Powersharing, Protection, and Peace

Scott Gates Peace Research Institute Oslo (PRIO)

Norwegian University of Science and Technology (NTNU)

Benjamin A. T. Graham University of Southern California

Yonatan Lupu

George Washington University

Håvard Strand University of Oslo

Peace Research Institute Oslo (PRIO)

Kaare W. Strøm University of California, San Diego

University of Oslo

Powersharing is often purported to lead to civil peace. We identify three types of powersharing:

inclusive, dispersive and constraining. Centering on the credible commitment problem, we analyze

the mechanisms of power allocation activated by powersharing. We focus on constraining

arrangements, which limit the power of any party or social group, and thereby serve to protect

vulnerable groups. Constraining powersharing institutions create credible commitments to minority

groups and the general public against governmental repression, whereas other types of powersharing

focus on elites. Such institutions limit the incentives for mass mobilization and raise the costs of

conflict, thus reducing the probability of civil war. Drawing on a new global data set of

powersharing institutions, we examine the relationship between powersharing and both the onset and

recurrence of civil war. In contrast to prior findings, our results indicate that only constraining

powersharing institutions have a pacific effect.

To address armed conflict in societies at risk, a common recommendation is to include competing

factions in institutionalized powersharing arrangements (Sisk 1996; Walter 2002; Hartzell and

Hoddie 2003, 2007; Roeder 2005; Mattes and Savun 2009; Wucherpfenning 2013). The hope is

that, ex ante, each faction will see the payoff from peaceful cooperation as superior to the expected

returns from violence, and that, ex post, the rewards from cooperative behavior will sustain this

expectation (Walter 2002; Mattes and Savun 2009). Proponents of powersharing argue that it

activates mechanisms that allow parties to credibly commit to a bargain, thus reducing conflict.

Commitment problems are especially relevant to civil wars because parties other than the

regime often must give up arms as part of a settlement (Walter 1997; Fearon 2004). As a result, at

least one party to a conflict cannot credibly commit to the future division of the benefits (Fearon

1995; Fortna 2003; Powell 2004, 2006). Given the opportunity, one or more of the parties could

attack the other or use repressive tactics to weaken it, thereby altering the division of future payoffs.

Ever-shifting balances of power afford advantage to one player relative to the other over time and

make it tempting to start a conflict when times are right. This might mean starting a civil war when

strong or attacking preemptively in anticipation of a shift in advantage to the opponent.

Given that powersharing promises to minimize the risk of conflict, it is no surprise that such

arrangements have found widespread favor among analysts and peace-makers. Such support,

however, may be overly optimistic. Governments bound by some powersharing arrangements may

nonetheless be able to dis-empower competitors by using repressive strategies that exacerbate social

divisions and conflict or exploit the weaknesses of minority groups. It is critical to specify the

institutions that most significantly constrain governments from abusing less powerful groups, thus

making conflict less likely, yet existing analyses of powersharing have not fully explored this

connection. Institutions such as independent judiciaries and protection of religious freedom, for

1

example, have been shown to reduce repression1 (Keith 2002; Powell and Staton 2009; Lupu 2013),

yet the literature on powersharing has not sufficiently accounted for this. The broader literature on

civil wars consistently finds that "the quality of country governance or institutions" affects the

likelihood of civil wars, yet it remains unclear which types of institutions are most important in this

context (Fearon 2011a, p. 3; Walter 2011).

This paper provides and tests a theory that identifies which types of powersharing institutions

are most likely to reduce the risk of the onset and recurrence of civil conflict. We make four

contributions to the literature on powersharing and peace. First, we identify the mechanisms by

which powersharing institutions allocate power among key actors. Second, we explain which types

of institutions can activate the mechanisms most likely to contribute to peace. The key to this

contribution is our focus on the need to protect the masses. Our argument focuses on constraining

powersharing institutions that limit the power of the dominant party or social group and thus protect

ordinary citizens and vulnerable groups against encroachment and abuse. Many existing definitions

of powersharing omit these institutions. We explain how these institutions alleviate commitment

problems, thus reducing actors’ abilities to renege on bargains, repress their opponents, and foment

conflict. In so doing, we provide a much-needed link between the literature on repression (Davenport

1995, 2007; Moore 1998, 2000) and the powersharing literature. Third, by drawing on the new

1 We use the term repression similarly to Davenport (2007), who defines it as “the actual or

threatened use of physical sanctions against an individual or organization, within the territorial

jurisdiction of the state, for the purpose of imposing a cost on the target as well as deterring specific

activities and/or beliefs perceived to be challenging to government personnel, practices or

institutions” (p.2).

2

Inclusion, Dispersion, and Constraints (IDC) data set,2 which we collected for the purpose of testing

this theory, our research avoids selecting our sample on the basis of a history of civil conflict.

Extant studies have asked whether powersharing institutions affect the recurrence of armed conflict

in post-conflict societies. Yet if powersharing institutions can reduce the risk of conflict recurrence,

they might also reduce the risk of conflict onset in the first place. We assess the relationship

between multiple forms of powersharing on both the occurrence and recurrence of civil conflict. We

do not define powersharing institutions based on prior conflict history, thus mitigating a potential

problem of reverse causality. Finally, by using factor-analytic indices created using these data, we

mitigate some of the measurement problems in existing research.

1. Civil Wars, Commitment Problems, and Powersharing.

Commitment problems often contribute to the causes of civil war and make its termination more

difficult. Unlike in inter-state wars, combatants in civil wars (other than the national government)

often do not retain independent, post-conflict armed forces. This significantly increases the

difficulty of committing to respect a settlement after the parties have laid down their arms (Walter

1997, 2002; Powell 2004, 2006; Lake 2003). A key concern among combatants is the possibility of

repression and other forms of reneging on conflict settlements (Humphreys and Weinstein 2008).

