Ethnonationalist Triads
Assessing the Influence of Kin Groups on
Civil Wars
By Lars-Erik Cederman, Luc Girardin, and
Kristian Skrede Gleditsch*
NATIONALISM has the potential to transgress and transform
state borders. Irredentism represents the most radical form of
border transformation, but border-transgressing effects, such as external
support by kin groups short of major military interventions, may
also make ethnonationalist civil wars more likely. Given the highly
asymmetric nature of such conflicts, which by definition feature nonstate
groups challenging well-armed governments, it can be expected
that the former will seek support from related groups in neighboring
countries.
Even a quick look at the ethnopolitical map reveals that there are
plenty of structural opportunities for such transborder influences.
Whereas relatively few cases of outright irredentism have occurred,
ethnonationalist civil wars have often featured external support from
kin groups across state borders.1
Examples include Kurdish transborder
cooperation against hostile state governments such as Turkey and Iraq.2
Transborder nationalism can also be blamed for having contributed to
ethnic conflict in Croatia, Bosnia, Kosovo, and other parts of the former
Yugoslavia.3
Although the Russian “near abroad” that emerged in
the post–cold war period has generally been more peaceful than ex*
We thank Paul Diehl, Harvey Starr, Jeremy Weinstein, Andreas Wimmer, participants at seminars
at the University of Nottingham and Trinity College Dublin, as well as three anonymous reviewers
and the editors for helpful comments and suggestions. This research was supported by grants from
the European Science Foundation (06ECRPFP004), the Research Council of Norway (180441/V10),
the Swiss National Science Foundation (105511-116795/1), and the U.K. Economic and Social Research
Council (RES-062-23-0259).
1
Horowitz 1985; Chazan 1991.
2
Kiris¸ci and Winrow 1997.
3
Brubaker 1996; Gagnon 2004.
World Politics 61, no. 3 (July 2009), 403–37
Copyright © 2009 Trustees of Princeton University
10.1017/S0043887109000148
404	 world politics
4
King and Melvin 1999–2000.
pected, tensions persist that could be exploited to poison ethnic politics
in the post-Soviet republics.4
Despite the existence of a rich case-based literature documenting
the impact of kin groups on ethnonationalist civil wars, we still know
little about their general effect on the likelihood of internal conflict.
Quantitative studies of civil war onset have thus far tended to ignore
such connections. This is the result of two blind spots, one relating
to overaggregation and the other to underaggregation. Regarding the
first problem, influential studies in the political economy tradition have
relied heavily on materialist interpretations of civil wars at the country
level while downplaying specific interactions involving ethnic groups.
The second blind spot pertains to the inclination of existing research
on civil war to treat civil wars as purely domestic phenomena. Since researchers
have assumed that civil wars take place within societies, they
assume as well that the key causes of conflict must also be found within
the boundaries of formally independent nation-states. They thus disregard
the role of transborder linkages and processes.
In order to capture the ethnic-kin effect at both the domestic and
the transnational levels of analysis, we propose a triadic model that
analyzes ethnonationalist civil war as a disaggregated, relational phenomenon
that includes links across state borders. Based on new data
from geographic information systems, we show that ethnic kin have a
discernible impact on internal conflict within a sample limited to Eurasia
and Northern Africa. Rather than being an unconditional effect,
this influence depends directly on the dyadic power balance between
the main conflict parties. In brief, the presence of external kin who
could potentially offer support is more likely to increase the risk of
conflict when the excluded minority is relatively large. This suggests
that relatively smaller groups are less likely to rebel and that the potential
assisting transborder kin groups tend to exercise caution: they give
serious consideration to power relations before deciding whether to get
involved in a conflict across the border. In contrast, there are no clear
results indicating that the governmental or nongovernmental status of
external kin groups matters for conflict propensity.
The next section charts the state of the literature before deriving
our main hypotheses.The section following that, on operational definitions,
then sets the stage for the analysis and our statistical results.
	 ethnonationalist triads	 405
5
For example, Posen 1993; and Hardin 1995.
6
Weiner 1971.
7
Weiner 1971, 683.
8
Brubaker 1996.
9
Horowitz 1985.
Previous Research on the Ethnic-Kin Effect and Its
Impact on Conflict
The qualitative literature on ethnicity and nationalism provides rich accounts
of border-transgressing processes involving support across state
borders. Whereas some studies of ethnic conflict following the cold
war relied on dyadic actor constellations under anarchy,5
the bordertransgressing
nature of ethnonationalism has been fully appreciated by
scholars who rely on explicit theories of nationalism.
In a classic article, Myron Weiner identifies a “Macedonian syndrome”
that involves an actor constellation featuring an irredentist state,
an anti-irredentist neighboring state, and a shared ethnic group that is
a minority in the latter state.6
Based on this triadic setup, Weiner outlines
an interactive process triggered by irredentist claims on the part of
the minority that leads to growing ethnonationalist polarization. Once
all actors have aligned themselves along the relevant ethnonationalist
cleavage, extremist entrepreneurs, creating and responding to real or
imagined threats, come to dominate politics within each of the three
collective units, with political violence the likely consequence. Viewing
his model as a corrective to country-level analysis, Weiner concludes
that “too many theories of development assume constancy or irrelevancy
as far as the international environment is concerned, and assume
also that internal political development or decay occur without regard
to external factors.”7
Extending this perspective to less drastic situations than outright
irredentism, Rogers Brubaker analyzes how state-led nationalization
projects trigger support of national minorities from external national
homelands, a process that may or may not involve violence.8
Applied
to the former Yugoslavia, this triadic analysis shows why Croatia’s bid
for independence (the nationalizing state) triggered calls for “Greater
Serbia” among the Serbs in Croatia (the national minority) with the
goal of joining Serbia (the national homeland).
Also mindful of the border-transgressing logic of ethnonationalist
conflict, Donald Horowitz shows that externally supported irredentist
campaigns are relatively rare, despite the frequent presence of structural
conditions favoring such phenomena.9
Border revisionism carries with
it considerable risks that invite caution on the part of possible sponsor-
406	 world politics
ing groups, especially if they are in charge of their own states. While
geopolitical considerations usually trump ethnic loyalties, less drastic
types of aid may be offered in cases that involve separatist groups. Even
at this moderate level, which falls well short of irredentism, the fear of
conflict contagion may dissuade many kin groups from lending a helping
hand to secessionist minorities.
Despite the suggestive logic of these border-transgressing mechanisms,
the qualitative literature on ethnonationalist conflict offers little
by way of systematic empirical assessments of a general effect on conflict
likelihood. Ironically enough, the quantitative literature turns out
to be equally uninformative. As we have argued above, this lacuna is
caused by the tendency of political economists both to downplay ethnicity
and to use comparative designs that exclude neighborhood effects
and dependence among observations.
The first of these two tendencies can be understood in the context
of the predominantly materialist logic propounded by political economists.
Eager to show that materialist, economic explanations trump
cultural, identity-related arguments, some of these scholars argue that
civil wars are driven predominantly by individual rebels’ opportunity
costs in terms of employment and by warlords’ personal enrichment
through looting of raw materials such as diamonds and oil.10
Another
influential argument shifts the materialist logic from the individual level
to the level of political institutions, postulating that internal conflict is
likely to erupt in weakly governed states that are unable to project their
territorial control, especially in rough terrain.11
According to these interpretations,
it is greed or the logic of insurrection that causes civil
wars rather than ethnic or other grievances. In brief, conflict erupts for
opportunistic, self-interested reasons rather than for political motives
or idealistic, high-minded projects such as nationalism.
Most scholars within this research tradition operationalize ethnicity
by relying on the ethnolinguistic fractionalization index (elf). Based
on the Herfindahl index, this measure indicates the likelihood of two
individuals in a country not belonging to the same ethnic group.12
Yet
ethnonationalist civil wars clearly are not the aggregated effects of individual-level
processes. It remains unclear, then, how and whether at
all this measure is related to conflict.13
Furthermore, the alleged irrel-
10
Collier and Hoeffler 2004; for an overview, see Ross 2006.
11
Fearon and Laitin 2003.
12
Taylor and Hudson 1972.
13
Cederman and Girardin 2007.
	 ethnonationalist triads	 407
evance of ethnicity may not be robust even measured in terms of the
elf index.This measure of diversity can be shown to be significantly related
to conflict in studies using alternative nonlinear specifications or
conflict data sets that include violent conflict at a lower level than that
of the customary threshold of one thousand battle deaths suggested by
the Correlates of War project.14
Although it is possible to construct theoretically more satisfactory
country-level measures of ethnic exclusion that appear to have an influence
on conflict,15
the best way to establish the relevance of ethnonationalist
conflict mechanisms is to disaggregate the analysis down to
the group level. This also avoids the aggregation problem in countrylevel
studies, which simply ignore all variation between groups within
countries and disregard all instances of groups without conflict with
governments in countries where we see conflict for one group. Introducing
a disaggregated model that features geocoded center-periphery
dyads, Buhaug, Cederman, and Rød are able to show that the probability
of ethnonationalist civil wars increases with the excluded group’s
relative demographic size, its distance from the capital, and the roughness
of the terrain within its settlement area.16
Based on these results, we have good reason to believe that the invisibility
of ethnicity in conventional studies depends on improperly
specified, overaggregated research designs. In contrast, the second gap
afflicting the literature pertains to underaggregated, statecentric analysis
that does not even attempt to measure transborder effects or, if this
is done, that uses very imprecise regional indicators.
