

Country Fixed-Effects No No No No Yes

Observations 834 834 834 834 834

The table reports Poisson ML estimates associating civil war with partitioning and other ethnicity-specific measures. In Panel A the

dependent variable is the number of civil wars that have taken place in the historical homeland of an ethnic group between 1970 and

2005. In Panel B the dependent variable is the number of conflict zones associated with the civil wars that have affected the

historical homeland of an ethnic group during the period 1970-2005. SPLIT is an indicator variable that identifies partitioning

ethnicities as those with at least 10% of the historical homeland belonging to more than one contemporary country. BORDER is an

indicator that identifies ethnic groups residing by the border. These groups also fall into more than one country but more than 90% of

the historical homeland lies in one country. All specifications include a set of region fixed-effects (constants not reported), log land

area, log land area under water (lakes, rivers, and other streams), and population density around independence (in 1960).

Columns (2)-(6) include a regression discontinuity (RD) cubic polynomial in latitude and longitude of the centroid of each ethnic

group. Columns (3)-(6) include a rich set of control variables (land suitability for agriculture, elevation, a malaria stability index, an

indicator of early development that equals one when a major city was in the ethnicity’s historical homeland in 1400, an oil indicator

and a diamond mine indicator). Columns (4)-(5) include a set of ethnic-family fixed-effects (constants not reported). Column (5)

includes a set of country fixed-effects (constants not reported). The Data Appendix gives detailed variable definitions and data

sources. Standard errors reported in parentheses are adjusted for double clustering at the country-dimension and the ethno-linguistic

family dimension. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level respectively.

Panel B - Dependent Variable: Number of Conflict Zones

Table 4: Partitioning and the Incidence of Civil Conflict (Poisson ML Estimates)

Panel A - Dependent Variable: Number of Civil Conflicts

(1) (2) (3) (4) (5) (6) (7) (8)

SPLIT 0.6072*** 0.9134*** 0.6834*** 0.6160*** 0.5858*** 0.6102*** 0.5791*** 0.6228**

Partitioning (>10%) (0.2006) (0.2373) (0.1621) (0.1896) (0.1914) (0.1775) (0.1861) (0.1786)

BORDER 0.5550*** 0.6173*** 0.3893** 0.2032 0.3689*** 0.3134*** 0.2888*** 0.3228**

Partitioning (0% - 10%) (0.1403) (0.1170) (0.1539) (0.1460) (0.1316) (0.1389) (0.1358) (0.1410)

Country Casualties 0.0001*** 0.0001*** 0.0001***

(0.0000) (0.0000) (0.0000)

Ethnic Family Casualties 0.0001*** 0.0001*** 0.0001***

(0.0000) (0.0000) (0.0000)

Log Likelihood -3192.98 -3108.18 -2916.83 -2677.24 -2745.51 -2556.05 -2343.46 -2667.39

SPLIT 0.6053** 0.7638*** 0.6779*** 0.4768*** 0.3458** 0.3558** 0.3209** 0.3905**

Partitioning (>10%) (0.2488) (0.1674) (0.1537) (0.1618) (0.1512) (0.1648) (0.1525) (0.1684)

BORDER 0.4836** 0.4191*** 0.3418*** 0.1133 0.1935* 0.1527 0.1352 0.1695

Partitioning (0% - 10%) (0.2430) (0.1246) (0.1072) (0.1678) (0.1024) (0.1375) (0.1248) (0.1461)

Country Casualties 0.3283*** 0.2460*** 0.1686***

(0.0682) (0.0617) (0.0570)

Ethnic Family Casualties 0.2362*** 0.1768*** 0.1045***

(0.0588) (0.0538) (0.0366)

adjusted R-squared 0.221 0.529 0.618 0.769 0.782 0.840 0.852 0.833

Region Fixed-Effects Yes Yes Yes Yes Yes No No No

Geography Yes Yes Yes Yes Yes Yes Yes Yes

RD Polynomial No No Yes Yes Yes Yes Yes Yes

Additional Controls No No Yes Yes Yes Yes Yes Yes

Ethnic Family Fixed-Effe No No No Yes No Yes Yes Yes

Country Fixed-Effects No No No No Yes Yes Yes Yes

Observations 834 834 834 834 834 834 834 834

Table 5 - Partitioning and Civil War Casualties

Panel A: Negative Binomial ML

Panel B: Log Linear OLS

Table 5 - Notes

***, **, and * indicate statistical significance at the 1%, 5%, and 10% level respectively.

Panels A and B report Negative Binomial ML and OLS estimates respectively, associating civil conflict casualties with ethnic partitioning.

The dependent variable is the number of casualties from civil conflict between 1970 and 2005 that have occurred in the historical

homeland of an ethnic group divided by the respective tribal area. In the OLS specifications the dependent variable is the log of one plus

the number of casualties per thousand of square kilometers, while in the Negative Binomial ML specifications the casualty rate is

expressed in levels. In columns (1)-(6) we use the best estimate for casualties from the PRIO database, while in column (7) and (8) we use

the high and the low war casualties estimates, respectively. SPLIT is an indicator variable that identifies partitioning ethnicities as those

with at least 10% of the historical homeland belonging to more than one contemporary country. BORDER is an indicator that identifies

ethnic groups residing by the border. These groups also fall into more than one country but more than 90% of the historical homeland lies

in one country.