Conflicts are prolonged when the government cannot commit to refraining from such actions

(Fearon 1998, 2004).

Civil war settlements that include powersharing arrangements are often argued to alleviate

the commitment problem (Hartzell and Hoddie 2003, 2007). Powersharing provisions constrain

actors’ ability to renege and thereby decrease the sense of insecurity (Mattes and Savun 2009).

When governments are threatened by opposition groups, they strategically respond by using

2 This data set is described in detail in Appendix 4.

3

repressive tactics (Davenport 1995, 2007; Moore 1998, 2000). A government's use of repression can

lead to popular dissent and civil unrest (Lichbach 1987; Moore 1998). The state's use of repression

(one-sided political violence) and the possibility of ensuing civil war (two-sided political violence)

are therefore closely linked (Besley and Persson 2009). Because domestic minorities and opposition

groups may fear that the government will renege on the settlement, powersharing "provisions are

intended to ensure domestic groups that they will not become victims of discrimination and violence

in the new state" (Mattes and Savun 2009, p. 140).

Existing definitions and operationalizations of powersharing vary significantly, as do the

samples on which these are analyzed. Walter (2002) introduced the disaggregation of powersharing

into political, territorial, and economic forms. Hartzell and Hoddie (2003, 2007) add a category of

military powersharing. Mattes and Savun (2009) and Jarstad and Nilsson (2008) use similar

indicators, but expand the set of cases. Mukherjee (2006) focuses on “political power-sharing”,

which is an arrangement that offers rebel leaders concessions such as government posts and the right

to form their own parties. He distinguishes these arrangements from “military power-sharing”,

which grants top military positions to rebel leaders, “economic power-sharing”, which reserves jobs

for members of the rebel group, and “territorial power-sharing”, which provides the former rebels

with a measure of territorial autonomy. Binningsbø (2013) expands the analysis to include all post-

conflict countries. Roeder (2005) takes a different approach by distinguishing between powersharing

and power-dividing institutions.

Empirical estimates of the effect of powersharing are also mixed (Binningsbø 2013). Most

find that at least some type of powersharing promotes peace, but scholars disagree as to the forms of

powersharing that are (most) effective and the conditions under which positive effects can be

expected (Walter 2002; Hartzell and Hoddie 2003, 2007; Mukherjee 2006; Jarstad and Nilsson 2008;

Mattes and Savun 2009; Wucherpfenning 2013). Roeder's (2005) results differ most fundamentally

4

from the others, indicating that powersharing dyads tend to be more conflict-ridden than others.

While we have learned much from these studies, they have several important limitations.

First, powersharing is typically defined by single institutional variables or by an additive index of

individual institutions constructed without empirical corroboration. If only one institution -- such as

proportional representation -- serves as a proxy for all powersharing arrangements, the variety of

such institutions (and the variety of mechanisms by which they operate) is not taken into account.

On the other hand, tests employing an additive index of powersharing will likely be biased because

this measurement procedure assumes the different components are independent from each other and

that their relative weights can be accurately specified a priori (and are generally equal). To address

this problem, we both expand and disaggregate the definition of powersharing.

Second, powersharing institutions tend to be defined in terms of the type of prior conflict

they address, such as conflict over territory or conflict over governance (Walter 2002; Hartzell and

Hoddie 2003, 2007). Yet the distinction between territory and government mixes the solution and

the problem, such that the competitive stakes (territorial autonomy or governmental authority) are

subsumed within the definition of powersharing institutions. This categorization of powersharing

institutions endogenizes them to conflict rather than causally linking them to conflict. We use

measures of powersharing that are not coded based on prior (or future) conflicts.

The third limitation in existing research is the focus on post-conflict states. Most existing

studies examine powersharing strictly as a strategy for resolving an existing conflict and preventing

it from recurring, which is often appropriate depending on the research question. If enduring

powersharing institutions have an effect on conflict, they might also affect the probability that a

conflict begins in the first place. Previously, data on powersharing institutions were previously not

available for non-post-conflict societies. We use a data set that includes all polities, regardless of

whether powersharing institutions were created before or after civil conflict.

5

Fourth, existing work focuses largely on elites. Yet opposition elites cannot start or re-start

civil war without some measure of public support. The effectiveness of powersharing institutions

depends on the extent to which they create credible commitments to protect the interests of both

elites and publics. We address this limitation by broadening the definition of powersharing to

include institutions that facilitate the sharing of power between the government and the masses and

by explaining how such institutions can resolve the commitment problem.

2. A Theory of Powersharing, Protection, and Peace.

2.1 Elites, Masses, and the Commitment Problem. Political elites are actors that can organize

violence, either through the state or through a rebel organization. Opposition Elites engaged in

organizing armed civil conflict have goals that, broadly speaking, tend to break down into two types:

to control the political center of a state or to control a non-central part of the state (which may entail

a demand for political independence). Most analyses of the commitment problem in the civil war

context focus on the government and elites. After opposition elites lay down their arms, it may be

difficult for the government to credibly commit to abiding by the terms of the peace settlement.

Yet the masses also play an important and, in the powersharing literature, often overlooked

role in the commitment problem. While opposition elites are often instrumental in leading armed

conflict against the government, they depend critically on masses for recruits (Collier 2000; Gates

2002) and for support more generally (Mukherjee 2006; Gates and Strøm 2013). A rich research

tradition has therefore explored patterns of participation in civil conflicts by the mass public (Olsen

1965; Popkin 1979; Skocpol 1979; Kalyvas and Kocher 2007). Opposition elites who might seek to

challenge the government via armed conflict face many costs. In some situations, the costs of

fighting are too large or elites lack the capacity to mobilize enough support for combat. If the public

they seek to recruit views war to be too costly, it may withdraw or withhold its support, in turn

making war too costly for the elites. Under these conditions, opposition elites will find peace more

6

attractive than returning to (or initiating) armed conflict. Potential parties to war will avoid "the ex

post inefficiency of war" and instead "prefer to conclude an ex ante negotiated agreement" (Mattes

and Savun 2009: 739).