Most comparative analyses have treated individual countries as independent
observations where conflict may or may not occur. However,
we have strong theoretical reasons to believe that the risk of civil war
can be influenced by international and transnational features ignored
by these studies. Many have pointed out that civil wars appear to cluster
regionally and that certain regions appear particularly conflict prone in
specific time periods. Although such clustering could stem from similarities
in the attributes among the countries that make them all the
more likely to experience civil war, there may also be a more funda-
14
See Sambanis 2004; Hegre and Sambanis 2006. Researchers have also used alternative conceptions
of ethnic relations, including polarization (see Esteban and Ray 1994; Montalvo and ReynalQuerol
2005) or dominance measures based on the share of the second largest ethnic group or all
groups other than the dominant (e.g., Ellingsen 2000; and Vanhanen 1999). We focus our discussion
on the elf index here since this is the most prominent measure and these alternative measures also
suffer from similar problems for country-level data.
15
See Cederman and Girardin 2007.
16
Buhaug, Cederman, and Rød 2008.
408	 world politics
mental spatial dependence at play among the observations, where the
risk of civil war increases for states that are either exposed to, or have
particular ties to, other countries involved in civil conflicts. Ties to another
state with an active or latent conflict can facilitate insurgent mobilization
through the ability to raise resources, personnel, or military
equipment.17
Likewise, groups may become more likely to engage in
violence if they observe successful insurgencies in other states that can
serve as models for emulation.18
Furthermore, the spillover effects or
externalities of conflict such as economic recessions and refugees may
serve to increase the risk of conflict in other affected areas.19
Although much of the existing work on the role of international factors
in civil war has focused on the diffusion implications of ongoing
conflict in another state, there are many consequences for conflict that
may apply in the presence of transborder ties, irrespective of whether
we actually see conflict in the other state. For example, ethnic kin may
support or finance insurgencies in other states in instances where they
are not aggrieved in the other state or have no interest in rebelling
against their own government.20
Poor relations between states may provide
governments with incentives to foster insurgencies in other states
to undermine rivals; or poor relations may encourage rebellion if insurgents
expect to receive support from outside parties.21
There are a number of empirical studies that provide strong support
for the importance of border-transgressing factors. When the World
Bank commissioned a set of case studies to evaluate the Collier and
Hoeffler model,22
many of the contributors found that international
factors appeared very important in accounting for conflict onset, despite
being largely absent in Collier and Hoeffler’s original work.23
A number
of statistical studies have shown that the positive conflict-clustering
effect does not disappear when controlling for well-known predictors
of civil war and survives various robustness tests, including unmeasured
heterogeneity and systematic sample differences on observed characteristics.
Hence, this cannot be dismissed as an artifact of the clustering
of similar country characteristics.24
A number of studies confirm that
interstate relations influence the risk of civil war onset as well as the
17
See, for example, Gleditsch 2007; and Salehyan and Gleditsch 2006.
18
Ayres and Saideman 2000.
19
Sandler and Murdoch 2004.
20
Davis and Moore 1997.
21
Gleditsch 2007; Kuperman and Crawford 2006; Thyne 2007.
22
Collier and Hoeffler 2004.
23
Collier and Sambanis 2005a, 2005b; Sambanis 2004.
24
Buhaug and Gleditsch 2008.
	 ethnonationalist triads	 409
likelihood of interventions in ongoing civil war.25
Although there are
in principle many transborder ties that could be relevant in civil war,
existing empirical studies suggest that ethnic ties are among the most
prominent linkages that can be mobilized in civil wars.26
While this incipient research on cross-border linkages in civil war is
very encouraging, it still suffers from a number of shortcomings resulting
from its reliance on highly aggregate country-level analysis. Some
studies have used very crude proxies of external ethnic ties that lack
face validity. Collier and Hoeffler,27
for example, look at the size of
immigrant communities from a given country in the United States as
a measure of opportunities for support from diasporas. This is a highly
problematic measure, as it ignores all communities not in the U.S. and
may pick up on refugee populations that are a result of previous or ongoing
conflict rather than a prior cause of conflict onset. Furthermore,
most of the existing studies on the role of transborder ethnic communities
have simply considered whether countries with such groups
are more likely to see conflict,28
but they have not actually examined
whether the transborder group itself was involved in the rebellion.
Most studies have been conducted at the country level; they therefore
encounter the common problems of overaggregation and treating conflict
and nonconflict cases nonsymmetrically. More specifically, a single
instance of a transborder group and a conflict in a state (say India) will
be taken as support for the theory, even if there may be a very large
number of other border-transgressing groups in the same state that
do not rebel—and therefore should be counted as evidence against the
theory. Overcoming this problem requires us to move down to the level
of dyads of specific individual groups and their interactions with the
government. Cetinyan provides one of the few fully dyadic studies of
transborder groups using the Minorities at Risk (mar) data.29
Although
he does not find any effect of groups having ethnic kin on the risk of
conflict, this finding may follow partly by construction, for reasons that
we will return to later.
Existing studies of transborder ethnic ties in civil war have overlooked
the strong mediating role of geography and distance. Although
such ties in principle could extend over large distances,30
it seems rea-
25
Austvoll 2005; Gleditsch 2007; Salehyan 2009; Thyne 2007.
26
Austvoll 2005; Saideman 2001.
27
Collier and Hoeffler 2004.
28
For example, Gleditsch 2007.
29
Cetinyan 2003.
30
For example, Shain and Barth 2003.
410	 world politics
sonable to assume that the cross-border ties most relevant for influencing
the risk of civil wars are those pertaining to groups that are
geographically close. Insurgents face many tactical advantages when
operating from bases outside a country’s territorial boundaries, since it
is costly for governments to pursue rebels across borders and violate the
sovereignty of other states, but these military advantages to insurgents
dissipate over large geographical distances. Geographically close communities
are likely to have a denser web of contact and communication
than more distant ones. Existing studies of international factors in civil
war suggest that geographical linkages are very important. The risk of
civil war increases when a neighboring state is fighting a civil war but is
not affected by conflict in distant countries.31
Moreover, Salehyan and
Gleditsch show that, although refugees from neighboring countries appear
to augur a higher risk of conflict, there is no evidence that refugees
from faraway areas are associated with civil war.32
Many of the data
sources used in previous research do not allow researchers to identify
where other segments are located and to measure the distance between
kin groups.
In sum, there are very strong reasons to believe that cross-border
groups have an effect on the risk of civil conflict in center-periphery
dyads, but at the same time this effect has been obscured by problems
related to data deficiencies and improper modeling or research design.
Theorizing Ethnonationalist Triads
Our literature review has shown that it is difficult or even impossible
to capture the effect of ethnic-kin groups based on conventional, country-level
analysis. Therefore, the first priority should be to construct
a plausible baseline model of dyadic ethnonationalist conflict within
a specific country; this will allow us to examine the triadic effects of
linkages to kin groups in other states. We build directly on Buhaug,
Cederman, and Rød,33
who offer such a disaggregated dyadic model
that focuses on ethnic groups’ access to power. 34
Measuring the effect
of ethnonationalist exclusion, they disaggregate both ethnicity and
conflict to the level of explicitly geocoded center-periphery dyads. Each
dyad is composed of an egip, that is, an ethnic group (or groups) in
power, and a meg, that is, a marginalized ethnic group without access
31
Buhaug and Gleditsch 2008.
32
Salehyan and Gleditsch 2006.
33
Buhaug, Cederman, and Rød 2008.
34
See Wimmer 2002.
	 ethnonationalist triads	 411
to executive power.35
Illustrating these definitions, Figure 1 displays the
basic center-periphery (bcp) configuration. The next section provides
details on the operationalization of both notions.
	 Within each dyadic relationship, Buhaug, Cederman, and Rød postulate
(1) that powerful ethnic groups excluded from power are most
likely to mobilize around an ethnonationalist program and to initiate
conflict against the government and conversely (2) that the government
will engage in repression to curb the power of such threatening
contenders. Thus, according to this formulation, civil wars occur when
peripheral contenders to the government are powerful enough to challenge
the center and sufficiently motivated to do so.36
The rebels’ opportunities and willingness to challenge the center are
related to many factors. As a first cut, we follow Buhaug, Cederman,
and Rød in highlighting the demographic and geographic dimensions.
Other things being equal, it can be expected that larger groups will be
able to stage successful collective action thanks to their superior num-
bers:
H1. The probability of conflict increases with the relative demographic
size of the excluded group.
Most obviously, this demographic effect depends on an opportunitydriven
mechanism that can be explained in terms of resource mobili-
zation.37
However, the rebels’ willingness to act should increase with
demographic size as well, because exclusion becomes normatively more
problematic as the share of the excluded population increases, whether
the political system is democratic or not. Permanent exclusion of large
minorities, let alone, as is sometimes the case, of the majority of the
population, represents powerful grievances that can trigger conflict.38
Despite their central importance to any ethnographic configuration,
raw demographic size is likely to be an imperfect predictor of the
power and motivations of marginalized ethnic groups. In particular,
35
Cederman and Girardin 2007. Note that this specification reifies the ethnic groups for empirical
tractability reasons. Thus, here we do not attempt to analyze the groups’ representatives and organizations,
which constitute the actual locus of agency. For a critique of group reification, see Cederman
2002; and Brubaker 2004.