In columns (2)-(5) we control for spatial spillovers including the number of civil war casualties per thousand kilometers at the countrylevel

and the ethnic-family level (excluding ethnicity i), respectively. All specifications include a set of region fixed-effects, log land area,

log land area under water (lakes, rivers, and other streams), and population density around independence (in 1960). Columns (3)-(8)

include a regression discontinuity (RD) cubic polynomial in latitude and longitude of the centroid of each ethnic group. Columns (4)-(8)

include a rich set of control variables (land suitability for agriculture, elevation, malaria stability index, an early development indicator

whether a major city was in the ethnicity’s historical homeland in 1400, an oil indicator and a diamond mine indicator). Columns (4), (6),

(7), and (8) include a set of ethnic-family fixed-effects (constants not reported). Columns (5)-(8) include a set of country fixed-effects

(constants not reported). The Data Appendix gives detailed variable definitions and data sources. Standard errors reported in parentheses

are adjusted for double clustering at the country-dimension and the ethno-linguistic family dimension.

(1) (2) (3) (4) (5) (6)

SPLIT 0.3206* 0.3807** 0.3837** 0.3129** 0.4225** 0.3055**

Partitioning (>10%) (0.1787) (0.1702) (0.1527) (0.1265) (0.1814) (0.1453)

BORDER 0.1958 0.1894* 0.1601** 0.1737** 0.1600** 0.1729*

Partitioning (0% - 10%) (0.1478) (0.1064) (0.0758) (0.0862) (0.0797) (0.0935)

Total Duration at Country 0.0912*** 0.0763*** 0.0571***

(0.0197) (0.0099) (0.0159)

Total Duration at Ethnic Family 0.0834*** 0.0696*** 0.0558***

(0.0155) (0.0117) (0.0121)

Log Likelihood -2377.90 -2195.90 -2088.96 -1884.05 -1972.15 -1758.84

SPLIT 0.3039** 0.2724*** 0.2927*** 0.2763*** 0.2673*** 0.2585**

Partitioning (>10%) (0.1258) (0.0819) (0.0756) (0.0913) (0.1001) (0.1048)

BORDER 0.2240 0.0912* 0.0834 0.0721 0.0523 0.0909

Partitioning (0% - 10%) (0.1377) (0.0485) (0.0520) (0.0904) (0.0615) (0.0898)

Total Duration at Country 0.5355*** 0.5383*** 0.4484***

(0.0770) (0.0722) (0.0997)

Total Duration at Ethnic Family 0.6210*** 0.5240*** 0.4251***

(0.0575) (0.0517) (0.0736)

adjusted R-squared 0.235 0.678 0.717 0.794 0.791 0.847

Region Fixed-Effects Yes Yes Yes Yes No No

Geography Yes Yes Yes Yes Yes Yes

RD Polynomial No No Yes Yes Yes Yes

Additional Controls No No Yes Yes Yes Yes

Ethnic Family Fixed-Effects No No No Yes No Yes

Country Fixed-Effects No No No No Yes Yes

Observations 834 834 834 834 834 834

Table 6 - Partitioning and Civil War Duration

Panel A: Negative Binomial ML

Panel B: Log Linear OLS

Table 6 - Notes

Panels A and B report Negative Binomial ML and OLS estimates respectively, associating civil war duration with ethnic

partitioning. The dependent variable is the total number of years that an area of an ethnic group has been under civil conflict over

the period 1970-2005. In the OLS specifications the dependent variable is the log of one plus the number of number of years under

conflict, while in the Negative Binomial ML civil war duration is expressed in levels. In columns (2)-(5) we control for spatial

spillovers including the civil war duration at the country-level and the ethnic-family level (excluding ethnicity i), respectively.

SPLIT is an indicator variable that identifies partitioning ethnicities as those with at least 10% of the historical homeland belonging

to more than one contemporary country. BORDER is an indicator that identifies ethnic groups residing by the border. These groups

also fall into more than one country but more than 90% of the historical homeland lies in one country.

All specifications include a set of region fixed-effects, log land area, log land area under water (lakes, rivers, and other streams),

and population density around independence (in 1960). Columns (3)-(6) include a regression discontinuity (RD) cubic polynomial

in latitude and longitude of the centroid of each ethnic group. Columns (4)-(6) include a rich set of control variables (land suitability

for agriculture, elevation, malaria stability index, an early development indicator whether a major city was in the ethnicity’s

historical homeland in 1400, an oil indicator and a diamond mine indicator). Columns (4) and (6) include a set of ethnic-family

fixed-effects (constants not reported). Columns (5) and (6) include a set of country fixed-effects (constants not reported). The Data

Appendix gives detailed variable definitions and data sources. Standard errors reported in parentheses are adjusted for double

clustering at the country-dimension and the ethno-linguistic family dimension. ***, **, and * indicate statistical significance at the

1%, 5%, and 10% level respectively.

