The Origins of Ethnic Power-Sharing Coalitions: How Uncertainty Fosters Cooperation
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
The Origins of Ethnic Power-Sharing Coalitions:
How Uncertainty Fosters Cooperation
Nils-Christian Bormann∗
April 11, 2015
Abstract
Why do ethnic elites share power in government coalitions? Existing explanations
are pessimistic about the prospects of ethnic power-sharing unless institutional incen-
tives encourage cooperation. In contrast, I argue that ethnic leaders usually form gov-
ernments with secure majorities because they anticipate defections by their supporters
and allies, and fear revolutionary threats by excluded groups. Moreover, ethnic elites
prefer coalition partners with whom they share cross-cutting cleavages since defections
by supporters along shared identity markers occur within the ruling coalition. To test
this hypothesis, I collected new data on linguistic, religious, and racial intra-group di-
visions. Using conditional choice models on formation opportunities in 134 ethnically
divided societies between 1946 and 2009 I find that, regardless of regime type, ethnic
elites most frequently opt for oversized multiethnic coalitions that share as many iden-
tity markers as possible. These findings challenge important assumptions about the
nature of ethnic politics and the efficacy of power-sharing institutions.
∗
Center of Comparative and International Studies, ETH Zurich, Switzerland. Email: bor-
mann@icr.gess.ethz.ch
1
Origins of Ethnic Coalitions Bormann
Why do ethnically diverse elites form governmental power-sharing coalitions? This is a
central question for scholars who study the prevention and resolution of civil wars (Lijphart,
1977; Walter, 2002; Hartzell and Hoddie, 2003) and policy-makers who attempt to solve
violent ethnic conflicts in places such as South Sudan and Iraq. Beyond war and peace, the
composition of the ruling coalition influences important outcomes such as economic growth
(Bueno de Mesquita et al., 2003), authoritarian regime stability (Pepinsky, 2009), and pa-
tronage politics (Bates, 1974; Chandra, 2007). Studying the origins of ethnic power-sharing
is of particular interest because conventional wisdom holds that it should be very difficult to
accomplish. Students of ethnically divided societies conceptualize political elites and their
co-ethnic supporters as rational actors who want to maximize their own share of power (e.g.,
Fearon, 1999; Posner, 2005), or even subordinate members of other groups (Horowitz, 2000).
In the absence of institutional rules, such as guaranteed government inclusion, minimum-
winning coalitions or even minority rule should predominate in ethnically divided societies
(Horowitz, 1993; Rabushka and Shepsle, 2008; Roessler, 2011). The resulting large-scale
ethnic exclusion is a fertile breeding ground for violent conflicts (see Cederman, Gleditsch
and Buhaug, 2013).
Yet despite an extensive literature that links so-called power-sharing institutions, such as
proportional representation (PR) or authoritarian parties, to desirable outcomes including
stability and economic growth (e.g., Norris, 2008; Gandhi, 2008), there is relatively little
direct evidence that the same institutions affect the formation of government coalitions. In
fact, Figure 1 shows no evidence for the link between institutions and coalitions that include
political elites from at least two distinct ethnic groups.1 The frequency of ethnic coalitions
(grey bars) in 2009 is larger than the frequency of one-group rule (black) in states without
power-sharing institutions in the preceding year (on the left) while the pattern reverses in
states with power-sharing institutions (on the right). The absence of ethnic coalitions in large
number of countries with power-sharing institutions questions the hypothesized effectiveness
of these institutions. That ethnic elites form coalitions where no power-sharing institutions
exist is even more puzzling from the pessimist perspective that prevails in the literature on
1
Data on power-sharing institutions comes from Cheibub, Gandhi and Vreeland (2010)
and Bormann and Golder (2013). Data on ethnic coalitions comes from the Ethnic Power
Relations dataset by Cederman, Wimmer and Min (2010).
2Origins of Ethnic Coalitions Bormann
ethnic politics (Horowitz, 2000, cf.).
Figure 1: Power-sharing institutions in 2008 and ethnic regime type in 2009.
0.5
One−group rule
Ethnic coalition
0.4
Share of governments in 2009
0.3
0.2
0.1
0.0
No power−sharing Power−sharing
Institutions Institutions
Challenging the common notion that power-sharing institutions are necessary for ethnic
cooperation to emerge, I argue that ethnic government coalitions are far more frequent than
suggested by existing theories because ethnic elites are uncertain about the true distribu-
tion of power. Since group boundaries are fuzzy and support for ethnic leaders is strategic
rather than honest, leaders of ethnic groups operate in an environment of incomplete in-
formation where they fear future revolutions from competing groups and defections from
their own ranks. Anticipating these challenges, ethnic elites form oversized coalitions with
leaders from other groups. Only elites from those groups with an overwhelming majority
of the population can afford to rule alone. I show that governments with secure majorities
rather than minimum-winning or minority governments are the most common form of rule
in ethnically divided societies regardless of regime type.2
My argument builds on recent work on self-enforcing coalitions in weakly-institutionalized
environments (Acemoglu et al., 2008; Driscoll, 2012; Francois, Rainer and Trebbi, 2014). Yet
2
Secure majorities consist of (1) oversized coalitions that include at least two groups
which represent a majority of the population that exceeds minimum-winning size, and (2)
mono-ethnic rule by a group that includes an overwhelming majority of the population.
3Origins of Ethnic Coalitions Bormann
my paper goes beyond these contributions both in its wider empirical scope and in its the-
oretical contributions. Regarding the latter, I extend the focus of these models from the
competition over power between groups with fixed boundaries to contests among ethnic
elites who cannot be completely certain about co-ethnic support because group boundaries
are less rigid. Linking the older sociological literature on cross-cutting and reinforcing cleav-
ages (Lipset and Rokkan, 1990 (1967)) with more recent work on individual ethnic identity
(Posner, 2005; Hale, 2008; Chandra, 2012), I argue that the presence of cross-cutting cleav-
ages enables individuals to shift their support between competing elites by highlighting the
salience of alternative ethnic markers, for example, from language to religion. In turn, elites
anticipate defections by their co-ethnic supporters and attempt to prevent them by inter-
nalizing cross-cutting cleavages into the ruling coalition. Supporter shifts due to changes in
the salient identity markers therefore remain inside the ruling coalition.