To the extent the government can credibly commit to protecting the masses, therefore,

opposition elites may find it too costly to re-engage in conflict, even if they have an incentive to do

so. Severing the link between elites and the masses, including by raising the cost of mobilizing

support for a conflict, can make commitments to peace more credible.

2.2 Constraining Institutions and the Commitment Problem. Political institutions could allow

actors to overcome the commitment problem in at least two ways: (1) by raising the cost of engaging

in conflict; or (2) by decreasing the potential benefits of conflict. In this section, we argue that a

type of powersharing institution we refer to as constraining institutions can activate both of these

mechanisms, allowing actors to make more credible commitments to peace, and reducing the

likelihood of conflict. We begin this section by describing these mechanisms, turn to defining the

different types of powersharing institutions, and conclude this section by explaining how

constraining institutions alleviate the commitment problem.

Protection. If and to the extent the government can credibly commit to protect the masses,

this can alleviate the civil conflict commitment problem. The protection mechanism increases the

value of the status quo to ordinary citizens, thus making them less likely to mobilize to support

revisionist elites. Another way of saying this is that protection increases the costs of fighting such

that many individuals will refuse to join the violence. This mechanism weakens civil support for

insurgency and increases insurgent leaders’ incentive to engage in political rather than military

competition. This mechanism also weakens the bonds between elites and their complicit public.

At the most basic level, political institutions that protect the general public from government

repression and guarantee political access elevate the extent to which ordinary citizens value

7

maintaining the peace. Such political systems offers a wider range of choices of political

contestation, providing non-violent forms of resolving disputes with the state, which are generally

less costly for citizens than violent conflict (Tilly 2003; Findley and Young 2011: 363). North,

Summerhill, and Weingast (2000, p. 7) point out that "establishing credible commitments requires

the creation of political institutions that alter the incentives of political officials so that it becomes in

their interest to protect relevant citizen rights." Opposition elites may nonetheless have the incentive

to go to war, yet their ability to do so will be diminished. This is a bit like the 60’s slogan – "what if

they had a war and no one came". If the cost of fighting is sufficiently large, political violence will

not be effective no matter how willing elites are, because the public will not mobilize when a

valuable and credible peaceful option exists. For the elites, this increases the costs of fighting,

particularly the cost of mobilizing supporters. By raising the cost of mobilizing public support for

war, the protection mechanism thus alleviates the problem of credibly committing to peace.

Lowering the stakes. Commitments to civil peace can also be made more credible by

lowering the stakes of political victory. Some powersharing arrangements can serve to limit the

overall power of government, thus reducing the payoff from control of the political system.

Lowering the stakes of political control allows the government to obtain minorities' consent to be

governed. “For losers the worst outcome is to have little voice in a system in which the government

has a great deal of power to implement policies unchecked. A better outcome (at least for losers)

would be to have one in which any government is more constrained” (Anderson et al. 2005: 126). In

systems with significant discretionary power for victors, losers will be tempted to seek victory

through means outside the political system rather than accept defeat within it. The stakes in such

polities are simply too large to consent to defeat.

The mechanism of lowering the stakes of political victory reduces the likelihood of conflict.

While the stakes of a dispute may have little effect on the likelihood of commitment problems in

8

many rationalist formulations, disputes are more likely to arise in the first place when stakes are

higher because parties have more to fight over. Lowering the stakes fundamentally reduces the

scope of government power, which in turn decreases the value of winning distributional conflicts.

Examples of this include peaceful distributional conflicts, such as elections, meaning that electoral

losers have diminished incentives to seek victory militarily. Yet this can also apply to violent

distributional conflicts. A losing side in a civil war may be less likely to take up arms again when

institutions are created that reduce the value of being on the winning side.

Which type of powersharing institution can activate these mechanisms? Because much of the

literature on powersharing focuses on elites, powersharing tends to be defined in terms of institutions

that facilitate the sharing of power among elites. Our definition of powersharing is broader, aiming

to also encompass institutions by which the government shares power with the masses. We define

political powersharing as an arrangement that mandates the participation of a broad set of decision

makers in the policy process.

Powersharing comes in three distinct forms, which differ fundamentally with respect to what

it means to "share". One common notion of powersharing implies an inclusiveness or jointness in

participation. This is the type of sharing in which individuals experience something together and

jointly, like sharing a house or a memorable occasion. In contrast, dispersing a good proportionally

to different individuals is also a form of sharing, as for example when family members share an

inheritance. In this context, the meaning of sharing is quite different. It takes place with the

dispersion of goods, which are to be consumed separately by their respective recipients. Yet another

form of sharing is evident when political activists demand that elites "share the wealth". In such

contexts the meaning of "sharing" relates to a call for limitations on a dominant actor or group. Thus,

sharing can refer to joint and inclusive consumption, the dispersion and individual consumption of a

good, or restrictions on some group’s control of the good in question.

9

Following these three different meanings of "sharing", the manner in which power is shared

politically can thus be divided broadly into three different forms. Constraining arrangements limit

the power of political office holders, and thereby serve to protect vulnerable groups, individuals, and

indeed civil society defined more broadly against encroachment and abuse. Inclusive arrangements

mandate that the representatives of designated parties or groups hold particular offices or participate

in particular decision-making processes. Dispersive institutions distribute authority among groups or

regions in a well-defined pattern (e.g., federalism).

While other types of powersharing arrangements focus on the sharing of executive and

legislative power, constraining arrangements are those in which those branches of government share

power with the judiciary, civil society, and ordinary citizens. Constraining powersharing

arrangements emphasize the protection of minority groups or individuals from the abuse of power

and guarantee human rights. Constraining powersharing institutions work to limit the size of the

political pie the government can enjoy. Governments often agree to constrain themselves by

creating institutions that credibly prevent them from abusing their power (North and Weingast 1989;

North 1993; Fearon 2011b). Constraining powersharing institutions act as a check or limitation on

government power, thus activating mechanisms that enforce government commitments.