(1) (2) (3) (4) (5) (6) (7)

SPLIT -0.7827*** .3854***

BORDER -0.7854*** -0.4533** -0.3907** -0.1334 -0.0959 -0.3124*** -0.2033*

Log Likelihood -478.964 -438.123 -393.55 -339.236 -321.435 -284.317 -309.682

SPLIT -0.1048*** -0.0422

Partitioning (>10%) (0.0375) (0.0337) (0.0309) (0.0283) (0.0299) (0.0241) (0.0271)

Partitioning (0% - 10%) (0.0305) (0.0330) (0.0276) (0.0283) (0.0261) (0.0230) (0.0239)

RD Polynomial No Yes Yes Yes Yes Yes Yes

90% of

p

1%.

r double

at the 1%,

-0.7018*** -0.8353*** -0.6913*** -0.4856*** -0.4000*** -0

Partitioning (>10%) (0.6591) (0.1737) (0.1807) (0.1275) (0.1590) (0.1060) (0.1198)

Partitioning (0% - 10%) (0.2242) (0.1821) (0.2050) (0.1365) (0.1345) (0.0524) (0.0849)

-0.0677** -0.0756** -0.0604** -0.0536* -0.0306

BORDER -0.1168*** -0.0825** -0.0709** -0.0674** -0.0665** -0.0585** -0.0616***

adjusted R-squared 0.399 0.478 0.563 0.730 0.764 0.751 0.785

Region Fixed-Effects Yes Yes Yes Yes No No No

Geography Yes Yes Yes Yes Yes Yes Yes

Additional Controls No No Yes Yes Yes Yes Yes

Ethnic Family Fixed-Effects No No No Yes Yes Yes Yes

Country Fixed-Effects No No No No Yes Yes Yes

Observations 834 834 834 834 834 825 834

Panels A and B report Negative Binomial ML and OLS estimates respectively associating regional development, as reflected in satellite

light density at night, with ethnic partitioning. In the OLS specifications the dependent variable is the log of one plus light density, while

in the Negative Binomial ML satellite light density is expressed in levels. SPLIT is an indicator variable that identifies partitioning

ethnicities as those with at least 10% of the historical homeland belonging to more than one contemporary country. BORDER is an

indicator that identifies ethnic groups residing by the border. These groups also fall into more than one country but more than

the historical homeland lies in one country.

All specifications include a set of region fixed-effects, log land area, log land area under water (lakes, rivers, and other streams), and

population density around independence (in 1960). Columns (2)-(7) include a regression discontinuity (RD) cubic polynomial in

latitude and longitude of the centroid of each ethnic group. Columns (3)-(7) include a rich set of control variables (land suitability for

agriculture, elevation, malaria stability index, an early development indicator whether a major city was in the ethnicity’s historical

homeland in 1400, an oil indicator and a diamond mine indicator). Columns (4)-(7) include a set of ethnic-family fixed-effects

(constants not reported). Columns (5)-(7) include a set of country fixed-effects (constants not reported). In column (6) we dro

observations at the top 1% of the distribution on light density, while in column (7) we winsorise the dependent variable at the

The Data Appendix gives detailed variable definitions and data sources. Standard errors reported in parentheses are adjusted fo

clustering at the country-dimension and the ethno-linguistic family dimension. ***, **, and * indicate statistical significance

5%, and 10% level respectively.

Table 7 - Partitioning and Development (Satellite Light Density)

Panel A: Negative Binomial ML

Panel B: Log Linear OLS

(1) (2) (3) (4) (5) (6)

SPLIT 0.5243*** 0.5029*** 0.2677*** 0.2813*** -0.0942*** -0.0647***

Partitioning (>10%) (0.1114) (0.1025) (0.0754) (0.0658) (0.0270) (0.0240)

BORDER

0.3702*** 0.2731** 0.1564* 0.1188 -0.0951*** -0.0599**

Partitioning (0% - 10%) (0.1351) (0.1233) (0.0919) (0.0792) (0.0329) (0.0289)

rho (spatial lag) 1.658 1.642 1.659 1.645 1.603 1.575

(0.015) (0.031) (0.015) (0.028) (0.070) (0.097)

chi2 447.17 226.47 343.96 237.90 106.54 82.52

Log Likelihood -1406.76 -1318.38 -1085.07 -949.07 -227.30 -105.94

Region Fixed-Effects Yes Yes Yes Yes Yes Yes

Geography Yes Yes Yes Yes Yes Yes

RD Polynomial No Yes No Yes No Yes

Additional Controls No Yes No Yes No Yes

Observations 834 834 834 834 834 834

The table reports general spatial lag specifications, associating partitioning with civil war intensity (in (1)-(2)), civil war

duration (in (3)-(4)), and regional development as proxied by satellite light density (in (5)-(6)). The spatial lag is decaying

linearly in the distance from the centroid of each ethnic group. Estimation is with maximum likelihood. The table also reports

the coefficient and the corresponding standard error of the spatial lag (rho). SPLIT is an indicator variable that identifies

partitioning ethnicities as those with at least 10% of the historical homeland belonging to more than one contemporary country.

BORDER is an indicator that identifies ethnic groups residing by the border. These groups also fall into more than one country

but more than 90% of the historical homeland lies in one country.

All specifications include a set of region fixed-effects, log land area, log land area under water (lakes, rivers, and other streams),

and population density around independence (in 1960). Columns (2), (4), and (6) include a regression discontinuity (RD) cubic

polynomial in latitude and longitude of the centroid of each ethnic group and a rich set of control variables (land suitability for

agriculture, elevation, a malaria stability index, an early development indicator whether a major city was in the ethnicity’s

historical homeland in 1400, an oil indicator and a diamond mine indicator). The Data Appendix gives detailed variable

definitions and data sources. Standard errors reported in parentheses are adjusted for spatial correlation. ***, **, and * indicate

statistical significance at the 1%, 5%, and 10% level respectively.