36
Gurr 2000.
37
McCarthy and Zald 1977;Tilly 1978. We acknowledge that the ability of a group to successfully
challenge the state can be influenced by resources other than demographic size and that some excluded
groups may have far more or less power relative to the state than would be expected from their relative
size, due to factors such as economic position or degree of group cohesion; see, for example, Hechter
1987. However, there are no systematic sources of data that would allow us to consider alternative
resources, so we focus here on potential power as given by demographic size.
38
Horowitz 1985; Petersen 2002; Wimmer 2002.
412	 world politics
even relatively small groups, such as the Chechens and the Aceh, can
stage surprisingly effective, long-lasting rebellions. On average, then,
the prospects of peripheral challenges to the central government can be
expected to be the most successful in the cases where the latter’s reach
is least developed. We propose the following two hypotheses:
H2a. The probability of conflict increases with the distance between
the excluded group and the capital.
H2b. The probability of conflict increases with the roughness of the
terrain within the settlement area of the excluded group.
Again, the causal mechanisms vary with both opportunity and willingness.
In the former sense, standard logistical arguments show how state
capacity declines with increasing distance and geographic obstacles.39
Beyond such straightforward materialist relationships, Buhaug, Cederman,
and Rød suggest that the motivation of rebel groups also varies
with distance and terrain.40
Rokkan’s multidimensional notion of
“peripherality” suggests that ethnic groups that are far removed from
the political center and live in inaccessible territories hold, on average,
more hostile attitudes toward central rule than those that have been
more thoroughly socialized to tolerate central control.41
In this regard,
distance and remoteness can be thought of as indicators that cultural
penetration by the central state is lacking.42
39
Boulding 1962.
40
Buhaug, Cederman, and Rød 2008.
41
Rokkan 1999.
42
Cederman 2008.
Figure 1
The Basic Center-Periphery (BCP) Configuration
	 ethnonationalist triads	 413
Having summarized the dyadic baseline model based on the bcp
configuration, we are now ready to extend it by introducing a third unit
category, namely, that of the ethnic kin group, which is the potential
deliverer of external support; see Figure 2. This extended center-periphery
(ecp) configuration conforms directly to the triadic logic proposed
by Weiner and Brubaker.43
Depending on the presence of a kin
relationship, assistance can come either from a peripheral group in the
neighboring country or from this country’s egip, whether it is the only
dominant group or part of the governmental coalition. Accordingly,
we classify the former transborder link as the symmetric type, because
it connects two peripheries, and the latter one as the asymmetric type,
because it links the periphery with the center of another state.
Why should the risk of conflict depend on whether a peripheral
group has a link to cross-border kin? On the one hand, the ethnic-kin
group may provide additional resources to a peripheral group and thus
make the relative power balance more favorable to the peripheral group
than would be apparent from the attributes of the group at the domestic
level. On the other hand, although there could be instances where
transborder kin simply help contribute to resources that are “scarce” at
the domestic level, it will often be the case that such groups can provide
qualitatively different forms of resources and contributions to insurgencies.
Since kin groups are based in other states, it will typically be much
more difficult for the central government to target supporters of insurgency
and deter contributions through severe repression than would be
the case for groups fully contained within the territory controlled by
the state. Furthermore, insurgencies that can seek shelter among kin
on the other side of international boundaries and operate from extraterritorial
bases will be more difficult for governments to conclusively
defeat. In this sense, cross-border links can provide insurgencies with
many advantages beyond the direct aggregation or resource-pooling
implications.
Based on the ecp configuration, we propose a series of hypotheses
about the influence of the transborder relationship on conflict probabilities
within the core dyad.The simplest version introduces this influence
as a constant effect, where border-transgressing groups generally
increase the risk of violence.The justification for such an unconditional
argument can be derived from noninstrumentalist interpretations of
nationalism. For example, Connor claims that rationalistic and materialist
approaches to ethnic nationalism “can be faulted principally for
43
Weiner 1972; Brubaker 1996.
414	 world politics
their failure to reflect the emotional depth of national identity.”44
More
specifically referring to external intervention in civil wars, Holsti suggests
that “reasons of affinity and sentiment rather than . . . power or
more hard-headed cost-benefit analyses.”45
Such observations find at
least partial support in Heraclides’ comparative case studies of seces-
sion.46
Most simply stated, this hypothesis can be stated as follows:
H3. The probability of conflict increases if the excluded group has
ethnic kin in a neighboring state
There is, however, good reason to believe that such a hypothesis is
too bluntly stated. Indeed, most scholars are inclined to make their
analysis of the ethnic-kin effect dependent on power-related considerations.
For example, in their study of “diaspora politics” in the former
Soviet Union, King and Melvin insist that
in the realm of ethnicity and international relations, identity politics is often
more about politics than about identity. Disputes over the allocation of scarce
resources, competing visions of foreign policy directions, domestic political contests,
and other prosaic features of political life frequently trump any putative
duty that political elites might feel toward individuals who share their language
or culture beyond their own frontiers.47
44
Connor 1994, 206.
45
Holsti 1996, 127.
46
Heraclides 1990.
47
King and Melvin 1999–2000, 109.
Figure 2
The Extended Center-Periphery (ECP) Configuration
Kin-group
support
Symmetric Link
Asymmetric Link
	 ethnonationalist triads	 415
Whether of the symmetric or asymmetric type, intervention in ethnonationalist
civil wars is a risky business even if the assistance falls
below outright irredentist warfare. For one, groups that are weak at the
outset will face a major risk of escalation by trying to challenge governments
through violence even if they count on the support of transborder
groups. In addition there may be carry-on effects of unsuccessful
insurgencies in other states. Horowitz’s classic study Ethnic Groups in
Conflict still offers the most subtle analysis of these considerations.48
Focusing on separatist ethnic conflict, this work highlights a number of
reasons why cross-border bonds often remain inactive despite considerable
minority mobilization and possibly even violent ethnonationalist
conflict:
Trans-border ethnic affinities more often promote restraint in supporting separatists
or intervention in behalf of a central government fighting to suppress
separatism. Fear of contagion and domino effects is widespread. Among separatists,
this creates a fear that the failure of a movement in one state will hurt
movements in others—hence the ties among them. Among states, fear of the
success of separatism works in the opposite direction.49
How can this caution be empirically assessed? The most obvious way
to do so is to measure the power of the meg that is to receive support in
order to discriminate between plausible and less plausible challengers at
the outset. Before deciding to offer assistance to ethnic kin, members of
the external kin group have to evaluate the chances of success. Accordingly,
Horowitz points out that “there is a certain circularity involved
in securing foreign support in the first place. No foreign state will risk
committing itself to a movement that appears weak.”50
In addition to the implications of demographic balance for the prospects
for success, it will also generally be the case that larger excluded
groups can claim more legitimate political grievances than smaller ones,
thus adding to the radicalizing effect of external kin groups.51
This may
increase the identification of transborder kin groups with their cause
and make it easier to mobilize political support. Furthermore, it is generally
easier for governments in other states to allow active support for
insurgencies perceived as legitimate.
In operational terms, the most straightforward test of this power-
48
Horowitz 1985.
49
Horowitz 1985, 274–75.
50
Horowitz 1985, 272.
51
See Jenne 2007.
416	 world politics
sensitive mechanism combines hypotheses H1 and H3 interactively in
the following way:
H4. The ethnic-kin effect on conflict increases in proportion to the
excluded group’s relative demographic weight in the primary dyad.
Our expectation is that this hypothesis, rather than H3, will be empirically
confirmed.52
Finally, we need to consider the nature of the external group. At least
in principle, it could matter a great deal whether it is another peripheral
group in a symmetric link or a central actor in an asymmetric relationship.
Other things being equal, there can be no doubt that groups that
control state resources, such as egips, are more powerful than those
that do not. We know from existing research that interstate disputes
between states are particularly likely when an excluded group is politically
privileged in other states.53
Direct interventions against another state are explicit acts of aggression
and pose large risks for governments. This will in turn make less
direct forms of support for insurgents more attractive, but the effects
on civil war are still ambiguous. It is not obvious whether the willingness
of egips to support insurgencies is higher than that of excluded
kin groups: “Even where ethnic affinities relate, not to peripheral minorities
in the external state, but to centrally influential groups, support
is by no means automatic.”54
There are many reasons why state
governments may be more cautious than actors from nonstate groups.
Most obviously, territorial revisionism can come back to haunt its
state-based promoter. Multiethnic states that have problems controlling
their own minorities must typically think twice before supporting
rebellions in neighboring states. More generally, international norms
governing interstate relations are biased in favor of stability and nonintervention.
Their impact has been most clearly observed in the case
of sub-Saharan Africa,55
but it is likely to apply to other parts of the
world as well.56
Caution also derives from the fear that cross-border
assistance may likewise lead to unwanted turmoil spreading across state
borders into the territory of the intervening state. Furthermore, when
52
A pure aggregation effect would suggest that the risk of conflict should hinge on the size of this
transnational kin group. However, the fact that transnational kin can provide qualitatively different
forms of resources suggests that the risk of conflict should increase even when transnational kin groups
are not necessarily particularly large.