Many institutional forms associated with liberal democracy might be conceived of as

constraints on the government, but in our definition of constraining powersharing institutions we

focus on those institutions that are relevant to the sharing of rights between the elites in government

and minorities in the public. A key constraining powersharing institution is the freedom of religion,

under which the state can neither sanction nor proscribe religious practices, thus allowing minority

religious groups to practice their religion. Another key constraining powersharing institution is an

effective judicial check on the authority of elected officials, whereby, for example, the high court has

the power of judicial review and judges have lifelong tenure. Bans on military officers serving in the

10

legislature also work to constrain the government and prevent abuses of minorities. Our definition

of constraining powersharing thus excludes constraints such as legislative veto powers and other

forms of horizontal accountability (O'Donnell 1998; Tsebelis 2002). Table 1 provides examples of

constraining, dispersive and inclusive powersharing institutions. Based on the IDC data, Table 1

also provides examples of polities that feature each type of powersharing institution.

Table 1: Examples of Powersharing Institutions

Examples of Institutions Examples of Polities with

Form of Powersharing

Constraining • Judicial Review • Military Legislator Ban • Freedom of Religion

• Ghana (1993-2010) • Taiwan (1975-2010) • Sierra Leone (1979-1991)

Dispersive

• Subnational Education Authority

• State Elections • Constituency Alignment

• Switzerland (1975-2010)

• Colombia (1992-2010) • Russia (1994-2010)

Inclusive • Mutual Veto • Reserved Legislative Seats • Reserved Executive Positions

• Burundi (1995-2010) • Yugoslavia (1975-1992) • Lebanon (1975-2010)

Constraining powersharing institutions are the most effective at preventing civil conflict

because they activate the protection mechanism for the masses and lower the stakes of political

victory for elites. Such institutions limit the monopolization of power, which in turn limits patterns

of exclusion and makes losers’ consent easier. The direct effects of constraining powersharing

institutions are felt by the public, securing their rights, increasing the value of the status quo, and

lowering their interest in change. These are commitments to protect the rights of minorities who do

not control the central executive and they reduce the stakes of electoral competition. They increase

the quality of life while groups are not governing, driving down the incentives to bear the costs/risks

of fighting. By limiting the scope of power of elected officials, constraining powersharing

institutions lower the stakes of political competition, thereby making it easier for the losers of

11

elections to consent to giving up power (Anderson et al. 2005). Political institutions that guarantee a

rich variety of avenues for nonviolent political expression and protect against repression from state

authorities provide credible commitments to civil peace. The link between constraining

powersharing institutions and civil peace comes from both the expectation and the fulfillment of

respect for minority and individual rights. Although elites may still be interested in conflict, they

will find it more costly to win support from publics whose rights are better protected.

A key constraining powersharing institution, the establishment of the right to religious

freedom limits the ability of elites to shape policy around their religious practices or preferences and

thus protects minorities. Many civil conflicts are fought based on identity, a big part of which is

often religion. According to Toft (2007), 42 religious civil wars were fought between 1940 and

2000. When a regime imposes laws that discriminate against rival religious groups, such groups

begin to fear for their faith (Philpott 2007). Conflicts can also result from elites socially constructing

identities around religious cleavages and mobilizing oppressed groups to produce new conflicts. As

Toft argues, "When political elites come under immediate threat, they will work to reframe issues of

contention as religious issues, essentially attempting to outbid each other in an effort to establish

religious credibility and thus attract domestic and external support" (p. 97-98). Protections for the

freedom of religion can help prevent these scenarios. Groups that are not being religiously repressed

will be less likely to mobilize behind elites that attempt to foment conflict based on their religious

identity. This will increase the cost of conflict to elites, thus making war less likely.

Another type of constraining powersharing institution that can provide a credible

commitment to protecting minority rights from government abuse is an independent judiciary. If

and to the extent the judicial branch can effectively and independently review and overturn

government actions that violate legal agreements, such an institution can make credible the

government’s promises to refrain from abusing minorities. “Namely, if an independent judiciary

12

exists extremists can be less concerned about a strong crackdown by the government in the future”

(Findley and Young 2011: 374). Leaders of states with independent courts are significantly less

likely to be able to abuse the human rights of their people (Keith 2002; Powell and Staton 2009;

Lupu 2013). Of course, not all judiciaries have such independence; the extent to which a judiciary is

truly a constraining powersharing institution depends on factors such as the processes for judicial

appointment and the power of judicial review.

The Constitutional Court of Benin, created following the adoption of the state's constitution

in 1990, provides an example of how this type of institution activates the protection mechanism and

lowers the stakes of political victory. The Constitution provides numerous human rights protections

and legally incorporates the African Charter on Human and Peoples’ Rights. Individuals have direct

access to the court when human rights abuses are alleged,3 and Article 121 of the Constitution grants

the Court the power to act on its own motion to determine the constitutionality of laws and

regulations that threaten the fundamental rights of citizens. Since its establishment, the Court has

used its powers to find violations of human rights in hundreds of cases. The government's record

with respect to human rights abuses has improved significantly over the same period.

Constraining powersharing institutions also have important effects on elites. By activating

the protection mechanism, constraining institutions lead individuals to be less likely to mobilize for

war, which makes mobilization more costly for elites. As Humphreys and Weinstein (2008, p. 436)

note, "If [armies] have motivated participation instead by mobilizing popular discontent with

government policies, post-conflict arrangements must take more seriously the establishment of

institutional arrangements that address discrimination, oppression, and inequality." Also, just as they

can protect the rights of minorities in the public, these institutions can protect the rights of minority

3 Organic Law of the Constitutional Court of Benin, Law 91-009 of Mar. 4, 1991, Art. 22.

13

elites. Finally, by limiting government power, constraining powersharing institutions make control

of governmental institutions less attractive to elites. Rather than offering a "piece of the pie", the

size of the pie is limited, thus reducing the returns from war. This argument leads to the following

hypothesis:

H: The probability of civil war is smaller in polities with more constraining

powersharing institutions.