Appendix Table A - Sensitivity Analysis

Spatial Lag Specifications

Casualties Duration Development

(1) (2) (3) (4) (5) (6)

SPLIT 0.6483*** 0.4832*** 0.4338*** 0.2319** -0.9342*** -0.4144***

Partitioning (>10%) (0.1762) (0.1300) (0.1495) (0.1019) (0.2080) (0.1359)

BORDER 0.4699*** 0.3270** 0.2374 0.1786** -0.5529*** -0.2811**

Partitioning (0% - 10%) (0.1642) (0.1546) (0.1527) (0.0815) (0.1706) (0.0734)

Log Likelihood -2700.54 -2331.48 -2051.15 -1582.62 -294.75 -234.46

Observations 770 770 770 770 770 770

SPLIT 0.6496*** 0.6957*** 0.3904*** 0.3530** -0.9189*** -0.4100**

Partitioning (>10%) (0.1391) (0.1869) (0.1462) (0.1545) (0.2185) (0.1689)

BORDER 0.3539** 0.2939*** 0.1369 0.1856*** -0.3840** -0.0684

Partitioning (0% - 10%) (0.1720) (0.1313) (0.1165) (0.0794) (0.2314) (0.1481)

Log Likelihood -2516.23 -2170.84 -1839.07 -1479.81 -329.33 -270.56

Observations 747 747 747 747 747 747

Region Fixed-Effects Yes No Yes No Yes No

Geography Yes Yes Yes Yes Yes Yes

RD Polynomial Yes Yes Yes Yes Yes Yes

Additional Controls Yes Yes Yes Yes Yes Yes

Country Fixed-Effects No Yes No Yes No Yes

Ethnic Family Fixed-Effects No Yes No Yes No Yes

The table reports negative binomial ML specifications, associating partitioning with civil war intensity (in (1)-(2)), civil war duration (in

(3)-(4)), and regional development as proxied by satellite light density (in (5)-(6)). In Panel A we exclude ethnic areas in Northern

Africa, while in Panel B we exclude ethnic areas in Southern Africa. SPLIT is an indicator variable that identifies partitioning ethnicities

as those with at least 10% of the historical homeland belonging to more than one contemporary country. BORDER is an indicator that

identifies ethnic groups residing by the border. These groups also fall into more than one country but more than 90% of the historical

homeland lies in one country.

All specifications include log land area, log land area under water (lakes, rivers, and other streams), population density around

independence (in 1960), a regression discontinuity (RD) cubic polynomial in latitude and longitude of the centroid of each ethnic group,

land suitability for agriculture, elevation, malaria stability index, an early development indicator whether a major city was in the

ethnicity’s historical homeland in 1400, an oil indicator and a diamond mine indicator. Columns (1), (3), and (5) include a set of region

fixed-effects (constants not reported). Columns (2), (4), and (6) include a set of ethnic-family fixed-effects and a set of country fixedeffects

(constants not reported). The Data Appendix gives detailed variable definitions and data sources. Standard errors reported in

parentheses are adjusted for double clustering at the country-dimension and the ethno-linguistic family dimension. ***, **, and *

indicate statistical significance at the 1%, 5%, and 10% level respectively.

Panel A: Excluding Northern Africa Countries

Panel B: Excluding Southern Africa Countries

Appendix Table B - Sensitivity Analysis

Excluding Northern African and Southern African Countries

Casualties Duration Development

(1) (2) (3) (4) (5) (6)

FRACT 1.1693** 1.0899** 0.7731** 0.6302** -1.5200*** -1.2020***

Continuous Partitioning (0.4901) (0.4176) (0.3290) (0.3008) (0.5043) (0.3930)

Log Likelihood -2963.87 -2562.32 -2240.25 -1761.59 -406.13 -321.32

Observations 834 834 834 834 834 834

Region Fixed-Effects Yes No Yes No Yes No

Geography Yes Yes Yes Yes Yes Yes

RD Polynomial Yes Yes Yes Yes Yes Yes

Additional Controls Yes Yes Yes Yes Yes Yes

Country Fixed-Effects No Yes No Yes No Yes

Ethnic Family Fixed-Effects No Yes No Yes No Yes

The table reports negative binomial ML specifications, associating a contninuous measure of ethnic partitioning with civil war intensity

(in (1)-(2)), civil war duration (in (3)-(4)), and regional development as proxied by satellite light density (in (5)-(6)).

All specifications include log land area, log land area under water (lakes, rivers, and other streams), population density around

independence (in 1960), a regression discontinuity (RD) cubic polynomial in latitude and longitude of the centroid of each ethnic group,

land suitability for agriculture, elevation, malaria stability index, an early development indicator whether a major city was in the

ethnicity’s historical homeland in 1400, an oil indicator and a diamond mine indicator. Columns (1), (3), and (5) include a set of region

fixed-effects (constants not reported). Columns (2), (4), and (6) include a set of ethnic-family fixed-effects and a set of country fixedeffects

(constants not reported). The Data Appendix gives detailed variable definitions and data sources. Standard errors reported in

parentheses are adjusted for double clustering at the country-dimension and the ethno-linguistic family dimension. ***, **, and *

indicate statistical significance at the 1%, 5%, and 10% level respectively.