53
See Davis and Moore 1997; Gleditsch, Salehyan, and Schultz 2008.
54
Horowitz 1985, 275.
55
Herbst 1989.
56
Zacher 2001.
	 ethnonationalist triads	 417
potential external government support for excluded groups is known
in advance, governments in countries experiencing conflict will face
greater incentives to provide some form of accommodation to excluded
groups, which in turn would make these groups less likely to resort to
violence. While the arguments for a cautioning, countervailing effect
on accommodation are strong, there is empirical evidence pointing in
the opposite direction.57
Thus, we state our hypothesis in favor of a
stronger conflict-inducing impact in the asymmetric cases as compared
with their symmetric counterparts.
H5. The risk of conflict increases if the kin group is governmental
(egip) rather than another peripheral group (meg).
Nevertheless, in the presence of the countervailing, power-enhancing
influence of state power, it is hard to be certain of the net effect. To
determine it more precisely, it would be necessary to consider further
factors influencing supporting state’s calculations that unfortunately
fall outside the scope of this study. A prominent example relates to
the exact geographic location of the groups in question. Horowitz observes
that “external aid seems longest-lived when it comes, not from
strong, established states with clear-cut interests, but from irregular
forces across porous, remote borders.”58
Moreover, risk-taking behavior
can also be expected to vary with the nature of the interstate relations
in question. While geopolitical rivalry could introduce temptations to
weaken the other side through the transborder support of insurgen-
cies,59
stable democratic relations should make such interventions less
likely.60
Operationalizing Ethnonationalist Triads
In this section we describe the operational definitions of our variables.
Because our analysis builds directly on the operationalization of the
dyadic baseline model proposed by Buhaug, Cederman, and Rød,61
we
provide a short summary of the data used in that study before turning
to the data on transborder ethnic affiliations. This dyadic setup traces
conflict between ethnically defined state authorities, that is, the “ethnic
group(s) in power,” and their challengers. In this connection, the spatial
dimension is crucial, because it helps us not only to identify and locate
57
Saideman 2001.
58
Horowitz 1985, 276.
59
Saideman 2002.
60
Belanger, Duchesne, and Paquin 2005; but see also Saideman 2007.
61
Buhaug, Cederman, and Rød 2008.
418	 world politics
the ethnic groups but also to estimate their demographic sizes. All in
all, the extended triadic model relies on information along six dimensions,
the first five of which were covered by Buhaug, Cederman, and
Rød:
1. the identity and location of ethnic groups,
2. demographic group sizes,
3. ethnic group(s) in power,
4. geocoded data on distances and terrain,
5. ethnic dyadic conflicts, and
6. transborder ethnic affiliations.
In the following, we describe briefly the data-collection efforts before
turning to the results of the analysis in the next section.
Geocoding Ethnic Groups
In their choice of a basic data set on ethnic groups, Buhaug, Cederman,
and Rød rely on the well-known Atlas Narodov Mira (anm).62
The anm
stems from a major project to chart ethnic groups undertaken by Soviet
ethnographers in the early 1960s. Their efforts bore fruit in the extensive
but still untranslated atlas, covering the entire world. The anm has
a number of strengths: it is complete and carefully researched, provides
high-quality maps, relies on a consistent classification of ethnicity, and
offers a uniform group list valid across state borders; this last item, in
particular, is important for our research purposes. However, the anm
is not without its problems. It is exclusively defined on the basis of
linguistic criteria and therefore disregards many other often potentially
important markers of ethnicity such as religion. It also contains many
groups that are politically irrelevant and insufficiently mobilized to
have any capacity for collective action. In particular, the anm list is often
seen as problematic for identifying relevant ethnic groups in Africa.63
Finally, the anm data, based as they are on information from the 1960s,
are clearly outdated in some cases.64
Realizing that this choice of data
source is far from uncontroversial, we nevertheless contend that the advantages
outweigh the disadvantages, at least until a more suitable data
set becomes available. In order to make the data of the anm available for
statistical analysis, all of the Atlas’s maps were converted by scanning
62
Bruk and Apenchenko 1964.
63
E.g., Posner 2005.
64
Moreover, the anm data do not reflect population changes due to migration or refugees. As
such, the Kashmiris are considered to be present in India only, despite there reportedly being over one
hundred thousand migrants from the province in Pakistan. However, in our case, this helps ensure that
the data do not reflect population movements that may be in part a result of conflict.
	 ethnonationalist triads	 419
to a digital format. Then, the digitized maps were geo-referenced as
polygons, meaning that they were aligned with underlying gis country
shapes for the period (1964). The resulting data set, labeled greg
(Geo-Referencing of Ethnic Groups), contains spatial information on
the location of more than sixteen hundred ethnic groups identified in
the anm.65
Estimating Group Sizes as Share of State Populations
Having determined the geographic location of all ethnic groups, the
next task is to construct a suitable measure of the power balance in the
center-periphery dyads. Because a number of states did not exist at the
time of the atlas’s publication, such as the post-Soviet republics of Russia,
Buhaug, Cederman, and Rød propose a spatial estimation method
based on an intersection of territorial country masks, the group polygons,
and population-density maps. Using the boundaries in a country
layer (shapefile) representing the post–cold war period as “cookie cutters,”
the method selects the group polygons (or parts thereof) that fell
within the borders of each state. The ethnic group layers were then
intersected with a gridded population-density layer, making it possible
to measure the size of the population that fell within a given ethnic
group’s “state-cropped” polygon(s).66
Identifying the Ethnic Group(s) in Power
The notion of “ethnic group(s) in power” (egip) that we introduced
above provides the crucial piece of information that makes the definition
of ethnic dyads possible. In our empirical analysis, we follow Cederman
and Girardin in considering a group, or a coalition of groups, to
be in power if their leaders serve (at least intermittently) in senior governmental
positions, especially within the cabinet.67
Thus we focus on
ethnic groups’ influence over the executive at the national level rather
than on their legislative or local power. In addition to the ethnic background
of senior cabinet members, specific institutional arrangements,
such as different types of power sharing and consociationalism, may
also be indicators of power inclusion. By power sharing, we mean any
arrangement that divides the access to power among the groups making
up the governing coalition. Accordingly, egips can consist of more
than one group. For example, we code all the four language groups
65
Weidmann, Rød, and Cederman 2010.
66
For details on this procedure, see Buhaug, Cederman, and Rød 2008.
67
Cederman and Giradin 2007.
420	 world politics
of Switzerland as constituting the egip.68
Wherever deemed appropriate,
period-dependent egips were introduced as a way to capture major
shifts in the political constellation of power access (see, for example,
Afghanistan and Yugoslavia). By definition, any group not coded as an
egip is a marginalized ethnic group (meg). We can now form centerperiphery
dyads as pairwise constellations of a country’s egip and each
of its megs. Given the enormous difficulty of identifying egips in subSaharan
Africa, we limit the empirical sample to Eurasia and North
Africa, thus covering roughly half of the world’s states.
Geocoding Distances and Terrain
Hypotheses 2a and 2b call for geographic data on the distances between
each meg and the corresponding capital, as well as the extent of
mountainous terrain in their home regions. The location of the ethnic
groups can be readily determined by means of geocoded polygons.
Groups belonging to the egip are by default coded with the capital
city as their location. For megs represented by two or more polygons,
Buhaug, Cederman, and Rød generated a weighted distance measure
that gives the average centroid-capital distance for all locations of the
group, weighted by the population size in each polygon. The population
weighting is necessary to prevent a distortion of distances due
to small clusters far away from the group’s core settlement area. To
reduce outlier influence and to account for an expected declining effect
of distance with higher values, we take the natural logarithm of
the distance variable.The second proxy for geographic opportunity and
willingness—extent of mountainous terrain—was computed in a manner
similar to that of group populations. Intersecting gridded mountain
data with the boundaries of the ethnic groups yielded the share of the
two-dimensional area of each polygon covered by mountains.The areaweighted
mountain variable takes on values between zero and one.
Determining the Onset of Dyadic Ethnic Conflict
The final step of data generation for the dyadic baseline model concerns
the dependent variable, namely, the outbreak of dyadic ethnic conflict.
The main source was the ucdp/prio Armed Conflicts Dataset, henceforth
acd,69
which is arguably the most established country-level data
set on armed conflict thanks to its comparably low minimum casualty
threshold of just twenty-five annual battle-related deaths and its sepa-
68
For a full list of egips, see Buhaug, Cederman, and Rød 2008.
69
Gleditsch et al. 2002.
	 ethnonationalist triads	 421
rate coding of multiple simultaneous conflicts within single countries.
Using an operational definition of ethnic conflict that focuses on all
conflicts in which the parties are organized primarily along ethnic lines,
Buhaug, Cederman, and Rød identified the megs that were involved in
each ethnic conflict.70
In cases where more than one group challenged
the capital at the same time, an onset of conflict was recorded in each of
the corresponding dyads. Ongoing years of conflict were coded as missing.
In case of a lull in the fighting or a peace agreement that lasts for
at least two consecutive calendar years, the next observation of conflict
in the dyad was coded as a new onset.