To test my hypotheses, I employ conditional choice models from the government forma-
tion literature (Martin and Stevenson, 2001; Glasgow, Golder and Golder, 2012) on potential
ethnic coalitions in 134 ethnically divided states taken from the Ethnic Power Sharing (EPR)
dataset (Cederman, Wimmer and Min, 2010).3 I also introduce new data on multiple lin-
guistic, religious, and racial intra-group cleavages to measure cross-cutting and reinforcing
cleavages within potential coalitions. The statistical tests confirm my two main theoretical
expectations. First, the results indicate that ethnic elites mostly form oversized coalitions
while mono-ethnic regimes are only possible where overwhelming majorities exist. Secure
majorities are by far the most likely type of government in ethnically divided societies while
minority rule is very rare. Second, most coalitions form around cross-cutting cleavages as
ethnic elites anticipate shifts in salient identity markers and erect reinforcing cleavages to
excluded groups. At the same time, the coalition behavior of ethnic elites does not dif-
fer significantly between democratic and authoritarian regimes or under different electoral
regimes.
Studying ethnic coalitions helps explain some puzzling findings in the power-sharing lit-
3
In parliamentary democracies formation opportunities are regulated by well-defined in-
stitutional rules such as elections. In contrast, the ethnic composition of governments in
different regimes form at more diverse occasions in the cases studied here, for example, after
coups. I will discuss the challenges of coding ethnic formation opportunities in greater detail
below.
4Origins of Ethnic Coalitions Bormann
erature. Whereas Mattes and Savun (2009) argue that political power-sharing in the central
government is the most relevant aspect of post-conflict settlements, Jarstad and Nilsson
(2008) report that implemented pacts that guarantee government inclusion have no positive
effect on conflict recurrence. My theoretical focus on uncertainty shows that assurances of
government inclusion should reduce the incentive of elites to cooperate, which explains the
difference between promised and implemented settlements. Additionally, my focus on eth-
nic coalitions allows me to study power-sharing in pre- and post-conflict environments as
well as across regime types. Since ethnicity is so closely related to individual perceptions of
identity (Hale, 2008, Ch.3), the concept of ethnic groups as bases for political power travels
globally. Coalitions between distinct ethnic elites therefore offer a new field for the empir-
ical evaluation of theoretical coalition dynamics beyond the common setting of European
parliamentary democracies (cf. Martin and Stevenson, 2001; Glasgow, Golder and Golder,
2012), for example, by helping to overcome the observation problems that make the study
of secretive authoritarian politics difficult.
Ethnic coalition formation
Why do ethnic elites share power in coalition governments? Almost every theory of coalition
formation draws on Riker’s (1967) formative work that captures the importance of the distri-
bution of power in the “size principle.” According to this logic, utility-maximizing actors aim
to form minimum-winning coalitions, that is the smallest possible coalition that maximizes
their own influence. Later studies of coalition formation in democratic regimes explore other
aspects of coalition formation such as ideological congruence or proposal sequencing but the
distribution of power among the key actors remains central to these studies (e.g., Axelrod,
1970; Baron and Ferejohn, 1989; Laver and Shepsle, 1996).
In the context of ethnically divided societies,4 it follows that ethnic coalitions should only
form where the largest ethnic group does not constitute more than 50% of the population
(Horowitz, 2002, 20). Yet even minority groups might dominate the government when insti-
tutional prescriptions for majority rule are absent such as in democracies that run plurality
4
These are multi-ethnic states where at least some politicians advance ethnic claims in
the national arena.
5Origins of Ethnic Coalitions Bormann
elections and in dictatorships (Horowitz, 2000, 433–434). This pessimistic view underwrites
analyses that attribute political violence and other suboptimal social outcomes to the lack of
cooperation between ethnically distinct groups (see, e.g., Easterly and Levine, 1997; Alesina,
Baqir and Easterly, 1999; Roessler, 2011; Esteban, Mayoral and Ray, 2012).
In reaching these conclusions, existing work on ethnic power-sharing usually builds on
three central assumptions: (1) ethnic elites and their followers are utility-maximizing ac-
tors; (2) authoritarian regimes and some democracies lack commitment technologies, that is
mechanisms that incentivize cooperation across ethnic boundaries; and (3) ethnic groups are
homogeneous and unified actors.5 While I share the first assumption of utility-maximizing
actors (also see Posner, 2005; Chandra, 2012; Francois, Rainer and Trebbi, 2014), I disagree
with the remaining two statements.
With respect to commitment technologies, I assume that violent revolutions and coups
in authoritarian regimes are substitutes for elections and partisan defections in democracies
(also see Bueno de Mesquita et al., 2003; Svolik, 2009; Francois, Rainer and Trebbi, 2014).6
Ethnic elites depend on sufficient support from their co-ethnics and rival elites from other
ethnic groups to obtain power regardless of regime type. While receiving disproportional
attention in academic writing and public discourse, ethnic minority regimes such as As-
sad’s Syria and South Africa during Apartheid should be the exception rather than the rule.
Excluded majorities usually find ways to cooperate against a minority government since
their demographic superiority often comes with additional access to power resources such as
economic or military influence. Majority coalitions have successfully overthrown minority
regimes in Afghanistan, Liberia, Uganda, South Africa, and Zimbabwe, and obtained their
own states in Pakistan and Sudan. Only if an ethnic minority government such as Mestizos
in Guatemala can balance demographic inferiority with vast superiority in coercive capacity
should it be able to defend its position in the long run. Overall though, larger excluded
groups are more likely to violently challenge the government directly than smaller groups
(Buhaug, 2006), and Cederman, Gleditsch and Buhaug (2013, 196–198) show that “stronger
rebels” are more likely to win civil wars. In sum, even in autocracies ethnic elites often
5
I follow Weber (1978) in defining ethnicity as a “putative belief in common ancestry”
that builds on shared identity markers such as language, religion, caste, and race.