Appendix Table C - Sensitivity Analysis

Employing a Continuous Measure of Ethnic Partitioning

Casualties Duration Development

Ethnicity Name

% of Initial

Homeland Country

# of

Partitions Ethnicity Name

% of Initial

Homeland Country

# of

Partitions

ABABDA 0.72 EGY 2 LAKA (ADAMAWA) 0.69 TCD 3

ABABDA 0.28 SDN 2 LAKA (ADAMAWA) 0.20 CMR 3

ADELE 0.48 GHA 2 LAKA (ADAMAWA) 0.11 CAF 3

ADELE 0.52 TGO 2 LAMBA 0.39 ZAR 2

AFAR 0.17 DJI 3 LAMBA 0.61 ZMB 2

AFAR 0.22 ERI 3 LAMBYA 0.17 MWI 3

AFAR 0.61 ETH 3 LAMBYA 0.33 TZA 3

ALUR 0.16 ZAR 2 LAMBYA 0.50 ZMB 3

ALUR 0.84 UGA 2 LIGBI, DEGHA (SE) 0.72 GHA 2

AMBA 0.87 ZAR 2 LIGBI, DEGHA (SE) 0.28 CIV 2

AMBA 0.13 UGA 2 LOBI 0.42 CIV 2

AMBO 0.41 AGO 2 LOBI 0.58 BFA 2

AMBO 0.59 NAM 2 LUGBARA 0.45 ZAR 3

AMER 0.56 ERI 2 LUGBARA 0.04 SDN 3

AMER 0.44 SDN 2 LUGBARA 0.51 UGA 3

ANA 0.33 BEN 2 LUNGU 0.31 TZA 2

ANA 0.67 TGO 2 LUNGU 0.69 ZMB 2

ANUAK 0.75 ETH 2 LUVALE 0.81 AGO 3

ANUAK 0.25 SDN 2 LUVALE 0.01 ZAR 3

ANYI 0.42 GHA 2 LUVALE 0.17 ZMB 3

ANYI 0.58 CIV 2 MADI 0.42 SDN 2

ASBEN 0.89 NER 2 MADI 0.58 UGA 2

ASBEN 0.11 DZA 2 MAKONDE 0.56 MOZ 2

ASSINI 0.51 GHA 2 MAKONDE 0.44 TZA 2

ASSINI 0.49 CIV 2 MALINKE 0.03 GMB 6

ATTA 0.51 MAR 2 MALINKE 0.13 CIV 6

ATTA 0.49 DZA 2 MALINKE 0.27 MLI 6

ATYUTI 0.13 GHA 2 MALINKE 0.04 GNB 6

ATYUTI 0.87 TGO 2 MALINKE 0.25 GIN 6

AULLIMINDEN 0.55 MLI 3 MALINKE 0.29 SEN 6

AULLIMINDEN 0.40 NER 3 MAMBILA 0.57 CMR 2

AULLIMINDEN 0.05 DZA 3 MAMBILA 0.43 NGA 2

AUSHI 0.27 ZAR 2 MANDARA 0.35 CMR 2

AUSHI 0.73 ZMB 2 MANDARA 0.65 NGA 2

AVATIME 0.51 GHA 2 MANGA 0.60 NER 2

AVATIME 0.49 TGO 2 MANGA 0.40 NGA 2

AZANDE 0.62 ZAR 3 MANYIKA 0.39 MOZ 2

AZANDE 0.15 CAF 3 MANYIKA 0.61 ZWE 2

AZANDE 0.23 SDN 3 MASAI 0.38 KEN 2

AZJER 0.24 LBY 3 MASAI 0.62 TZA 2

AZJER 0.00 NER 3 MASALIT 0.13 TCD 2

AZJER 0.75 DZA 3 MASALIT 0.87 SDN 2

Appendix Table 1 - Partitioned Ethnicities and Countries they Belong to

BABUKUR 0.82 ZAR 2 MASHI 0.12 AGO 2

BABUKUR 0.18 SDN 2 MASHI 0.88 ZMB 2

BAJUN 0.37 KEN 2 MASINA 0.82 MLI 3

BAJUN 0.63 SOM 2 MASINA 0.09 BFA 3

BALANTE 0.73 GNB 2 MASINA 0.09 MRT 3

BALANTE 0.27 SEN 2 MATAKAM 0.70 CMR 2

BANYUN 0.48 GNB 2 MATAKAM 0.30 NGA 2

BANYUN 0.52 SEN 2 MBERE 0.02 TCD 3

BANZIRI 0.14 ZAR 2 MBERE 0.24 CMR 3

BANZIRI 0.86 CAF 2 MBERE 0.74 CAF 3

BARABRA 0.31 EGY 2 MBUKUSHU 0.74 AGO 3

BARABRA 0.69 SDN 2 MBUKUSHU 0.15 BWA 3

BARARETTA 0.18 ETH 3 MBUKUSHU 0.12 NAM 3

BARARETTA 0.44 KEN 3 MBUNDA 0.89 AGO 2

BARARETTA 0.38 SOM 3 MBUNDA 0.11 ZMB 2

BARGU 0.77 BEN 4 MENDE 0.18 LBR 3

BARGU 0.03 NER 4 MENDE 0.82 SLE 3

BARGU 0.19 NGA 4 MINIANKA 0.01 CIV 3