2.3 Why Other Forms of Powersharing Do Not Resolve the Commitment Problem

Dispersive powersharing. Dispersive powersharing distributes power by decentralizing decisions.

Power is thereby dispersed across regions or sectors of society. Most often the policy process is

divided by giving control of particular territories and processes to particular groups (Brancati 2006).

Institutions commonly associated with dispersive arrangements include federal structures, wherein

local or autonomous governmental units possess budgetary and policy autonomy. In terms of our

metaphorical pie, dispersive institutional arrangements can be thought of as the distribution of

smaller pies (or tarts) to multiple groups, rather than the sharing of a larger one.

Dispersive powersharing arrangements are primarily elite oriented. These institutions grant

power and influence to regional elites both within their own region and in national politics (i.e., by

granting representation to states/provinces in national legislatures). Such arrangements activate the

mechanisms of opposition elite cooptation and elite security. Elite cooptation is offering a potential

adversary sufficient private incentives not to take up arms (O'Donnell 1979; Gandhi and Przeworski

2006; Magaloni 2008). By making participation in the existing government attractive enough, this

can lead elites to choose peace rather than war. Sometimes the incentive might be as simple as a

bribe. Over time, cooptation can improve opposition elites' loyalty to the regime (Wintrobe 1998),

particularly as coopted elites become more invested in maintaining the status quo. The case of RUF

leader Foday Sankoh being offered a position within the post-conflict government that granted him

14

control of diamonds in Sierra Leone serves as an example of such cooptation (Binningsbø and

Dupuy 2009).

Dispersive arrangements also grant personal security to elites. Former belligerents typically

perceive the security dilemma in very personal terms since the military leaders of an insurgent army

are themselves prime targets. Disarmament may permit the government to take advantage of the

situation and attack or assassinate former rebel leaders. The government has an incentive to renege

on the agreement and to preemptively attack the rebels who have disarmed. Without a guarantee that

the government is willing to honor the peace treaty, they will not agree to disarm. An example of this

mechanism can be found in Nigeria. Campbell (2013) characterizes the Nigerian post-civil war

patron-client system of powersharing as based on several relevant formal and informal rules, such as

(1) "no president for life", (2) patrons at the pinnacle are never killed by their rivals, and (3) money

accumulated by a political figure in office is sacrosanct (p. 31). In addition, there are explicit rules

about "zoning" or division of the spoils and rotation of political offices (particularly the presidency)

between different regions, a form of dispersive powersharing. The mechanism of elite security can

also be activated by the type of amnesty laws that were implemented after transitions in countries

like Argentina and Chile.

The mechanisms of opposition elite cooptation and security ineffectively address the

commitment problem. There are several key reasons for this, all of which stem from the basic

problem that changes in power over time create an incentive to use violence to secure a better

bargain. Most importantly, because these mechanisms focus on opposition elites, they do not create

credible commitments for the government to refrain from repressing the broader public. Thus, even

if some elites are protected or coopted, others outside the elite may be repressed and have an

incentive to resume conflict. Second, if a powersharing arrangement overcompensates one set of

elites to entice them to sign a peace agreement, once the weak side has laid down its arms, the more

15

powerful side may seize the opportunity to attack. The same kind of incentive to resort to violence

may also occur when a party is weak at the time it agrees to the terms of a peace treaty but

subsequently becomes more powerful. Third, elites who receive concessions may not be constrained

from using these concessions to strengthen their coalition and continuing to oppose the regime

(Magaloni 2008). Fourth, attempts to coopt competing elites can create perverse incentives by

appearing to reward violent behavior. They may therefore increase the risk of conflict by inducing

other elites to take up arms in hopes of gaining similar advantages for themselves (Tull and Mehler

2005; Roeder 2005). Finally, the cooptation mechanism can lead to repressive substitution by

leaders. Leaders can coopt some of those who oppose them in order to focus repressive activities on

others (Moore 2000). When autocrats are threatened by groups that have been coopted, they tend to

respond with concessions, but when threatened by other groups they tend to respond with repression

(Conrad 2011). Elites may be able to anticipate this behavior on the part of the autocrat, thus

continuing to conspire or rebel, even when they hold formal political office (Magaloni 2008). Thus,

mechanisms designed to induce or coopt elites to agree to peace are still vulnerable to the

commitment problem.

Others have argued that regional autonomy may reduce the threat of conflict by appealing to

a complicit public. Government concessions for greater regional autonomy may serve to undercut a

particular group’s support for armed rebellion. In addition, if dispersive institutions are created in

conjunction with credible commitments to prevent repression, dispersive powersharing may work

through civil society as well as through elite mechanisms. Yet federal structures can also sow the

seeds of conflict. While the new province may give a certain identity-group policymaking autonomy,

the new political unit may create a viable new minority. Moreover, the creation of a new province

based on ethnic identity may create a perverse incentive for an elite member of the new minority to

organize armed conflict in the hope of creating a new homeland. Such autonomy can increase the

16

risk of future violence unless it takes the form of full partition (Chapman and Roeder 2007; cf.

Downes 2004). Nigeria and India have each had a history of repeated breakups of existing states,

driven by these kinds of incentives. Although dispersive institutions can serve to grant rights to

publics within national sub-units, the central government may face difficulty in credibly committing

not to violate those rights subsequently.

Inclusive Powersharing. Inclusive powersharing arrangements grant members shares in the

exercise of political power, so that each group or party can contribute to important public decisions.