6
Conceptually, I distinguish autocracies and democracies by the method of leader selection
and the degree to which elites in most democratic regimes are accountable to their supporters.
6Origins of Ethnic Coalitions Bormann
require the support from a majority of the population.
While the necessity of majority support for governance in dictatorships has been more
controversial than in democracies, several experts of ethnic politics argue that first-past-
the-post democracies are prone to minority rule due to their winner-takes-it-all institutional
setup (Lijphart, 1977; Horowitz, 2000). However, these authors fail to appreciate that there
is no direct relationship between plurality elections and minority rule. In fact, Sartori (1997)
convincingly argues that first-past-the-post elections are essentially equivalent to elections
run under proportional representation (PR) where ethnic groups are spatially separated.7 As
this is the case in many ethnically divided societies in Africa and Asia, I argue that majority
pressures should exist in all types of electoral democracies as well as in authoritarian regimes.
Coalition formation under uncertainty
So far I have argued that ethnic elites are most likely to form majority-sized coalitions no
matter the institutional rules under which they operate. Yet even if ethnic leaders form
majority governments, existing theoretical models of coalition formation predict that these
coalitions should be minimum-winning (Riker, 1967). Therefore, a large number of ethnic
groups should still be excluded from power-sharing pacts. Research on civil war alliances
(Christia, 2012) and government formation in ethnically divided societies (Rabushka and
Shepsle, 2008) provides evidence for this minimum-winning dynamic. However, the con-
ditions under which Riker’s logic holds have received less attention in these analyses. As
discussed above, one central assumption that undergirds the minimum-winning logic is that
ethnic leaders obtain the full support from their co-ethnics and know how large this sup-
port is. With this fixed and indivisible conceptualization of ethnic groups, existing research
presupposes that ethnic elites have complete information over the distribution of power.
In contrast, I argue that ethnic groups are neither fixed nor indivisible units that support
their political elites unconditionally. Although ethnicity is perhaps the most wide-spread
social cleavage that affects politics globally, the boundaries between ethnic groups are neither
impenetrable nor unchangeable (Barth, 1969). Ethnic boundaries are least telling of political
allegiance where identity markers allow membership in multiple sub-groups. Posner (2005)
7
National minority groups constitute local majorities. Also see Calvo and Rodden (2015,
4).
7Origins of Ethnic Coalitions Bormann
argues that individuals stress the one identity marker that guarantees their inclusion in
the smallest possible winning coalition (also see van der Veen and Laitin, 2012). Similarly
political elites compete over predominance within their ethnic group (Pearlman, 2009). This
implies that political elites who draw on the support of their co-ethnics to gain and stay in
power operate in an environment of incomplete information: they only know the approximate
distribution of power gauged from the headcount of different ethnic groups.
There is ample evidence that coalitions in multiethnic states are indeed unstable. Recent
research shows that ethnic voting is less likely when material benefits are neither distributed
along ethnics lines nor excludable (Dunning and Nilekani, 2013; Ichino and Nathan, 2013).
Brass (1968) describes the fluidity of coalitions in Indian state parliaments, and Ferree (2012)
shows that governments in Sub-Saharan Africa that include ethnic groups with majority
sub-segments are unstable. Once the ethnic distribution of power becomes uncertain, the
minimum-winning logic no longer applies:
The uncertainty of the real world and the bargaining situation forces coalition
members to aim at subjectively estimated minimum-winning coalition rather than
at an actual minimum. In decision-systems large enough so that participants
do not know each other or what each is doing, the actual size and weight of a
coalition may be in doubt, if only because of lack of communication or because of
participants’ inability to estimate each other’s weights (Riker, 1967, 77–78).
It is notable how well this quote captures the political competition in ethnically divided
states where communication between members from different ethnic groups is more difficult
than within the group but group membership is still not fixed (Hale, 2008, Ch.3). Even in
states where ethnic tensions run high, group members hardly ever throw their support con-
sistently behind just one ethnic party.8 As a consequence, ethnic elites anticipate defections
from their supporters and allies, and form secure majorities by which I mean single-group
governments with vast numerical superiority such as most monoethnic governments in West-
ern Europe and Latin America, and oversized coalitions elsewhere.
8
The Kurds in Iraq, for example, split their support between two major parties, the
Kurdistan Democratic Party and the Patriotic Union of Kurdistan. Yet sometimes, elites
from both parties cooperate in the Kurdistan Alliance to further their interests vis-a-vis
Sunni and Shi’a Arab elites.
8Origins of Ethnic Coalitions Bormann
My argument does not only challenge the pessimist view on ethnic power-sharing but also
disagrees with alternative theories in which ethnic leaders share their power out of good will
or “statesmanship” as originally argued by Lijphart (1969). Grand coalitions that include
leaders from all groups should be rare as the value of the coalition decreases in its degree
of inclusiveness. In spite of Persian and Russian majority status, power-sharing between
Persian and Azeri elites in Iran and between Russian and Ukrainian leaders in the Soviet
Union are examples of oversized but not grand coalitions as numerous other ethnic groups
are still excluded from power in both states. That the Russian-Ukrainian coalition was not a
result of statesmanship is further supported by Russian elite behavior after the dissolution of
the Soviet Union. The increased relative population share in the new state and the resulting
decrease in uncertainty about ethnic dominance made Russian leaders confident enough to
govern without the inclusion of other ethnic elites into the central government after 1991.
Recent theories of power-sharing in weakly institutionalized environments share more
parallels with my theory by highlighting the fear of future violence as driver for power-
sharing coalitions but they also differ in some important aspects. Acemoglu et al. (2008), for
example, assume that a revolutionary threat only exists at the time of coalition formation
but is not acute thereafter. Their model might therefore underestimate the stability and size
of coalitions. In contrast, Francois, Rainer and Trebbi (2014, 11) assume that all excluded
groups join a revolution once it breaks out. These authors likely overestimate inclusiveness
and underestimate the risk of civil war as they find a pattern of near-grand coalitions in
fifteen Sub-Saharan African countries.