BARGU 0.02 BFA 4 MINIANKA 0.72 MLI 3

BASHI 0.09 BDI 3 MINIANKA 0.27 BFA 3

BASHI 0.83 ZAR 3 MOMBERA 0.72 MWI 2

BASHI 0.08 RWA 3 MOMBERA 0.28 ZMB 2

BATA 0.29 CMR 2 MPEZENI 0.11 MWI 2

BATA 0.71 NGA 2 MPEZENI 0.89 ZMB 2

BAYA 0.20 CMR 2 MUNDANG 0.80 TCD 2

BAYA 0.80 CAF 2 MUNDANG 0.20 CMR 2

BERABISH 0.80 MLI 2 MUNDU 0.30 ZAR 2

BERABISH 0.20 MRT 2 MUNDU 0.70 SDN 2

BERTA 0.75 ETH 2 MUSGU 0.76 TCD 2

BERTA 0.25 SDN 2 MUSGU 0.24 CMR 2

BIDEYAT 0.21 LBY 4 NAFANA 0.74 GHA 2

BIDEYAT 0.40 TCD 4 NAFANA 0.26 CIV 2

BIDEYAT 0.03 EGY 4 NALU 0.41 GNB 2

BIDEYAT 0.36 SDN 4 NALU 0.59 GIN 2

BIRIFON 0.52 GHA 3 NAMA 0.18 ZAF 2

BIRIFON 0.47 BFA 3 NAMA 0.82 NAM 2

BOBO 0.20 MLI 2 NAUDEBA 0.87 BEN 2

BOBO 0.80 BFA 2 NAUDEBA 0.13 TGO 2

BOKI 0.22 CMR 2 NDAU 0.86 MOZ 2

BOKI 0.78 NGA 2 NDAU 0.14 ZWE 2

BONDJO 0.14 ZAR 2 NDEMBU 0.26 AGO 3

BONDJO 0.86 COG 2 NDEMBU 0.39 ZAR 3

BONI 0.67 KEN 2 NDEMBU 0.35 ZMB 3

BONI 0.33 SOM 2 NDOGO 0.01 ZAR 3

BORAN 0.46 ETH 2 NDOGO 0.18 CAF 3

BORAN 0.54 KEN 2 NDOGO 0.81 SDN 3

BRONG 0.84 GHA 2 NDUKA 0.23 TCD 2

BRONG 0.16 CIV 2 NDUKA 0.77 CAF 2

BUEM 0.40 GHA 2 NGAMA 0.30 TCD 2

BUEM 0.60 TGO 2 NGAMA 0.70 CAF 2

BULOM 0.85 SLE 2 NGERE 0.65 CIV 3

BULOM 0.15 GIN 2 NGERE 0.29 LBR 3

BUSA 0.14 BEN 2 NGERE 0.06 GIN 3

BUSA 0.86 NGA 2 NGUMBA 0.65 CMR 2

BWAKA 0.81 ZAR 3 NGUMBA 0.35 GNQ 2

BWAKA 0.15 CAF 3 NGWAKETSE 0.86 BWA 2

BWAKA 0.04 COG 3 NGWAKETSE 0.14 ZAF 2

CHAGA 0.24 KEN 2 NSENGA 0.15 MOZ 3

CHAGA 0.76 TZA 2 NSENGA 0.78 ZMB 3

CHAKOSSI 0.27 GHA 2 NSENGA 0.06 ZWE 3

CHAKOSSI 0.73 TGO 2 NSUNGLI 0.78 CMR 2

CHEWA 0.34 MWI 3 NSUNGLI 0.22 NGA 2

CHEWA 0.50 MOZ 3 NUKWE 0.44 AGO 4

CHEWA 0.16 ZMB 3 NUKWE 0.24 BWA 4

CHIGA 0.12 RWA 3 NUKWE 0.05 ZMB 4

CHIGA 0.87 UGA 3 NUKWE 0.26 NAM 4

CHOKWE 0.81 AGO 2 NUSAN 0.30 BWA 3

CHOKWE 0.19 ZAR 2 NUSAN 0.37 ZAF 3

COMORIANS 0.82 COM 2 NUSAN 0.33 NAM 3

COMORIANS 0.18 MYT 2 NYAKYUSA 0.12 MWI 2

DAGARI 0.67 GHA 2 NYAKYUSA 0.88 TZA 2

DAGARI 0.33 BFA 2 NYANGIYA 0.17 SDN 2

DARI 0.78 TCD 2 NYANGIYA 0.83 UGA 2

DARI 0.22 CMR 2 NYANJA 0.64 MWI 2

DAZA 0.27 TCD 2 NYANJA 0.36 MOZ 2

DAZA 0.73 NER 2 NYASA 0.05 MWI 3

DELIM 0.55 ESH 2 NYASA 0.68 MOZ 3

DELIM 0.45 MRT 2 NYASA 0.27 TZA 3

DENDI 0.60 BEN 3 NZANKARA 0.14 ZAR 2

DENDI 0.39 NER 3 NZANKARA 0.86 CAF 2

DIALONKE 0.36 MLI 3 PANDE 0.38 CAF 2

DIALONKE 0.58 GIN 3 PANDE 0.62 COG 2

DIALONKE 0.06 SEN 3 POPO 0.72 BEN 2

DIDINGA 0.04 KEN 3 POPO 0.28 TGO 2

DIDINGA 0.89 SDN 3 PUKU 0.31 CMR 3

DIDINGA 0.07 UGA 3 PUKU 0.49 GNQ 3

DIGO 0.62 KEN 2 PUKU 0.19 GAB 3

DIGO 0.38 TZA 2 REGEIBAT 0.34 ESH 2

DIOLA 0.14 GMB 3 REGEIBAT 0.66 MRT 2

DIOLA 0.07 GNB 3 RESHIAT 0.83 ETH 3