Transborder Ethnic Affiliations
Finally, we need to consider the data on external ethnic affiliations in
some detail, because this dimension was not covered by the previous
study. As already mentioned, we rely on the anm’s global group list,
which facilitates the coding of transborder ethnic links. Of course,
there are many other data sources on ethnic affiliation, including Gurr’s
Minorities at Risk (mar) data and Fearon’s more recent list of ethnic
groups.71
Although these data sources may be useful and appropriate
for many analyses or comparisons of individual ethnic groups, the
most commonly used data sources on ethnic affiliation presents many
limitations when it comes to identifying the cross-border presence of
ethnic groups. These data sources typically do not have standardized
group codes that are comparable across countries, and in many cases we
find similar labels used for groups that may lack any common identity
across national boundaries. For example, Fearon’s data identify “Black”
as a separate ethnic group in Brazil, Canada, Nicaragua, and Uruguay,
and—for unknown reasons—the label “Blacks” appears in Ecuador. In
this case, it would clearly be problematic to treat English-, Portuguese-,
and Spanish-speaking blacks as one ethnic group with a transborder
presence. Likewise, the mar data identify “Foreigners” as a minority at
risk in Switzerland, but it would be highly problematic to treat “Foreigners”
as a cohesive group and assume linkages to communities in
other states without further information about the specific composition
of the group.
	Although we believe that the anm data offer certain advantages over
the alternatives, the list of ethnic groups in the anm data contain some
of the problems noted above with regard to transborder comparability.
70
Buhaug, Cederman, and Rød 2008.
71
Gurr 1993; Fearon 2003.
422	 world politics
We have therefore made some changes to the data to facilitate meaningful
cross-national comparisons. In particular, all the ethnic groups
in Switzerland (that is, French Swiss, German Swiss, and Italian Swiss)
are listed as separate ethnic groups in the anm, distinct from their kin
in their respective nation-states, while Germans in Belgium and Italians
in Yugoslavia are considered “the same” as their kin in their respective
nation states. For consistency, we recode all the Swiss ethnic groups
to the same group identifier as the ethnic groups in their respective
nation-states and other neighboring countries. Likewise, Austrians are
considered a distinct ethnic group from Germans in the anm, and we
recode all instances of this in Austria and elsewhere to Germans. Furthermore,
Jews in Israel are considered a distinct ethnic group from
Jews in other states, for example, Russia or the USSR. We recode all
Jews as the same ethnic group.
The most fundamental change that we make to the anm data pertains
to the role of Arabs. The anm data code Arabs from different states in
the Middle East and North Africa as separate ethnic groups. Although
we recognize the wide variation in spoken Arabic and that the issue of
a common Arab identity is controversial among Arabs, it seems problematic
to assume that Palestinians in Israel and Arabs in Iran have no
ties to neighboring communities. Moreover, many other encompassing
categories are included as single groups in the anm data, including
Chinese, despite the large differences and lack of mutual intelligibility
between dialects such as Mandarin and Cantonese within the greater
Chinese macrolanguage. We therefore include all Arab groups as a single
ethnic group to ensure meaningful border-transgressing linkages.
Empirical Estimates and Results
To test our hypotheses on the onset of ethnic conflict, we consider a
series of regression models. Our basic unit of analysis is a basic centerperiphery
(bcp) dyad composed of an excluded peripheral group (meg)
paired against the center or government of a state (egip). We limit ourselves
to ethnic groups that encompass at least fifty thousand individuals
in a state. Our dependent variable is the onset of an ethnic conflict
in a peripheral group, as defined in the previous section. We consider
two different samples in our analysis. Our first is a static data set for
which we have one observation per dyad. We investigate whether the
dyad experiences conflict at any time during the period 1945–2005 (or
the period of a state’s existence for dyads in countries that cease to exist
as independent states prior to 2005), with values for the independent
	 ethnonationalist triads	 423
variables based on the characteristics prevailing at the beginning of the
period.
Static Analysis
Our static data contain 541 center-periphery dyads in sixty countries
in Europe and Asia. Panel data often tend to violate the assumption
of constant variances, given the stacked structure of the data. Our data
have a panel structure where not only are the individual observations
pooled over time for each dyad and across space, but multiple dyads
may also originate within the same country.This in turn makes it likely
that we may have large variation across individual dyads and countries,
and we therefore report Huberized robust standard error estimates,
clustering on countries.72
Our first model—model 1 in Table 1—is essentially a replication of
the empirical operationalization of the bcp model proposed by Buhaug,
Cederman, and Rød.73
The right-hand-side variables considered include
the demographic balance of the peripheral group to the egip, the
distance of the group from the capital, the extent to which the group
populates mountainous terrain, and the per capita income of the country
at large. More specifically, the measure of the dyadic power balance
between the egip and the marginalized group is operationalized using
the share of the periphery’s population sp relative to the population of
the egip sc, or more precisely ln[sp /(sp+sc)]. Relatively larger excluded
groups have higher rates. The measure is logged due to the skewed
nature of the ratio, as most peripheral groups are considerably smaller
than the center. Furthermore, using the logged rate implies that the effect
of further increases in group size declines as the size of the group
grows larger. As specified above, the distance from the capital is measured
using the logged distance from the geographical centroid of the
polygon for each ethnic group to a country’s capital. The mountain
share measure indicates the proportion of the polygon for an ethnic
group that is composed of “mountainous terrain.” Finally, we consider
the national-level per capita income, given the attention that this has
received in the cross-country comparative literature.74
72
Most of our explanatory variables are dyad specific, and it could be argued that we should cluster
on individual dyads or consider possible differences across dyads rather than countries. Group-level
clustering leads to standard error estimates generally much smaller than those found for clustering on
countries. Hence, the country-level robust standard error estimates should be regarded as the more
conservative estimates.
73
Buhaug, Cederman, and Rød 2008.
74
We refer to Buhaug, Cederman, and Rød 2008 for further details and discussion of data sources
and definitions for these control variables.
424	 world politics
The results for model 1 shown in Table 1 are very similar to those
reported by Buhaug, Cederman, and Rød,75
thus confirming hypotheses
H1, H2a, and H2b. Dyads with larger peripheral ethnic groups are
much more likely to see the onset of an ethnic conflict. Furthermore,
violent conflict is significantly more likely for dyads with excluded ethnic
groups that are far from the center or the capital of the state. Groups
that populate a greater share of mountainous terrain are also more likely
to be involved in civil conflict. At the dyadic level, once we have taken
into account the demographic balance and the geographical position of
the group, national-level gdp per capita income no longer has a statistically
significant negative impact on the likelihood of conflict.
75
Buhaug, Cederman, and Rød 2008.
Table 1
Logit Estimates of the Probability of Ethnic Conflict Onset,
Static Samplea
Variable	 Model 1	 Model 2	 Model 3	 Model 4
Group-Level Variables
Ln demographic balance 4	 0.471	 0.489	 0.254	 0.260
	 (0.135)***	 (0. 143)***	 (0. 189)*	 (0.185)*
Ln distance from capital	 0.506	 0.487	 0.481	 0.503
	 (0.189)***	 (0.188)***	 (0.180)***	 (0.201)***
Mountain share	 0.810	 0.781	 0.829	 0.848
	 (0.388)**	 (0.377)**	 (0.424)**	 (0.433)*
Transnational Variables
Contiguous transnational group		 –0.289	 2.341	 2.249
		 (0.466)	 (1.326)**	 (1.294)**
Ln dem. balance X contiguous group			 0.658	 0.653
			 (0.265)***	 (0.260)***
Transnational group EGIP				 0.226
				 (0.631)
Country-Level Variables
Ln GDP per capita (lagged)	 –0.122	 –0.106	 –0.157	 –0.176
	 (0.242)	 (0.256)	 (0.252)	 (0.239)
Intercept	 –2.816	 –2.592	 –3.414	 –3.399
	 (2.273)	 (2.282)	 (2.430)*	 (2.403)*
N	 541	 541	 541	 541
Wald χ2
	 21.38	 21.72	 26.21	 26.93
* p < 0.1; ** p < 0.05; *** p < 0.01 (one-tailed)
a
Entries are logit coefficient estimates, with standard errors in parentheses.
	 ethnonationalist triads	 425
As a test of hypothesis H3, model 2 introduces a dichotomous measure
of whether the excluded ethnic group has transborder kin in another
state. For reasons explained above, we limit ourselves to groups
that are located in states within five hundred kilometers of one another,
using data on the minimum distances between states.76
We also disregard
groups that have a small transborder contingent, encompassing
fewer than fifty thousand individuals, and code these as not having
a cross-border presence. As can be seen, the coefficient estimate for
the simple dichotomous indicator of transborder presence is actually
negative, although not statistically significant. Hence, as we expected,
our results do not provide any support for H3, that is, the idea that
transborder excluded groups are generally more likely to rebel, even
after taking into account other dyadic characteristics that make groups
more likely to become involved in violence against the center of a state
or the government. Comparing the likelihood ratios or the Wald χ2
for the two models indicates that adding the dichotomous measure for
whether groups have a cross-border presence does not yield a statistically
significant increase in the overall fit of the model.