In sum, I predict that in ethnically divided societies, oversized coalitions should be more
common than either minority, minimum-winning, or grand coalitions, and that this pattern
should hold across different regime types.
Coalition formation and ethnic cleavage configurations
Thus far, I have argued that elites will form oversized coalitions because they fear future
defections. However, in almost all states several coalition options exist across multiple real
and latent cleavages that are oversized but do not include elites from all groups. Which
coalitions will elites then choose?
9Origins of Ethnic Coalitions Bormann
Lipset and Rokkan (1990 (1967)) pointed to the multi-dimensionality of ethnic identities
and the underlying potential for change in the cleavages that structure political conflict. At
the same time, they stress critical junctures that locked in certain cleavage configurations
in the long run. More recently, a number of scholars have studied which ethnic identity
marker is most salient for individuals under changing political circumstances (Laitin, 1998;
Posner, 2005; Chandra, 2005; van der Veen and Laitin, 2012). The ability of individuals to
support elites who speak the same language at one point in time and other elites with whom
they share religious affiliations at a later point is analogous to shifting the vote between
two ideologically similar candidates in democratic elections. It is unlikely that liberal voters
support a conservative party but they might alternate between social-democratic and other
liberal parties.9 While this shift in support is less consequential in states where ethnic
cleavages are reinforcing, it has important implications for the winning coalition in states
with cross-cutting cleavages (Chandra, 2005; Posner, 2005).
Ethnic leaders fear the restructuring of salient cleavages because it realigns their support
base and therefore threatens their power position. Conversely, excluded elites that represent
smaller groups have every incentive to stress less salient cleavages that cross-cut the border
between the government coalition and the opposition in order to delegitimize their own
exclusion. Should support realignments pose a threat to the power access of governing
elites, it is likely that those leaders threatened by defections anticipate their possibility and
take counter-measures wherever they can.
To illustrate this line of argument, consider the case of India where the Congress Party had
dominated Indian politics since independence on a platform of cross-caste, cross-linguistic/regional,
and cross-religious appeal.10 While Congress was out of government for short spells in the
1970s and late 1980s, the first opposition party that posed an effective challenge to Congress’
dominance, and managed to stay in power for a full legislative term between 1999 and 2004,
was the Bharatiya Janata Party (BJP). The more conservative BJP portrays itself as a de-
fender of the Hindu identity of the Indian state. Its strategy is to drive a wedge in the
9
It is entirely possible that individuals shift their support from elites with an ethnic agenda
to other elites who run on a non-ethnic platform. While more realistic, this possibility is
beyond the scope of this paper.
10
After a split in 1977 the largest emerging faction was known as Congress (I), named
after its leader Indira Gandhi.
10Origins of Ethnic Coalitions Bormann
Congress coalition by polarizing the Hindu-Muslim cleavage in India. Thus, its Hindu-only
message is in principle directed at a significant portion of the Congress electorate by stress-
ing Hindu identity over caste or linguistic markers. In practice though, the BJP has mainly
attracted upper-caste or -class Hindus, and the need for coalitions with smaller, regional
partners has moderated the message of the BJP (Chandra, 2005, 238). Nonetheless the
success of the BJP can be explained in part by taking a chunk of voters out of the Congress
coalition by appealing to overlapping identity markers.11
What does the threat of support realignment imply for the type of government that
ethnic elites form? In anticipation of defections along cross-cutting cleavage constellations,
group leaders attempt to undercut such attempts from the very beginning. Instead of simply
forming the smallest government coalition that promises sufficient support to deter external
challengers, elites also consider the ethnic compatibility of their coalition partners. Where
elites internalize cross-cutting cleavages into the government coalition, support realignment
may shift the balance of power within the coalition but not between the government and the
opposition. This argument suggests that ethnic elites prefer to form coalitions among groups
that share more identity markers with each other and are, thus, internally less diverse to more
heterogeneous coalitions that share several cross-cutting cleavages with excluded groups.12
To conclude the theoretical discussion, I summarize my two central expectations in the
following hypotheses:
H1 : Ethnic elites are more likely to form secure majorities than minimum-winning coalitions,
minority governments, and grand coalitions.
H2 : Ethnic elites are more likely to form a coalition that includes a small number of ethnic
cleavages than a coalition that encompasses a large number.
11
For a discussion of the BJP and changes in the Indian party system refer to Varshney
(2002, 244–5) and Guha (2007, 634–641).
12
Note that this does not mean that elites choose to minimize the number of groups in
the coalition. Consider the case of three groups that only differ linguistically and share no
markers with excluded groups. According to my argument elites should prefer this outcome
to a coalition of two groups that differ on the linguistic and religious dimensions, and each
share identity markers with excluded groups.
11Origins of Ethnic Coalitions Bormann
Research design and data
While there are different ways to model ethnic power-sharing this paper estimates the choice
of ethnic coalitions directly with so-called conditional choice models (Train, 2003). This
approach was originally introduced into political science by Martin and Stevenson (2001)
in the context of parliamentary government formation, and it models the choice situation
of elites in one country at a given point in time by comparing all potential government
coalitions rather than only the observed outcome. In other words, conditional choice models
estimate the choice of a government relative to alternatives in the same country. This
within-country comparison is superior to country-level analyses that estimate the share of
the included population (Ch.5 Wimmer, 2013) or the presence of a coalition government
(Reilly, 2005) because it takes into account the differences in countries’ ethnic configuration.
Consider, for example, France with its dominant French population and fractionalized Chad.
A government in France that only includes French politicians makes up a large percentage
of the population and it is very unlikely that French politicians would form a coalition with
Basque or Corsican elites. Conversely, Chad’s largest ethnic group, the Sara, constitute just
one quarter of the country’s population which is why a coalition with any of the five other
politically relevant ethnic groups is far more likely. Country-level investigations cannot take
the complexity of these different choice environments into account.