DIOLA 0.78 SEN 3 RESHIAT 0.06 KEN 3

DUMA 0.63 GAB 2 RESHIAT 0.11 SDN 3

DUMA 0.37 COG 2 RONGA 0.60 MOZ 3

DZEM 0.74 CMR 3 RONGA 0.35 ZAF 3

DZEM 0.03 GAB 3 RONGA 0.05 SWZ 3

DZEM 0.24 COG 3 RUANDA 0.02 BDI 5

EGBA 0.41 BEN 3 RUANDA 0.06 ZAR 5

EGBA 0.52 NGA 3 RUANDA 0.89 RWA 5

EGBA 0.07 TGO 3 RUANDA 0.02 TZA 5

EKOI 0.38 CMR 2 RUANDA 0.02 UGA 5

EKOI 0.62 NGA 2 RUNDI 0.76 BDI 4

ESA 0.03 DJI 3 RUNDI 0.04 RWA 4

ESA 0.52 ETH 3 RUNDI 0.20 TZA 4

ESA 0.44 SOM 3 RUNGA 0.74 TCD 3

EWE 0.44 GHA 2 RUNGA 0.26 CAF 3

EWE 0.56 TGO 2 SABEI 0.56 KEN 2

FANG 0.37 CMR 4 SABEI 0.44 UGA 2

FANG 0.07 GNQ 4 SAHO 0.43 ERI 2

FANG 0.54 GAB 4 SAHO 0.57 ETH 2

FANG 0.02 COG 4 SAMO 0.12 MLI 2

FON 0.86 BEN 3 SAMO 0.88 BFA 2

FON 0.14 TGO 3 SANGA 0.26 CMR 3

FOUTADJALON 0.01 MLI 4 SANGA 0.19 CAF 3

FOUTADJALON 0.11 GNB 4 SANGA 0.55 COG 3

FOUTADJALON 0.88 GIN 4 SEKE 0.34 GNQ 2

FOUTADJALON 0.01 SEN 4 SEKE 0.66 GAB 2

FUNGON 0.81 CMR 2 SHAMBALA 0.10 KEN 2

FUNGON 0.19 NGA 2 SHAMBALA 0.90 TZA 2

GADAMES 0.25 LBY 3 SHEBELLE 0.58 ETH 2

GADAMES 0.27 TUN 3 SHEBELLE 0.42 SOM 2

GADAMES 0.48 DZA 3 SHUWA 0.62 TCD 3

GIL 0.80 MAR 2 SHUWA 0.17 CMR 3

GIL 0.20 DZA 2 SHUWA 0.21 NGA 3

GOMANI 0.86 MWI 2 SONGHAI 0.57 MLI 3

GOMANI 0.14 MOZ 2 SONGHAI 0.36 NER 3

GREBO 0.33 CIV 2 SONGHAI 0.07 BFA 3

GREBO 0.67 LBR 2 SONINKE 0.68 MLI 3

GRUNSHI 0.68 GHA 2 SONINKE 0.03 SEN 3

GRUNSHI 0.32 BFA 2 SONINKE 0.29 MRT 3

GUDE 0.83 CMR 2 SOTHO 0.24 LSO 2

GUDE 0.17 NGA 2 SOTHO 0.76 ZAF 2

GULA 0.61 TCD 2 SUBIA 0.11 BWA 4

GULA 0.39 CAF 2 SUBIA 0.53 ZMB 4

GUN 0.48 BEN 2 SUBIA 0.06 ZWE 4

GUN 0.52 NGA 2 SUBIA 0.30 NAM 4

GURENSI 0.74 GHA 3 SUNDI 0.37 ZAR 2

GURENSI 0.13 TGO 3 SUNDI 0.63 COG 2

GURENSI 0.13 BFA 3 SURI 0.71 ETH 2

GURMA 0.15 BEN 4 SURI 0.29 SDN 2

GURMA 0.12 NER 4 SWAZI 0.45 ZAF 2

GURMA 0.01 TGO 4 SWAZI 0.55 SWZ 2

GURMA 0.72 BFA 4 TABWA 0.57 ZAR 2

GUSII 0.53 KEN 2 TABWA 0.43 ZMB 2

GUSII 0.47 TZA 2 TAJAKANT 0.15 MAR 4

HAMAMA 0.80 TUN 2 TAJAKANT 0.14 ESH 4

HAMAMA 0.20 DZA 2 TAJAKANT 0.66 DZA 4

HAUSA 0.14 NER 2 TAJAKANT 0.05 MRT 4

HAUSA 0.86 NGA 2 TAMA 0.30 TCD 2

HIECHWARE 0.81 BWA 2 TAMA 0.70 SDN 2

HIECHWARE 0.19 ZWE 2 TAWARA 0.57 MOZ 2

HLENGWE 0.82 MOZ 3 TAWARA 0.43 ZWE 2

HLENGWE 0.00 ZAF 3 TEDA 0.34 LBY 3

HLENGWE 0.18 ZWE 3 TEDA 0.35 TCD 3

HOLO 0.84 AGO 2 TEDA 0.31 NER 3

HOLO 0.16 ZAR 2 TEKE 0.31 ZAR 3

IBIBIO 0.11 CMR 2 TEKE 0.03 GAB 3

IBIBIO 0.89 NGA 2 TEKE 0.66 COG 3

IFORA 0.30 MLI 2 TEKNA 0.53 MAR 2

IFORA 0.70 DZA 2 TEKNA 0.47 ESH 2

IMRAGEN 0.10 MAR 3 TEM 0.17 BEN 2

IMRAGEN 0.74 ESH 3 TEM 0.83 TGO 2

IMRAGEN 0.16 MRT 3 TENDA 0.57 GIN 2

ISHAAK 0.20 ETH 2 TENDA 0.43 SEN 2