As specified by hypothesis H4, model 3 also includes an interactive
term between the dyadic demographic balance and whether a group
has a transborder presence. This specification reflects our argument
that the border-transgressing linkages can increase the risk of conflict,
but only conditional on dyadic characteristics likely to be associated
with conflict. As can be seen, we find a large significant positive coefficient
for the interactive term. Substantively, the results for model 3
imply that although larger groups are more likely to become engaged
in violence with the center, the effect of increases in the demographic
power balance is much greater for groups that have a cross-border presence.
Comparing the Wald χ2
for models 1 and 3 indicates that the
addition of the terms for contiguous transborder groups and the interaction
between border-transgressing groups and the demographic
power balance makes a significant contribution and improves notably
on the baseline model 1.This provides strong support for our argument
that transborder links can dramatically increase the risk of civil war,
but only in dyads where the transborder link combines with domestic
characteristics that make violence more likely.
The net impact of a variable that enters a regression model with an
interactive term will depend on the value of the other variable in the
interactive term, as well as the scaling of the two individual variables or
76
Gleditsch et al. 2008.
426	 world politics
raw components of the interactive term.77
Figure 3 graphically presents
the net effects of differences in the demographic ratio on the predicted
probability of conflict in a center-periphery dyad implied by the baseline
model 1 and model 3 with the interactive term, for groups with and
without a cross-border presence. All other covariates in the model are
held at their median values. The thin dashed line indicates the predictions
from the baseline model 1, where we disregard whether groups
have a transborder presence. The longer dashed line indicates the predicted
probabilities from model 3 for a group without a transborder
presence in any neighboring country, while the solid line indicates
how the predicted probabilities from model 3 change with the demographic
balance for a group with kin in a neighboring state. As can be
seen, the increase in the risk of civil war in larger group sizes is indeed
much more pronounced for groups that have transborder kin in model
3. Moreover, model 3 yields notably higher predicted probabilities for
relatively larger groups with a cross-border presence than the purely
domestically oriented model 1. Consider, for example, the case of the
Kurds in Turkey, where the center-periphery dyad has a demographic
balance ratio of about 0.136. Based on the other values for this dyad,
the estimates for model 1 imply a predicted probability of conflict of
about 0.32. By contrast, model 3 implies a higher predicted probability
of conflict—that is, 0.43—given the presence of Kurds in neighboring
Iran, Iraq, and Syria. The appendix lists all groups with transborder kin
involved in conflict, ordered by the demographic balance ratio.
As foreseen by hypothesis H5, model 4 adds a term indicating
whether the peripheral group in a dyad is an egip or ethnic group in
power in at least one of the contiguous states in which it is present.
Although the results indicate a positive coefficient, this does not reach
statistical significance and is not distinguishable from zero.This is consistent
with our discussion above.
So far we have seen that the terms for border-transgressing characteristics
in model 3 when added to the purely domestic dyadic baseline
model 1 are clearly statistically significant, and that the introduction of
the transborder features in model 3 can generate quite different predictions.
However, is there any evidence that the predictions of model
3 are generally more accurate than those of model 1, in the sense of
assigning higher predicted probabilities to cases in which we observe
conflict and assigning lower probabilities to those in which we do not?
77
See Braumoeller 2004.
	 ethnonationalist triads	 427
Neither model 3, with the interaction between demographic balance
and transborder kin, nor the baseline model 1 returns predicted
probabilities above 0.5—that is, instances where conflict is predicted to
be more likely than not—for any dyad where we actually observe con-
flict.78
However, since conflict is a rare event that occurs in only about
10 percent of the 541 dyads in our sample, it may be more appropriate
to consider whether the model does a good job of identifying a higher
likelihood of conflict and helps in discriminating between those dyads
within which we actually see conflict and those within which we do
not. If an event is very rare, we can typically maximize the share of
78
The baseline model 1 does not generate predicted probabilities above 0.5 for any groups. Model
3 does yield predicted probabilities above 0.5 for two dyads (Jhats/Awans in Pakistan and Russians in
Kazakhstan), but neither of these groups is actually involved in conflict.
Figure 3
Predicted Probability of Conflict by Demographic Balance Rate for
a Median Dyad Profile, Static Sample
0.8
0.6
0.4
0.2
0.0
Pr.ConflictOnset
0.0 0.2 0.4 0.6 0.8
Demographic Balance Rate
model 4 w/ transnational group
model 4 w/o transnational group
base model 1
428	 world politics
correct predictions by simply assuming that the event will never occur.
In practice, however, we generally prefer to know whether certain
features are associated with the rare events and therefore help identify
when these events are more likely.The State Failure Task Conflict Task
Force, for example, considered a prediction threshold of 0.25 in its efforts
to forecast failing states. For a prediction threshold of 0.25, model
3 correctly identifies nine out of the fifty-five dyadic conflicts, while
model 1 correctly identifies only five conflicts.79
From a decision-theoretic viewpoint, the appropriate prediction
threshold C depends on the relative costs of a false negative or on a case
for which a true event is not predicted and the costs of a false positive
or an incorrect prediction of the event in an instance where it does not
happen. A low threshold minimizes the missed events but risks generating
an unacceptable number of false positives. Rather than choosing
a single-prediction threshold, we could evaluate whether there are systematic
differences between models across a range of plausible thresholds.
King and Zheng suggest using Receiver-Operating-Characteristic
(roc) plots to compare the performance of two competing models over
a range of prediction threshold Cs.80
An roc plot plots a continuous
curve comparing the share of true and false positives from a model
for a given prediction threshold. The performance of a model relative
to a random guess based on the share of events in the sample can be
assessed by comparing the height of the curve to a 45 degree line. Figure
4 shows roc plots for model 1 (dashed line) and model 3 (solid
line). With a few isolated exceptions, the curve for model 3 is generally
higher than that of model 1, and the difference is quite substantial
in most of the relevant areas of the curve, suggesting that transborder
characteristics provide important information in identifying the center
periphery dyads within which conflict is more common.
Dynamic Analysis
So far we have considered only a static analysis where each centerperiphery
dyad is observed once and we consider whether conflict occurs
at any point. However, conflicts may be more or less frequent, and
many of the features in our model can change over time. In particular,
79
More specifically, in addition to those conflicts identified by model 1, model 3 correctly predicts
conflict in the center periphery dyads involving Azerbaijanis and Kurds in Iran, Kurds in Turkey,
Baloch in Pakistan, and Shan in Myanmar. All of these groups have transnational kin. The number of
false positives is the same for both models, so model 3 improves on the predictions of model 1 without
increasing the number of incorrectly called conflicts.
80
King and Zheng 2001.
	 ethnonationalist triads	 429
groups may become included or excluded, and demographic balances
may be changed by border changes and changes in the state system.Table
2 replicates models 1–4 for a sample of annual dyadic observations,
where we consider how the covariates in the model predict the onset of
conflict in a given year. Since the risk of conflict is likely to be highly
dependent on whether a dyad has experienced conflict previously, we
include a count of the length of time that a dyad has remained at peace,
using a cubic smoothing spline to allow for possible time dependence
as suggested by Beck, Katz, and Tucker.81
We also include a linear time
trend. As can be seen from Table 2, the conclusions from the analysis
based on the static sample are generally upheld in the dynamic analysis,
with the exception that the negative coefficient for the log of per capita
gdp now becomes significantly negative in the baseline model without
the border-transgressing variables.
81
Beck, Katz, and Tucker 1998.
0.0 0.2 0.4 0.6 0.8 1.0
False Positive Rate
TruePositiveRate
1.0
0.8
0.6
0.4
0.2
0.0
Figure 4
ROC Plot for Predictive Success Based on the Static Sample
baseline
w/ transnational ties
430	 world politics
More specifically, model 6 with the dichotomous term for whether
groups have a transborder presence still indicates that conflict is not
generally more likely for such groups, again disconfirming hypothesis
H3. However, the interactive specification in model 7 still suggests a
strong interactive effect between demographic balance and cross-border
Table 2
Logit Estimates of the Probability of Ethnic Conflict Onset,
Dyadic Samplea
Variable	 Model 5	 Model 6	 Model 7	 Model 8
Group-Level Variables
Ln demographic balance r	 0.399	 0.430	 0.224	 0.215
	 (0.135)***	 (0. 145)***	 (0. 190)	 (0.189)
Ln distance from capital	 0.602	 0.547	 0.545	 0.508
	 (0.135)***	 (0.124)***	 (0.119)***	 (0.115)***
Mountain share	 1.135	 1.114	 1.012	 0.955
	 (0.305)***	 (0.310)***	 (0.287)***	 (0.262)***
Transnational Variables
Contiguous transnational group		 –0.446	 1.716	 1.892
		 (0.420)	 (1.273)*	 (1.285)*
Ln dem. balance X contiguous group			 0.574	 0.595
			 (0.250)**	 (0.256)**
Transnational group EGIP				 –0.415
				 (0.434)
Country-Level and Control Variables
Ln GDP per capita (lagged)	 –0.289	 –0.274	 –0.301	 –0.288
	 (0.263)*	 (0.282)	 (0.299)	 (0.289)
Year	 0.065	 0.067	 0.064	 0.065
	 (0.010)***	 (0.011)***	 (0.011)***	 (0.011)***
Peaceyears	 –0.271	 –0.267	 –0.261	 –0.259
	 (0.067)***	 (0.062)***	 (0.065)***	 (0.066)***
Spline1	 –0.001	 –0.001	 –0.001	 –0.001
	 (<0.001)***	 (<0.001)**	 (<0.001)***	 (<0.001)**
Spline2	 <0.001	 –0.001	 –0.001	 –0.001
	 (<0.001)**	 (<0.001)**	 (<0.001)**	 (<0.001)**
Spline3	 >–0.001	 <0.001	 <0.001	 >–0.001
	 (<0.001)	 (<0.001)	 (<0.001)	 (<0.001)	
Intercept	 –132.906	 –135.398	 –130.995	 –133.020
	 (18.820)***	 (19.621)***	 (21.032)***	 (20.125)***
N	 22,961	 22,961	 22,961	 22,961
Wald χ2
	 374.39	 383.71	 521.36	 615.78
* p < 0.1; ** p < 0.05; *** p < 0.01 (one-tailed)
a
Entries are logit coefficient estimates, with standard errors in parentheses.
	 ethnonationalist triads	 431
presence, thus lending support to hypothesis H4. Finally, with respect
to hypothesis H5, the coefficient estimate for whether the group is egip
in any of the neighboring states now actually returns a negative coefficient
estimate, although this is not statistically significant.