The unit of analysis in conditional-choice models is the formation opportunity that in-
cludes between three and over a million government choices depending on the number of
groups in a state.13 Table 1 shows the basic data setup for one formation opportunity in
Iraq after 2003 where a Shia-Kurdish coalition had replaced Saddam Hussein’s Sunni minor-
ity regime. Conditional choice models estimate the probability that one of the alternatives
in Table 1 is selected and do so for each formation opportunity. Put differently, these models
accommodate a highly unbalanced set of coalition choices across formation opportunities.
A large number of choices in one state does not disproportionally influence the estimated
coefficients at the expense of the formation opportunity in a different state with fewer choices
13
The number of coalition choices at a formation opportunity is 2n − 1 as the empty
coalition is excluded. Since the number of choices rises exponentially, a state like Sudan
with 16 relevant groups presents 65535 coalition opportunities compared with 3 in Trinidad
and Tobago. In a state with 20 groups there would be more than one million choices.
12Origins of Ethnic Coalitions Bormann
since each formation opportunity only counts as one unit. Similarly, the estimator accounts
for a changing number of relevant groups within states over time.
Table 1: Ethnic coalition formation opportunities in Iraq.
Formation Coalition Member(s) Pop. Actual
Opportunity Size Coalition
(1) Sunni 0.19 0
(2) Shia 0.63 0
(3) Kurds 0.17 0
(4) Sunni & Shia 0.82 0
(5) Sunni & Kurds 0.36 0
(6) Shia and Kurds .8 1
(7) Sunni, Shia & Kurds 1 0
Using conditional choice models has two additional advantages over alternative estimation
strategies. First, there is no trade-off between coalition-, country-, and group-level inference
that would be present in a regular multinomial logit model on the country-level. Conditional
choice models make it possible to recover the probability of inclusion for individual groups,
for specific coalitions, and general types of governments. Second, relying on the mixed logit
enables me to estimate entire parameter distributions rather than simple point estimates, and
thereby model unobserved heterogeneity that, for example, reflects diverging preferences of
elites.14 Moreover, the mixed logit does not suffer from bias due to violation of the irrelevance
of independent alternative (IIA) assumption (Glasgow, Golder and Golder, 2012).15
I test my hypotheses on a sample of formation opportunities from the updated Ethnic
Power Relations (EPR-ETH) dataset (Cederman, Wimmer and Min, 2010; Cederman, Gled-
itsch and Buhaug, 2013), which codes politically relevant ethnic groups in all states where
leaders make claims on behalf of their groups, or where the state politically discriminates
any group.16 Although political coalitions are usually formed between political organiza-
tions that represent ethnic groups, these organizations are at least implicitly included in
EPR through the coding requirement that demands political relevance. The EPR codebook
14
Consider the different political outlooks by Nelson Mandela and Robert Mugabe.
15
The online appendix includes a more detailed description of the mixed logit model.
16
Discrimination is usually negative as in Apartheid-South Africa or in the United States
before 1965. However, it can also be positive as exemplified by the case of Hispanics in the
United States who became politically relevant through the Voting Rights Act of 1965.
13Origins of Ethnic Coalitions Bormann
notes that “[an] ethnic group is considered politically relevant if at least one political organi-
zation claims to represent it in national politics. . . ” (Min, Cederman and Wimmer, 2008).
While it would be preferable to have data on political organizations, existing datasets usually
only include a much smaller number of groups or states.17 Moreover, some regimes included
in this study legally ban all political organizations. For most military dictatorships, it would
be impossible to collect organizational actor information.
I construct my dependent variable, the actual government choice, from information on the
political access of group representatives to executive power provided by EPR-ETH. When
group representatives are considered as being included in the executive, I code the group
as a participant in the actual government.18 Token membership by ethnic elites but cannot
or do not effectively represent a group does not qualify for an “inclusion” coding. In other
words, only if elites effectively represent their group members, do I count a coalition.
I assess the relative power of groups by their relative size, and coalition size by the sum
of group sizes included in the coalition. Assessing the relative influence of political elites
through the size of their ethnic group is not only common in academic writings on conflict
(Cederman, Wimmer and Min, 2010; Metternich, 2011). In authoritarian regimes ethnic
headcounts during the census are often the basis for bureaucratic appointments, and thus
the basis for relative influence over the distribution of resources (Jackson and Rosberg, 1984;
Diamond, 1988; Bayart, 2009).
Drawing on its constructivist understanding of ethnicity, EPR-ETH codes the reconfig-
uration of ethnic group boundaries. In a number of states, ethnic groups split into smaller
segments – for example, Blacks in South Africa after the end of Apartheid – or merge into
larger ones, for example, along territorial lines.19 Ethnic groups might become or cease to
be politically relevant in national politics at any point in time when elites start or stop to
make claims on behalf of these groups. In the same vein, relative group sizes change over
17
Birnir (2006) codes organizations for a subset of ethnic groups in the Minorities at Risk
dataset by Gurr (2000) which includes a third of the ethnic groups in EPR and does not
have the same temporal reach.
18
Included groups are coded as holding “monopoly” power or being “dominant” in mo-
noethnic governments. Coalition governments feature one or more “senior partners” and
“junior partners.” Refer to Cederman, Wimmer and Min (2010, 93 & 100-1).
19
Horowitz (2002, 20) describes the ethnic polarization of political allegiances between
northern and southern groups in many African states.
14Origins of Ethnic Coalitions Bormann
time, for example, in Lebanon due to differential birth rates between ethnic groups.
I code new formation opportunities whenever the EPR dataset records a change in group
size, in the set of politically relevant ethnic groups in a state, or in the power position of any
group.20 I argue that these events indicate a change in the bargaining environment for ethnic
elites and therefore offer the potential for a renegotiation over the ethnic composition of the
government. The data then include a new formation opportunity that concludes all possible
combinations of ethnic groups, and highlight the realized government as in Figure 1.21 Due
to computational limitations the analysis does not include China, Russia/the Soviet Union,
and India.22 In sum, the sample includes 338,591 potential coalitions across 467 formation
opportunities in 134 ethnically divided states between 1946 and 2009.