ISHAAK 0.80 SOM 2 THONGA 0.58 MOZ 3

IWA 0.33 TZA 2 THONGA 0.42 ZAF 3

IWA 0.67 ZMB 2 TIENGA 0.22 NER 3

JERID 0.90 TUN 2 TIENGA 0.78 NGA 3

JERID 0.10 DZA 2 TIGON 0.32 CMR 2

JIE 0.24 KEN 2 TIGON 0.68 NGA 2

JIE 0.76 UGA 2 TIGRINYA 0.51 ERI 3

KABRE 0.39 BEN 2 TIGRINYA 0.44 ETH 3

KABRE 0.61 TGO 2 TIGRINYA 0.05 SDN 3

KANEMBU 0.73 TCD 3 TLOKWA 0.14 BWA 3

KANEMBU 0.25 NER 3 TLOKWA 0.77 ZAF 3

KANEMBU 0.02 NGA 3 TLOKWA 0.09 ZWE 3

KAONDE 0.21 ZAR 2 TOMA 0.29 LBR 2

KAONDE 0.79 ZMB 2 TOMA 0.71 GIN 2

KAPSIKI 0.65 CMR 2 TONGA 0.84 ZMB 2

KAPSIKI 0.35 NGA 2 TONGA 0.16 ZWE 2

KARA 0.85 CAF 2 TRIBU 0.25 GHA 2

KARA 0.15 SDN 2 TRIBU 0.75 TGO 2

KARAMOJONG 0.27 KEN 2 TRIPOLITANIANS 0.74 LBY 2

KARAMOJONG 0.73 UGA 2 TRIPOLITANIANS 0.26 TUN 2

KARE 0.75 ZAR 2 TUBURI 0.25 TCD 2

KARE 0.25 CAF 2 TUBURI 0.75 CMR 2

KGATLA 0.13 BWA 2 TUKULOR 0.39 SEN 2

KGATLA 0.87 ZAF 2 TUKULOR 0.61 MRT 2

KISSI 0.12 LBR 3 TUMBUKA 0.74 MWI 2

KISSI 0.02 SLE 3 TUMBUKA 0.26 ZMB 2

KISSI 0.86 GIN 3 TUNISIANS 0.87 TUN 2

KOBA 0.89 BWA 2 TUNISIANS 0.13 DZA 2

KOBA 0.11 NAM 2 UDALAN 0.82 MLI 3

KOMA 0.57 ETH 2 UDALAN 0.05 NER 3

KOMA 0.43 SDN 2 UDALAN 0.13 BFA 3

KOMONO 0.49 CIV 2 VAI 0.76 LBR 2

KOMONO 0.51 BFA 2 VAI 0.24 SLE 2

KONGO 0.77 AGO 3 VENDA 0.70 ZAF 2

KONGO 0.23 ZAR 3 VENDA 0.30 ZWE 2

KONJO 0.81 ZAR 2 VILI 0.20 AGO 4

KONJO 0.19 UGA 2 VILI 0.22 ZAR 4

KONKOMBA 0.24 GHA 2 VILI 0.11 GAB 4

KONKOMBA 0.76 TGO 2 VILI 0.47 COG 4

KONO 0.74 SLE 2 WAKURA 0.28 CMR 2

KONO 0.26 GIN 2 WAKURA 0.72 NGA 2

KONYANKE 0.30 CIV 2 WANGA 0.79 KEN 2

KONYANKE 0.70 GIN 2 WANGA 0.21 UGA 2

KORANKO 0.39 SLE 2 WUM 0.88 CMR 2

KORANKO 0.61 GIN 2 WUM 0.12 NGA 2

KOTA 0.41 GAB 2 YAKA 0.16 AGO 2

KOTA 0.59 COG 2 YAKA 0.84 ZAR 2

KOTOKO 0.67 TCD 2 YAKOMA 0.40 ZAR 2

KOTOKO 0.33 CMR 2 YAKOMA 0.60 CAF 2

KPELLE 0.48 LBR 3 YALUNKA 0.25 SLE 2

KPELLE 0.52 GIN 3 YALUNKA 0.75 GIN 2

KRAN 0.16 CIV 2 YAO 0.13 MWI 3

KRAN 0.84 LBR 2 YAO 0.65 MOZ 3

KREISH 0.10 CAF 2 YAO 0.22 TZA 3

KREISH 0.90 SDN 2 YOMBE 0.13 AGO 3

KUNDA 0.84 MOZ 3 YOMBE 0.48 ZAR 3

KUNDA 0.15 ZMB 3 YOMBE 0.39 COG 3

KUNG 0.10 BWA 2 ZAGHAWA 0.14 TCD 2

KUNG 0.90 NAM 2 ZAGHAWA 0.86 SDN 2

KUNTA 0.85 MLI 2 ZEKARA 0.83 MAR 2

KUNTA 0.15 DZA 2 ZEKARA 0.17 DZA 2

KWANGARE 0.84 AGO 2 ZIMBA 0.16 MWI 2

KWANGARE 0.16 NAM 2 ZIMBA 0.84 MOZ 2

Appendix Table 1 reports the name of partitioned ethnic groups (as coded by Murdock (1959)) and the percentage of the historical

homeland of the split ethnic groups that fall into more than one country. Section 2 in the paper gives details on our approach in

identifying partitioned ethnicities.