Mutual veto and grand coalitions are key forms of inclusive powersharing. According to Lijphart,

the most important institution of this type is the grand coalition (Lijphart 1985). The grand coalition

may take different forms, including: "a grand coalition cabinet in parliamentary systems, a grand

coalition of a president and other top officeholders in presidential systems, and broadly inclusive

councils or committees with important advisory and coordinating functions" (Lijphart 1985, p. 7).

Another example of inclusiveness is a mandated set of seats in the national legislature that is

reserved for members of a minority group, an institution that allows such minorities to participate in

national policymaking. An inclusive arrangement is therefore one in which the parties in a divided

society jointly decide how the pie will be apportioned.

Inclusive powersharing arrangements guarantee policymaking influence and access to power,

which activate the elite cooptation mechanism. Advocates of inclusive arrangements focus on how

this form of powersharing makes peace more attractive, binding the elite of all represented groups in

society. As we have argued above, elite cooptation does not effectively overcome the commitment

problem. Like some dispersive powersharing institutions, inclusive powersharing institutions do not

create mechanisms that allow the government to credibly commit to protecting the masses, and thus

can lead to an unstable peace. Regimes with these types of arrangements are more prone to use

repressive tactics like torture than regimes without inclusive institutions (Vreeland 2008; Conrad

17

2014; Frantz and Kendall-Taylor 2014). Because inclusive powersharing need not involve the

protection mechanism, the broader public may face repression from the government and may be

more willing to mobilize behind the elites of their faction.

Inclusive arrangements can also result in two other problems. First, inclusive powersharing

strengthens ethnic divisions by cementing them to guaranteed positions in government (Horowitz

2003). The hardening of cleavages also serves to strengthen identity and attachment to one’s own

group, thereby making it easier for elites to appeal to norms of group solidarity and kinship (Gates

2002). Second, inclusive powersharing may also increase the risk of conflict because it rewards

violent behavior (Tull and Mehler 2005). Once rebels are rewarded by being included in an inclusive

powersharing arrangement, leaders of groups not included in government will see armed conflict as

a means to power and influence. In addition, once those elites who are coopted receive their rewards,

which are often monetary, it may be difficult for the government to prevent them from using these

rewards to seek the regime's downfall (Magaloni 2008).

3. Research Design and Methodology.

3.1 Data Format. We test our hypotheses using new data, collected specifically for this purpose and

in recognition of the research design limitations of previous research. Our data precisely date major

institutional changes, such as the introduction of new constitutions or the overthrow of regimes. This

allows us to create a country-day data set. By precisely dating institutional change (as well as

conflict onset and termination), we can establish temporal precedent with a much higher degree of

accuracy than is possible with conventional country-year data.4

4 Both our dependent variable (conflict onset) and key independent variables (powersharing) are

available on a country-day basis. Other variables are coded on a country-year basis. For such

variables, we include in our analysis the annual values they take on during the date in question (e.g.,

18

3.2 Dependent Variable. Our dependent variable is the onset of an armed intrastate conflict. We

use the UCDP Conflict Termination data set (Kreutz 2010), which has information on start and end

dates for all UCDP conflicts. The threshold in the UCDP definition of internal conflict is at least 25

deaths from armed violence in a year. There is no agreed-upon definition of what it means for a

conflict to end. Therefore, in many cases it is unclear whether two periods of armed violence

constitute one or two conflicts. For cases in which violence reemerges, we have relied in our main

estimations on an often-used rule that demands a minimum of two years of inactivity before we code

renewed violence as a new conflict onset. We use the start and end dates in Kreutz (2010) to

measure the time between two distinct episodes of conflict. In the robustness tests reported below,

we address this issue with alternative specifications.

3.3 Samples. Our theoretical discussion does not distinguish between post-conflict and non-post-

conflict societies. Our arguments are intended to apply to both types of societies. We therefore

include all states in our primary analysis (Model 1 below). Because much of the existing literature

has focused only on post-conflict societies, however, we will also separately test our arguments in a

sample limited to those polities. This sub-sample (Model 2 below) contains only conflict onsets that

happen in states that previously experienced a conflict. We are thus able to contrast the post-conflict

sample with a full sample of all countries, so that we can estimate and compare the effects of

powersharing institutions under both conditions.

3.4 Statistical Model. To analyze the relationship between powersharing institutions and conflict,

we use a conditional logit model (which is described in detail in Appendix 1 because of space

considerations). The data structure used in this analysis closely follows that of Raknerud and Hegre

if the date is January 15, 1978, we use the values of the conflict onset and powersharing variables for

January 15, 1978 and the values for 1978 for all other variables).

19

(1997) and Hegre et al. (2001). We begin with a list of conflict onsets, w, and sort them in ascending

chronological order. At each point in time when an onset occurs, tw, we take a cross-sectional

snapshot of all other independent countries at that time. The dependent variable is binary, where "1"

denotes an observation of a conflict that started on that specific day.

3.5 Key Independent Variables. The IDC data set codes 19 indicators of powersharing. These

variables were chosen because they represent the key indicators of powersharing, based both on the

powersharing literature and on our theory of the three types of powersharing. The variables included

in the data set as indicators of constraining powersharing are those powersharing arrangements that

protect the rights of minority groups and ordinary citizens. Our coding of such institutions

recognizes that they may not be effective at creating credible commitments if they exist only on

paper and are not implemented in practice. These include both de jure and de facto protection of

religious freedom, the barring of the military from engagement in elected politics, the banning of

ethnic or religious parties, and the existence of an effective and independent judicial check on the

authority of elected officials. These measures are designed to avoid capturing other forms of

constraint on executive authority, such as legislative veto players and other forms of horizontal

accountability. The inclusive powersharing indicators cover two of Lijphart’s components of

consociationalism: grand coalitions and mutual veto. They also include the reservation of seats or

executive positions for specific minority groups to ensure their inclusion in central-government

decision-making. The indicators of dispersive powersharing are grouped along three dimensions: (1)

the powers allocated to regional governments; (2) the election of regional governments by regional

electorates; (3) the representation of regional constituencies in the central government. A full list of

the indicators is provided in Appendix 2.