My explanatory variables proxy power relations through the number of groups in a po-
tential coalition, and four dummy variables that note whether the largest group in a state
is included, whether one group rules by itself, and whether a coalition is minimum-winning
or oversized.23 In line with H1 , oversized coalitions and dominant rule by one group, what I
call secure majorities, should be more likely than minority- and minimum-winning-coalitions.
In addition, the largest ethnic group in a state is likely to be an important player in the
government formation process, and therefore more likely to be included than not. Finally,
elites should prefer to rule with few other groups over ruling with many coalition partners
in order to maximize their own share of power.
In order to measure cleavages for each potential government constellation, I collected new
data on the linguistic, religious, and racial sub-segments of each ethnic group in the EPR
data.24 As suggested by constructivist theories of ethnicity, individuals usually possess mul-
20
Note that this might include a relative change within the government coalition or the
granting of territorial autonomy to excluded groups that does not affect the ethnic compo-
sition of the government.
21
The online appendix provides an alternative strategy to identify formation opportunities
based on changes in the institutional setup and leadership alternations that likely indicate
a change in the bargaining environment. Analyses based on this sample do not alter the
fundamental conclusions.
22
While the exponential growth in the number of combinations does not allow me to
include Russia and China that each feature more than 40 groups into my dataset, India’s 20
groups and more than 4.4 million potential government choices at 6 formation opportunities
stretch the demands on computing power but do not exceed it.
23
Minority coalitions constitute the residual category.
24
I derived most of the information from the Ethnologue catalog of languages by Lewis
(2009) and the Joshua Project, an online database that codes the religious affiliation of
15Origins of Ethnic Coalitions Bormann
tiple identity markers but not all of them are politically salient at the same time (Chandra,
2012). The politically relevant groups included in the EPR data usually differ from other
groups on at least one but not necessarily on all ethnic dimensions. Using two ethnic groups
from Nigeria as examples, Figure 2 shows that the data provide information on the relative
share of up to three sub-segments per identity marker and ethnic group.25 The Yoruba and
Hausa-Fulani are primarily divided by language but members of both groups adhere to the
Sunni Muslim faith. To assess how much potential exists for members of a group to defect to
elites who stress alternative identity segments, I first count the number of unique cleavages,
that is the number of different ethnic segments minus one, for each potential coalition. In
another operationalization, I divide this count by the total number of unique cleavages in a
state. The higher the count (or the share), the more likely it becomes that the coalition is
to suffer defections (H2 ).
Figure 2: Example of linguistic and religious identity segments for the Hausa-Fulani and
Yoruba in Nigeria.
Hausa-Fulani
Adamawa
Hausa
Fulfulde
Nigerian
Fulfulde
Sunni Muslims
(Maliki)
Yoruba
Yoruba
Sunni Muslims Anglican Other
(Maliki) Christians Christians
To control for the impact of the past, I code the lag of the ethnic composition of the
government by the population share that was also represented in the previous government.
Ethnologue groups. See Joshua Project: Unreached Peoples of the World (2011).
25
Since the racial affiliation of both groups is coded as Sub-Saharan African, Figure 2
omits this dimension.
16Origins of Ethnic Coalitions Bormann
Alternative measures in the appendix replace this variable by a simple dummy that records
whether the government remained the same at a new formation opportunity.26
Analysis
Before turning to multivariate choice models of coalition formation, I begin to explore the
global patterns of ethnic coalitions descriptively. Between 1946 and 2009, ethnic leaders form
coalitions in more than half of all formation opportunities (246). Out of these, Lijphart’s
grand coalition, in which elites of all relevant ethnic groups in a state are represented, is
selected at 76 different formation opportunities. The Central African Republic features the
smallest ethnic coalition between the Yakoma and Mbaka in 1982. It represents just below 9%
of the total population. Figure 3 shows the distribution of all realized government types in all
ethnically divided states over the sample period. Black bars denote monoethnic governments
while grey bars identify ethnic coalitions. The pattern revealed by the bar graph is definite:
the vast majority of ethnic governments are majority governments, and the combination of
grand and oversized coalitions constitute the modal government type. Moreover, the average
share of the included population in monoethnic governments is about 84% while the median
is even higher. In fact only 14% of all mono-ethnic majority governments represent less
than sixty per cent of the population. This is strong evidence that monoethnic governments
tend to present an overwhelming majority of the population and that these governments are
unlikely to ever be challenged by ethnic minority groups for predominance in the state.
Turning to multivariate comparisons, I evaluate my hypotheses in three mixed logit
models presented in Table 2. Each model includes estimates for the average effect, which is
equivalent to effect estimates in common choice models such as the multinomial logit, and its
standard deviation (SD) that provides information on the empirical variance of the choice.
Both of these estimates come with their own uncertainty estimate. Where the estimated
SD is significant and large, a substantial number of elites make choices that differ from the
average.27
26
As all initial coalitions are missing by construction when lagged variables are included
in the analysis, the sample shrinks to 328 formation opportunities.
27
A statistically significant standard deviation does not imply that the estimate of the
average effect is imprecise. An in-depth explanation of the mixed logit can be found in the
17Origins of Ethnic Coalitions Bormann
Figure 3: Realized government types in 134 states, 1946–2009.
As indicated by the positive mean effects for single-group majorities and oversized coali-
tions, governments that include groups representing more than 50% of the ethnically relevant
population in a state are more likely than minority governments (the baseline category). Po-
litical elites that represent majority ethnic groups often opt to govern alone, although in a
substantive minority of cases, they choose not to embrace one-group majority rule as indi-
cated by the significant estimate for the standard deviation of single-member governments.
The strong positive average effect likely reflects the outcome of successful nationalist projects
in which one groups dominates the state, as is the case in many Latin American countries.