Figure 5 displays the implied predictions for the dyadic model for a
median dyad profile. Conflict is a very rare event in our sample of annual
observations (less than 0.5 percent of the observations), and given
the strong time dependence, the predicted probability of conflict will be
low in a dyad with the median value of successive years of peace in our
sample, which is twenty-four years. However, although the absolute
predicted probabilities of conflict are obviously much lower, the relative
differences between the predictions for different dyad profiles given
domestic and cross-border specifications are similar to what we found
in the static analysis.
We have also conducted a number of robustness checks. First, we
have examined a version of model 3 using the size of the transborder
group rather than the binary indicator to examine whether the risk of
conflict increases with the absolute size of the transborder group. We
find no evidence, however, for the specification according to which the
effect depends on the size of the border-transgressing component of
the group. Second, we have experimented with alternative group size
thresholds to see whether our results depend on the specific criterion.
Doubling the cutoff threshold to one hundred thousand, however, does
not lead to results that differ notably from those found with a threshold
of fifty thousand. Third, we have reestimated the model using all
transborder groups rather than just groups in countries within five hundred
kilometers of the outer boundaries of countries. In our Eurasian
sample only a relatively low share—about 5.5 percent—of the observations
with transborder groups have no kin in a neighboring state. As
such, the results do not look remarkably different when we include all
groups, although we strongly believe that geographical dimensions are
very important. Fourth, to check whether our results might reflect the
capabilities of states where kin groups are located, we also conducted
additional robustness tests adding the capabilities of the kin states as
well as the ratio of kin state capabilities to those of the countries in
question. The capabilities of kin states are in our view a poor proxy for
the potential capabilities of transborder groups, since there is no reason
to expect that ethnic groups can rely on the resources of states where
their kin are located. However, adding this does not change our main
results—the coefficient estimate for the capabilities of states with kin is
negative, but not statistically significant.
432	 world politics
Conclusion
In this article we have demonstrated that border-crossing ethnic affiliations
have a considerable impact on the likelihood of ethnonational
civil wars. Furthermore, our findings indicate that rather than being
automatic and given by the presence of matching ethnic identities, this
effect depends directly on the power balance in the primary conflict
dyad.
We believe that this research represents considerable progress as
compared with the prevailing quantitative literature, whose statecentric
research designs have generally failed to capture triadic ethnonationalist
configurations. By tracing conflict at the dyadic level, our study
disaggregates the analysis to the group level, while at the same time
extending the focus across state borders. In doing so, the present apFigure
5
Predicted Probability of Conflict by Demographic Balance Rate for a
Median Dyad Profile, Dynamic Sample
0.06
0.05
0.04
0.03
0.02
0.01
0.00
Pr.ConflictOnset
0.0 0.2 0.4 0.6 0.8 1.0
Demographic Balance Rate
model 4 w/ transnational group
model 4 w/o transnational group
base model 1
	 ethnonationalist triads	 433
proach enables more precise testing of relational mechanisms in triadic
ethnic configurations that have thus far been only explicitly treated by
qualitative research on ethnonationalism.
Although these results are promising, they are not intended to suggest
that the last word has been had about the influence of ethnic kin
on conflict. Future research will tell us whether our conclusions are robust
beyond the current sample, which is limited to Eurasia and North
Africa. In principle, it is possible to incorporate other parts of the world
using the current data, but the difficulties of establishing reliable coding
of ethnic-kin relationships, especially in sub-Saharan Africa, are so
serious that we deem it more fruitful to rely on alternative data sources.
Thanks to the new data set on Ethnic Power Relations, we will be able
in the future to extend the sample to the rest of the world while at the
same time limiting it to politically relevant ethnic groups and assessing
their access to power over time in a more precise manner than was possible
in this study.82
The geographic dimension could also be explored
more fully. We would expect transborder support to be dependent on
geographic factors, very much as seems to be the case in domestic
dyads. Such possibilities include the effect of borders, distances between
relevant groups, and terrain. More generally, additional research
is needed on the details of the border-transgressing bond, especially
as regards the nature of the actor-specific mechanism. Disaggregated
research designs, such as those employed by Cunningham, Gleditsch,
and Salehyan,83
will be able to tell us what drives conflict—whether
refugee camps, arms smuggling, demonstration effects, irredentist foreign
policies, or other types of external interventions.
For the time being, however, we conclude that ethnicity matters in
civil wars and that its influence is felt beyond state borders. Attempts
to overlook such mechanisms are likely to block scientific progress and
may lead to dangerously distorted policy advice.
82
See Min, Cederman, and Wimmer 2008.
83
Cunningham, Gleditsch, and Salehyan 2009.
434	 world politics
Appendix
Groups with Transborder Kin (+50,000) Involved in Conflict,
Ordered by Dyadic Power Balance
Country	 Group	 Dyadic Power Balance
Israel	Arabs (Palestinians)	 0.364
Macedonia	Albanians	 0.208
Iran	Azerbaijanis	 0.192
Iraq	 Kurds	 0.174
Iran	 Kurds	 0.160
Turkey	 Kurds	 0.136
Iran	Arabs	 0.112
Moldova	Ukrainians	 0.109
Yugoslavia	Albanians	 0.088
Myanmar	Shan	 0.083
Myanmar	 Karen	 0.073
Croatia	Serbs	 0.071
Pakistan	 Baloch	 0.070
Spain	 Basques	 0.069
Azerbaijan	Armenians	 0.045
Myanmar/Burma	Mon (Talaing)	 0.037
USSR	Azerbaijanis	 0.035
Georgia	Ossetians	 0.028
Myanmar	 Kachins	 0.023
USSR	Armenians	 0.016
Myanmar	 Wa	 0.004
United Kingdom	Irish	 0.002
References
Austvoll, Martin. 2005. “Transnational Ethnic Affinities and Interventions in Internal
Conflicts.” Master’s thesis, University of Oslo.
Ayres, R. William, and Stephen M. Saideman. 2000. “Is Separatism as Contagious
as the Common Cold or as Cancer? Testing the International and Domestic
Determinants of Secessionism.” Nationalism and Ethnic Politics 6, no.
3: 92–114.
Beck, Nathaniel, Jonathan N. Katz, and Richard M. Tucker. 1998. “Taking Time
Seriously:Time-Series Cross-Section Analysis with a Binary Dependent Variable.”
American Journal of Political Science 42, no. 4: 1260–88.
Belanger, Louis, Érick Duchesne, and Jonathan Paquin. 2005. “Foreign Interventions
and Secessionist Movements: The Democratic Factor.” Canadian Journal
of Political Science 38, no. 2: 435–62.
Boulding, Kenneth. 1962. Conflict and Defense: A General Theory. New York:
Harper and Row.
Braumoeller, Bear F. 2004. “Hypothesis Testing and Multiplicative Interaction
Terms.” International Organization 58, no. 4: 807–20.
	 ethnonationalist triads	 435
Brubaker, Rogers. 1996. Nationalism Reframed: Nationhood and the National Question
in the New Europe. Cambridge: Cambridge University Press.
———. 2004. Ethnicity without Groups. Cambridge: Harvard University Press.
Bruk, Salomon I., and V. S. Apenchenko, eds. 1964. Atlas Narodov Mira. Moscow:
Academy of Science USSR.
Buhaug, Halvard, Lars-Erik Cederman, and Jan Ketil Rød. 2008. “Disaggregating
Ethno-Nationalist Civil Wars: A Dyadic Test of Exclusion Theory.” International
Organization 62, no. 3: 531–51.
Buhaug, Halvard, and Kristian Skrede Gleditsch. 2008. “Contagion or Confusion?
Why Conflicts Cluster in Space.” International Studies Quarterly 52, no.
2: 215–33.
Cederman, Lars-Erik. 2002. “Nationalism and Ethnicity.” In Walter Carlsnaes,
Thomas Risse, and Beth Simmons, eds., The Handbook of International Relations.
London: Sage: 409–28.