The indicators are coded from constitutions, peace treaties, and a variety of secondary

sources. Many of the indicators are based on de jure institutions included in these documents, but we

20

also include some de facto indicators, particularly those that tell us whether core powersharing

institutions, such as may be included in the national constitution, are actually implemented. Because

of space considerations, we describe the 19 indicators and 3 indices in greater detail in Appendix 4,

which includes information on the data sources, coding procedures, summary statistics, histograms

of the indices, and values of the indices for a random sample of country-years. In Appendix 4, we

also analyze these 19 indicators via factor analysis and find that they cluster around three latent

variables, each of which corresponds to one type of powersharing. Based on the factor loadings, we

construct an index measure of each type of powersharing. We use these index measures in our

analysis.

3.6 Control Variables. It is possible that certain types of institutions are more likely to be created

under certain conditions that, in turn, affect the probability of conflict. There are three aspects to the

issue: (a) some factors might explain the creation of powersharing institutions and the onset of

conflict; (b) when conflict recurrence is analyzed, aspects of the prior conflict may have affected the

end of that conflict and the creation of post-conflict institutions (see Chapman and Roeder (2007)

and Mattes and Savun (2009) for detailed discussions of this issue); and (c) in models of conflict

recurrence, states “select” into the sample by experiencing a prior conflict, which can also create

bias. Existing work on powersharing and conflict recurrence addresses problems (a) and (b) by

controlling for factors that might explain the creation of powersharing institutions and the onset of

conflict, which we include in both Models 1 and 2, as well as by controlling for the factors that

predict civil war settlement or the end of the prior conflict, which we include in Model 2.

Nonetheless, by only analyzing post-conflict states, the inference in much existing work is

threatened by problem (c). By testing our hypothesis on both post-conflict states and a broader

sample of states (many of which did not experience a prior conflict), we go further than existing

analyses in addressing problem (c). If we find consistent results regarding the relationship between

21

powersharing and conflict in both models we can better rule out biases resulting from selection into a

prior conflict.

Economic Development and Population. We control for the natural log of GDP per capita and

the natural log of population. We use data from World Development Indicators (2011),

supplemented with data from the Penn World Tables. Economic downturns may exacerbate the risk

of conflict. We control for economic growth using the World Development Indicators,

supplemented with data from Gleditsch (2002).

Electoral Democracy. Democracies are less conflict prone than other states (Hegre et al. 2001). In

addition, one of the most consistent findings of the repression literature is that democracies are less

likely to abuse the rights of their citizens (Davenport 2007), so publics in these countries may have

fewer reasons to take up arms than those in other states. Many measures of democracy, such as

Polity, include both institutional and electoral dimensions of democracy. Including such a measure

in our analysis would be a problem because these institutional features share some common elements

with our powersharing data. Thus, we primarily use the measure of electoral democracy provided by

Alvarez et al. (1996), because it largely takes into account indicators other than our powersharing

measures. To the extent that electoral democracy correlates with any of our measures of

powersharing, it is not because we use the same indicators to measure both. In cases for which

Alvarez et al. (1996) do not provide data, we supplement by using the measure provided by Geddes,

Wright, and Frantz (2013). The pairwise correlations between our three powersharing indices and

Electoral Democracy are given in Table 2 below.

Ethno-Linguistic Fractionalization. Identity heterogeneity is likely to influence political

institutions. Ethnicity may also affect civil conflict, although this point is hotly debated (Fearon and

Laitin 2003; Cederman and Girardin 2007; Cederman et al. 2011). We control for ethno-linguistic

fractionalization using data from Roeder (2005).

22

Table 2: Cross-Correlations

Variable Inclusive Dispersive Constraining Electoral Democracy

Inclusive Powersharing 1.00 Dispersive Powersharing 0.07 1.00 Constraining Powersharing 0.04 0.36 1.00 Electoral Democracy -0.03 0.31 0.48 1.00

Regime Stability. Regime instability can be detrimental to the effectiveness of powersharing and

can foster uncertainty and subsequent conflict. Following Hegre et al. (2001), we operationalize

regime instability as temporal proximity to prior regime change, and we operationalize regime

change based on the Gates et al. (2006) definition. We use a decay function to account for this

because the expected relationship between the effect of a preceding regime change and the risk of

conflict is non-linear (Gates et al. 2006).5

Interregnum. The IDC data set codes certain polities as experiencing interregnums. These are

periods in which it is unclear who is in power in a given country, or when a country is temporarily

without any government or ruler. An example of this is Somalia beginning in 1991. Many

powersharing indicators could not be reliably coded with respect to such periods. Rather than

dropping these from our sample, which might introduce bias, we code each specific powersharing

institution as absent and include an indicator variable indicating the units for which this is the case.

3.6.1 Additional Controls for Post-Conflict Model.

5 The decay function is αt

tf−

= 2)( , where t represents the number of days since the previous regime

change and α represents the half-life parameter. The value of α tells us how long it takes before the

effect of a preceding regime change on the risk of conflict is halved. We find 12 months to be the

optimal half-life parameter for this sample. The measure is equal to 1 immediately after a regime

change has occurred and approaches 0 as the regime change becomes more temporally distant.

23

To test our hypothesis in the sub-sample of post-conflict states (Model 2), we add several control

variables that capture characteristics of the prior conflict. The intensity and duration of the prior

conflict may affect both the type of settlement reached to end it and the likelihood of new conflict

onsets. We control for these factors using the UCDP/PRIO Armed Conflict Dataset (Gleditsch et al.

2002). Third parties, such as other states and the United Nations, often intervene in civil conflicts.