Contrary to predictions of general models of coalition formation and experts on ethnic con-
flict, minimal-winning coalitions are not more likely than minority governments. Both the
estimated average and the dispersion effect are insignificant across all three models in Ta-
ble 2. Where ethnic group elites cooperate, they overwhelmingly form oversized coalitions.
Moreover, there is hardly any deviation from this pattern as indicated by the insignificant
estimates for the standard deviation. Overall then, these models provide strong evidence in
support of H1 that expects ethnic elites to form secure majorities.
Turning to the control variables, the estimate for the largest group dummy is positive
appendix.
18Origins of Ethnic Coalitions Bormann
Table 2: Mixed logit models of coalition formation in 133 states, 1946–2009.
(1) (2) (3)
Mean SD Mean SD Mean SD
Single-Group Maj. 3.023∗∗∗ 4.684∗∗∗ 3.365∗∗∗ 5.620∗∗∗ 2.955∗∗∗ 3.340∗∗∗
(0.664) (0.911) (0.718) (1.106) (0.750) (0.979)
M.-W. Coalition -0.017 -0.376 0.378 0.602 -0.336 0.400
(0.458) (0.604) (0.595) (0.765) (1.172) (1.410)
Oversized Coal. 1.579∗∗∗ -0.399 2.231∗∗∗ -0.749 2.327∗∗∗ 0.792∗
(0.345) (0.576) (0.394) (0.491) (0.451) (0.380)
Largest Group 1.215∗∗ 1.505∗∗∗ 1.107∗ 1.592∗∗∗ 0.390 -0.613
(0.426) (0.335) (0.441) (0.434) (0.443) (0.418)
No. of Groups -0.551∗∗ 1.231∗∗∗ -0.199 1.389∗∗∗ 0.011 1.207∗∗∗
(0.186) (0.167) (0.187) (0.288) (0.206) (0.166)
Cleavage Count -0.712∗∗ 1.212∗∗∗ -1.048∗∗∗ -1.149∗∗∗
(0.218) (0.247) (0.269) (0.249)
% of Incumbents 2.553∗∗∗ 1.578∗∗
(0.434) (0.558)
Formation Opportunities 467 467 328
Potential Governments 338591 338591 295288
` -1021.64 -990.213 -727.204
χ2 443.938 451.171 298.480
Country-clustered standard errors in parentheses.
∗
p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
19Origins of Ethnic Coalitions Bormann
in Models 1 and 2 which might point to a formateur advantage (also see Francois, Rainer
and Trebbi, 2014, 1). More often than not, the largest group in a state is included in
the governing coalition, though again there are a substantial number of exceptions as the
significant standard deviation shows. In Model 1, the negative estimate for the number of
groups in a coalition implies that ethnic elites try to maximize their own relative share of
power with respect to other ethnic groups, even while they are building secure majorities.
However, the high variability in the estimate points to consociational leadership behavior in
Switzerland, India, and Ghana.
Once I add a cleavage count variable to the specification in Model 2, the group count
estimate becomes indistinguishable from zero while the cleavage effect is negative and statis-
tically significant. Political elites try to minimize the ethnic connectivity of their coalitions
with excluded groups. The fewer cleavages included in the government coalition, the more
overlap exists between different coalition partners on a subset of the ethnic identity dimen-
sions. When members of the government groups choose to stress alternative identity markers,
they are more likely to shift their support within ruling coalition than to elites of excluded
groups. This result yields support for H2 and calls for further research on the interaction of
ethnic voting, individual ethnic identity, and elite responses to potential defections.
Model 3 adds a variable that measures the population share included in both the previous
and newly formed government. Its positive mean estimate indicates that incumbency exerts
a strong influence on the subsequent ethnic composition of governments. However, the
large and significant estimate of its standard deviation accounts for the instability of ethnic
relations in states like Sierra Leone, the Democratic Republic of Congo, and Afghanistan.
Including an incumbency proxy weakens some of the other regressors – especially the largest
group dummy which is no longer statistically significant – but does not change the substantive
insights regarding H1 and H2 .
Alternative explanations and robustness checks
The most important alternative explanation to the theory advanced by this paper is that
ethnic coalitions result from formal rules that encourage or prescribe elite cooperation. In
addition to the standard government characteristic variables, Model 3 now includes a dummy
20Origins of Ethnic Coalitions Bormann
variable on institutions that is interacted with all government-type variables. Lijphart and
others expect that democracies and PR electoral rules should make minority rule unlikely
while increasing the probability of ethnic coalition formation relative to single-group rule
(Lijphart, 2002; Norris, 2008). Figure 4 displays the distribution of predicted probabilities
across all formation opportunities for various government types on the x-axis. For each type,
there are two probability distributions that distinguish democratic and autocratic systems
(Przeworski et al., 2000; Cheibub, Gandhi and Vreeland, 2010), and electoral systems within
democracies (Bormann and Golder, 2013). The bars inside the boxes indicate the median
predicted probability while the whiskers represent the 90% confidence intervals.
The most striking insight provided by Figure 4a is that there are hardly any differences
between democracies (white boxes) and dictatorships (black) in the overall patterns of gov-
ernment formation. Only minority rule by one ethnic group is more likely in dictatorships
than in democracies, and the difference is statistically significant. The graph also confirms
that ethnic elites are far more likely to form either single-group majority regimes or oversized
coalitions than any other type of government. Again, no notable differences between demo-
cratic and autocratic rule exist within these categories. Although it seems as if oversized
coalitions are even more likely in dictatorships than in democracies, the relationship is not
statistically significant. Whether or not governments are elected has little effect on their
ethnic inclusiveness.28
Turning to institutional variation within democracies, Figure 4b compares majoritarian
and PR electoral systems, and lend little support to the assertion that PR (black boxes)
promotes ethnic power-sharing. To the contrary, majoritarian electoral systems (white)
are more often associated with oversized coalitions and less often connected to oversized
majorities than their PR counterparts. Even more surprisingly, ethnic leaders opt for single-
member majorities significantly more often than for oversized coalitions under PR as the
confidence intervals of either category do not overlap with the median effect of the other.