Cederman, Lars-Erik. 2008. “Articulating the Geo-Cultural Logic of Nationalist
Insurgency. In Stathis N. Kalyvas, Ian Shapiro, and Tarek Masoud,
eds., Order, Conflict, and Violence. Cambridge: Cambridge University Press:
242–70.
Cederman, Lars-Erik, and Luc Girardin. 2007. “Beyond Fractionalization: Mapping
Ethnicity onto Nationalist Insurgencies.” American Political Science Review
101, no. 1: 173–85.
Cetinyan, Rupen. 2002. “Ethnic Bargaining in the Shadow of Third-Party Intervention.”
International Organization 56, no. 3: 645–77.
Chazan, Naomi, ed. 1991. Irredentism and International Politics. Boulder, Colo.:
Lynne Rienner.
Collier, Paul, and Anke Hoeffler. 2004. “Greed and Grievance in Civil Wars.”
Oxford Economic Papers 56, no. 4: 563–95.
Collier, Paul, and Nicholas Sambanis, eds. 2005a. Understanding Civil War: Evidence
and Analysis, vol. 1, Africa. Washington, D.C.: World Bank.
———. 2005b. Understanding Civil Wars, Evidence and Analysis, vol. 2, Other Regions.
Washington, D.C.: World Bank.
Connor, Walker. 1994. Ethnonationalism: The Quest for Understanding. Princeton:
Princeton University Press.
Cunningham, David, Kristian Skrede Gleditsch, and Idean Salehyan. 2009. “It
Takes Two: A Dyadic Analysis of Civil War Duration and Outcome.” Journal
of Conflict Resolution 53, no. 2.
Davis, David R., and Will H. Moore. 1997. “Ethnicity Matters: Transnational
Ethnic Alliances and Foreign Policy Behavior.” International Studies Quarterly
41, no. 1: 171–84.
Ellingsen, Tanja. 2000. “Colorful Community or Ethnic Witches’ Brew? Multiethnicity
and Domestic Conflict during and after the Cold War.” Journal of
Conflict Resolution 44, no. 2: 228–49.
Esteban, Joan, and Debraj Ray. 1994. “On the Measurement of Polarization.”
Econometrica 62, no. 4: 819–52.
Fearon, James D. 2003. “Ethnic and Cultural Diversity by Country.” Journal of
Economic Growth 8, no. 2: 195–222.
Fearon, James D., and David D. Laitin. 2003. “Ethnicity, Insurgency, and Civil
War.” American Political Science Review 97, no. 1: 75–90.
436	 world politics
Gagnon, V. P. 2004. The Myth of Ethnic War: Serbia and Croatia in the 1990s.
Ithaca, N.Y.: Cornell University Press.
Gleditsch, Kristian Skrede. 2007. “Transnational Dimensions of Civil War.” Journal
of Peace Research 44, no. 3: 293–309.
Gleditsch, Kristian Skrede, David Hugh Jones, Doreen Kuse, Michael D. Ward,
and Nils Weidmann. 2008. Minimum Distance Data, version July.
Gleditsch, Kristian Skrede, Idean Salehyan, and Kenneth Schultz. 2008. “Fighting
at Home, Fighting Abroad: How Civil Wars Lead to Interstate Disputes.”
Journal of Conflict Resolution 52, no. 4:479–506.
Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg,
and Håvard Strand. 2002. “Armed Conflict 1946–2001: A New Dataset.”
Journal of Peace Research 39, no. 5: 615–37.
Gurr,Ted Robert. 1968.“A Causal Model of Civil Strife: A Comparative Analysis
Using New Indices.” American Political Science Review 62, no. 4: 1104–24.
———. 2000. Peoples versus States: Minorities at Risk in the New Century. Washington,
D.C.: United States Institute of Peace Press.
Hardin, Russell. 1995. One for All: The Logic of Group Conflict. Princeton: Princeton
University Press.
Hechter, Michael. 1987. Principles of Group Solidarity. Berkeley: University of
California Press.
Hegre, Håvard, and Nicholas Sambanis. 2006. “Sensitivity Analysis of Empirical
Results on Civil War Onset.” Journal of Conflict Resolution 50, no. 4: 508–35.
Heraclides, Alexis. 1990. “Secessionist Minorities and External Involvement.” International
Organization 44, no. 3: 341–78.
Herbst, Jeffrey. 1989. The Creation and Maintenance of National Boundaries in
Africa. International Organization 43, no. 4: 673–92.
Holsti, Kalevi J. 1996. The State, War, and the State of War. Cambridge: Cambridge
University Press.
Horowitz, Donald L. 1985. Ethnic Groups in Conflict. Berkeley: University of
California Press.
Jenne, Erin K. 2007. Ethnic Bargaining: The Paradox of Minority Empowerment.
Ithaca, N.Y.: Cornell University Press.
King, Charles, and Neil J. Melvin. 1999–2000. “Diaspora Politics: Ethnic Linkages,
Foreign Policy, and Security in Eurasia.” International Security 24, no. 3:
108–38.
King, Gary, and Langche Zeng. 2001. “Improving Forecasts of State Failure.”
World Politics 53, no. 4 (July): 623–58.
Kiris¸ci, Kemal, and Gareth M Winrow. 1997. The Kurdish Question and Turkey:
An Example of a Trans-State Ethnic Conflict. London and Portland, Ore.: Frank
Cass.
Kuperman, Alan, and Timothy W. Crawford. 2006. Gambling on Humanitarian
Intervention: Moral Hazard, Rebellion and Civil War. New York: Routledge.
McCarthy, John D., and Mayer N. Zald. 1977. “Resource Mobilization and Social
Movements: A Partial Theory.” American Journal of Sociology 82, no. 6:
1212–41.
Montalvo, José G., and Marta Reynal-Querol. 2005. “Ethnic Diversity and Economic
Development.” Journal of Development Economics 76, no. 2: 293–323.
Min, Brian, Lars-Erik Cederman, and Andreas Wimmer. 2008. “Ethnic Power
	 ethnonationalist triads	 437
Relations: A New Dataset on Access to State Power of Politically Relevant
Ethnic Groups, 1946–2005.” Paper presented at the annual conference of the
American Political Science Association, Boston, August 28–31.
Petersen, Roger. 2002. Understanding Ethnic Violence: Fear, Hatred, and Resentment
in Twentieth-Century Eastern Europe. Cambridge: Cambridge University
Press.
Posen, Barry R. 1993. “The Security Dilemma and Ethnic Conflict.” In Michael
E. Brown, ed., Ethnic Conflict and International Security. Princeton: Princeton
University Press: 103–24.
Posner, Daniel N. 2005. Institutions and Ethnic Politics in Africa. Cambridge:
Cambridge University Press.
Rokkan, Stein. 1999. State Formation, Nation-Building, and Mass Politics in Europe:
The Theory of Stein Rokkan, ed. Peter Flora. Oxford: Oxford University
Press.
Ross, Michael. 2006. “A Closer Look at Oil, Diamonds, and Civil War.” Annual
Review of Political Science 9: 965–300.
Saideman, Stephen M. 2001. The Ties That Divide: Ethnic Politics, Foreign Policy,
and International Conflict. New York: Columbia University Press.
———. 2002. “Discriminating in International Relations: Analyzing External
Support for Ethnic Groups.” Journal of Peace Research 39, no. 1: 27–50.
———. 2007. “Ties versus Institutions: Revisiting Foreign Interventions and Secessionist
Movements.” Canadian Journal of Political Science 40, no. 3: 733–47.
Salehyan, Idean. 2009. Rebels without Borders: Transnational Insurgencies in World
Politics. Ithaca, N.Y.: Cornell University Press.
Salehyan, Idean, and Kristian Skrede Gleditsch. 2006. “Refugee Flows and the
Spread of Civil War.” International Organization 60, no. 2: 335–66.
Sandler, Todd, and James Murdoch. 2004. “Civil War and Economic Growth:
Spatial Dispersion.” American Journal of Political Science 48, no. 1: 138–51.
Shain, Yossi, and Aharon Barth. 2003. “Diasporas and International Relations
Theory.” International Organization 57, no. 3: 449–79.
Taylor, Charles L., and Michael C. Hudson. 1972. World Handbook of Political and
Social Indicators. New Haven: Yale University Press.
Thyne, Clayton. 2007. “Cheap Signals with Costly Consequences: The Effect of
Interstate Relations on Civil War.” Journal of Conflict Resolution 50, no. 6:937–
61.
Tilly, Charles. 1978. From Mobilization to Revolution. New York: McGraw-Hill.
Vanhanen, Tatu. 1999. “Domestic Ethnic Conflict and Ethnic Nepotism: A
Comparative Analysis.” Journal of Peace Research 36, no. 1: 55–74.
Weidmann, Nils B., Jan Ketil Rød, and Lars-Erik Cederman. 2010. “Representing
Ethnic Groups in Space: A New Dataset.” Journal of Peace Research 47.
Weiner, Myron. 1971. “The Macedonian Syndrome: An Historical Model of
International Relations and Political Development.” World Politics 23, no. 4
(July): 665–83.
Wimmer, Andreas. 2002. Nationalist Exclusion and Ethnic Conflict: Shadows of
Modernity. Cambridge: Cambridge University Press.
Zacher, Mark W. 2001. “The Territorial Integrity Norm: International Boundaries
and the Use of Force.” International Organization 55, no. 2: 215–50.