Uncertainty is again a possible explanation for this finding. In contrast to PR systems, small
changes in the vote distribution can lead to large shifts in parliamentary representation under
majoritarian rule, and thus result in loss of power. Elites hedge against this possibility by
28
Using an alternative index of democracy, the Polity IV measure by Marshall, Jaggers
and Gurr (2011) and adding anocracies as a third category does not change this insight.
21Origins of Ethnic Coalitions Bormann
Figure 4: Estimated government type probabilities in democracies and dictatorships (top
panel) and under PR and majoritarian electoral rules within democracies (bottom panel).
Predicted probabilities by regime type
0.8
Democracy
Dictatorship
Predicted Probability
0.6
0.4
0.2
0.0
Minority Majority Minority Minimum−Winning Oversized
Single−Member Coalition
(a) Democracy data from Cheibub, Gandhi and Vreeland (2010).
Predicted Probabilities by Electoral System
Majoritarian
0.8
PR
Predicted Probability
0.6
0.4
0.2
0.0
Minority Majority Minority Minimum−Winning Oversized
Single−Member Coalition
(b) Electoral systems data from Bormann and Golder (2013).
22Origins of Ethnic Coalitions Bormann
forming oversized coalitions that allow them to continue their rule even if they are losing
support.29
To test the robustness of my results, I rerun the main specifications with a simpler
but biased model, the conditional logit, that is able to compute estimates for the complex
configuration of ethnic coalitions in India. The results do not differ in any meaningful way
from the ones in Table 2.30 The results also remain robust when I use an alternative sample
of formation opportunities that records a potential change in the ethnic composition of the
government whenever major institutional changes occur.31 To ensure that the results are
not driven by cases in which ethnicity has arguably less political salience, I rerun Model 3
by on various subsets of countries. Neither the removal of OECD states, nor the exclusion
of Western countries more generally, or states from the former Soviet Union challenges my
findings. I am thus confident that sample idiosyncrasies are not driving my results.
In alternative models, I test whether the effect of cleavages on coalition formation holds
up when using alternative operationalizations. Rather than the total count I employ the
relative share of cleavages included in the coalition, and continue to find support for my
hypotheses. Additional tests assess whether the cleavage finding is driven by any one ethnic
dimension. Some scholars expect that cooperation across religious divisions should be less
likely (Huntington, 1996; Laitin, 2000; Toft, 2007) while others point to the strong visibility
of racial divisions that impedes cooperation (Caselli and Coleman, 2013). Indicators of
linguistic, religious, or racial fractionalization and polarization do not show any systematic
relation with patterns of ethnic coalition formation.
Conclusion
In this paper, I have argued that governmental power-sharing, the coalition among different
ethnic groups in the executive, is a result of elite uncertainty about future support by their
co-ethnics and coalition partners. In contrast to a largely pessimistic view on the prospects
29
In the appendix, I also explore the variation between autocratic regimes with and without
a ruling party, and with and without a legislature. Once more, I fail to find any evidence for
the purported power-sharing effects of these institutions.
30
While easier to estimate via standard maximum-likelihood techniques, conditional logit
models do not capture the variance of the average effect (Glasgow, Golder and Golder, 2012).
31
See the online appendix for more details.
23Origins of Ethnic Coalitions Bormann
of ethnic cooperation that prevails in the literature, my findings indicate that ethnic elites
form secure majorities, mostly oversized coalitions, due to their desire to obtain and remain
in power in an environment of incomplete information.
This paper does not only answer the question what type of government ethnic elites form
but also addresses which specific coalition will be chosen. My results support and comple-
ment an important theoretical argument from the literature on ethnic politics that argues
that rational individuals stress the identity marker that allows them to be a member of the
smallest possible winning coalition (Posner, 2005; van der Veen and Laitin, 2012). Focusing
on political elites, I argue that leaders anticipate potential defections by their co-ethnics.
Therefore, they attempt to form coalitions between groups with cross-cutting cleavages that
keep supporter switches inside the ruling coalition. Indeed, I find that coalitions that en-
compass a smaller number of cleavages are more likely than those including a larger number.
My elite-centered explanation diverges from previous studies of power-sharing by de-
emphasizing the role of formal institutions in overcoming commitment problems. In fact,
I do not find any evidence that the difference between democracies and dictatorships, or
institutional variation within these regime types affects the likelihood of ethnic coalition
formation. This finding has two major implications for the literature on power-sharing and
conflict research. First, if institutions that are commonly considered to induce cooperative
behavior do not affect ethnic coalition formation, the key mechanism for conflict resolution
is void. As Diermeier and Krehbiel (2003, 127) point out: “It cannot be stressed enough
that (. . . ) behavior within the institution – not just the institution in isolation – deter-
mines whether institutions are outcome-consequential, or, as is more often uttered, whether
institutions matter.” 32
Second, if my theory is correct that coalitions are most likely when leaders are uncertain,
institutions that reduce uncertainty such as guaranteed government inclusion are less likely
to induce cooperation among ethnic leaders. At the same time, existing research argues that
high uncertainty also increases the risk of civil war (Walter, 2009; Mattes and Savun, 2010).
If both these accounts of conflict and my theory were correct, one implication would be that
elites embrace power-sharing when conflict is likely. This poses a dilemma for policy-makers
32
See Pepinsky (2014) for a similar plea to students of power-sharing in authoritarian
regimes.
24Origins of Ethnic Coalitions Bormann
who want to promote power-sharing for normative reasons of political equality. Future
research needs to pay more attention to the conditions under which ethnic coalitions form,
when formal institutions are helpful, and how the interaction of coalitions and institutions
affect civil war risk.
Beyond conflict research, coalitions are a fundamental concept in political science (cf.
Humphreys, 2008) but their quantitative study mostly focuses on parliamentary democracies
in Europe (see Laver, 1998). Studying ethnic coalitions provides one empirical alternative by
which to test the many theoretical models of coalition formation that often predict the for-
mation of minimum-winning governments, and thereby further our theoretical understanding
of coalition formation.
25You can also read