These interventions can both shape the course of the conflict and have important effects on the

parties' ability to credibly commit to a settlement (Walter 1997; Fortna 2004). We control for this

factor using the UCDP Peace Agreement Dataset (Harbom et al. 2005). We control for whether a

peace agreement was signed to end the prior conflict, as this could affect both whether new

institutions were created and whether the conflict recurs. We also control for whether the agreement

included provisions for the disarmament of the warring parties, as this can exacerbate the

commitment problem. For both variables, we use the UCDP Peace Agreement Dataset (Harbom et

al. 2005). Like powersharing institutions, peacekeeping missions endeavor to alter the incentives of

belligerents to promote peace. Peacekeeping can often enhance the credibility of peace agreements

through international commitments. Thus, they may serve as either complements or substitutes to

powersharing, and their presence is determined both by international factors and factors specific to

the conflict itself. We control for whether a post-conflict peacekeeping mission was instituted using

data from Fortna (2008).

3.7 Missing Data. There are some observations in the IDC data set for which information on one or

more indicator is missing. Missingness is either due to a lack of credible information6 or a situation

6 To reduce missing data, coders adhered to a “preponderance of the evidence” rather than a “beyond

a reasonable doubt” standard when assigning values to a particular indicator. Nonetheless, for some

observations credible information could not be obtained. See Appendix 4.

24

where the indicator in question is not relevant. In our main specification, we impute the missing

indicator data using the Amelia 2 program (Honaker, King, and Blackwell 2011), before running the

factor analysis. We use the resulting powersharing indices in our statistical models. In a robustness

test reported in Appendix 3, we address the missing data problem with an alternative specification.

4. Results.

The results of our analysis using are shown in Table 3. Our main test is Model 1, which includes the

180 states in the IDC data. In this model, the coefficient of constraining institutions is significant

and negative, meaning that these institutions are associated with a lower probability of conflict onset,

as our theory suggests. By contrast, neither dispersive nor inclusive institutions have a significant

relationship with conflict onset. An increase in 1 unit in the index of constraining powersharing

(e.g., from -1 to 0 or 0 to 1) is associated with an 18% reduction in the probability of conflict onset.

Examples of states with a value of roughly 0 in the constraining powersharing index are

Czechoslovakia and Romania during most of the 1970s and 1980s (index value near 0), which had

minor protections for religious freedom, as well as some judicial powers, but where judges did not

have lifelong terms and did not have the power to review legislative and executive decisions. An

example of state with a value of roughly 1 in the constraining powersharing index is Botswana

during the entire period of the data (1975-2010), which has explicit prohibitions on state

establishment of religion and on restricting religious freedom, as well as lifelong tenure for judges

and the power of judicial review. The results of Model 2, which includes only post-conflict states,

are substantially similar with respect to powersharing institutions. This indicates that the

relationship between constraining powersharing institutions and conflict is similar for conflict onset

and conflict recurrence. This also indicates that selection into a prior conflict is not likely to explain

the relationship we find between constraining powersharing and conflict.

Economic Growth is associated with a lower probability of conflict onset, but is not

25

Table 3: Risk of Civil Conflict

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) All States Post-Conflict

States Constraining Powersharing -0.197** -0.205** (0.0783) (0.0985) Dispersive Powersharing -0.539 -0.0524 (2.735) (0.190) Inclusive Powersharing 0.101 0.0205 (0.0616) (0.0704) Economic Growth -2.213*** -0.790 (0.763) (0.883) Electoral Democracy 0.0370 -0.0535 (0.164) (0.201) Population (logged) 0.402*** 0.253*** (0.0408) (0.0578) GDP per Capita (logged) -0.407*** -0.285*** (0.0689) (0.0861) Ethno-Linguistic Fractionalization 1.015*** 0.432 (0.259) (0.333) Temporal Proximity to Prior Change in Institutions 0.368 0.229 (0.226) (0.268) Interregnum -0.736 -1.637 (0.599) (1.022) Prior Conflict Characteristics • Intensity 0.199 (0.277) • Duration 7.22e-05*** (1.66e-05) • Third Party Intervention -0.511 (0.339) • Disarmament 0.221 (0.298) • Peace Agreement 0.404 (0.372) • Peacekeeping 0.127 (0.315) Observations Ll 40,640 18,793

26

statistically significant for the recurrence of armed conflict. Population is statistically significant and

positively associated with a larger probability of conflict onset and recurrence, as we would expect.

GDP per capita is negatively associated with a larger probability of onset and recurrence. Ethno-

Linguistic Fractionalization is also positive and statistically significant, which is consistent with the

recent literature on the relationship between ethnicity and conflict (Cederman and Girardin 2007;

Cederman et al. 2011). Interestingly, in Model 2, most of the characteristics of the prior conflict are

not individually statistically significant (although they are jointly significant). Most of these

variables are significant in an alternate model that excludes powersharing institutions. This implies

that, to some extent, these prior conflict characteristics affect the choice of post-conflict

powersharing institutions, as we would expect, but that individually they may not affect the

probability of conflict recurrence once such institutions are put into place. In addition, by

controlling for these factors in Model 2, we are able to better estimate the extent to which

powersharing institutions affect conflict recurrence independently of the prior conflict characteristics

that may have led to such institutions being implemented.

4.1 Robustness Tests.

Time Since Conflict. Civil conflicts are often intermittent, recurring after pauses of days, weeks or

months. When the killing stops for a period of time, the conflict could be over or there simply may

be a lull in fighting. It can often be difficult to determine whether multiple intermittent periods of

violence are part of one conflict or constitute multiple conflicts. The ceasefire of April 2012 in

Syria, which lasted hours, clearly did not separate two conflicts. What about the 2002 ceasefire in Sri

Lanka, after which the conflict dipped below 25 battle casualties over a calendar year, but resumed

again in 2003? We could think of this as two conflicts, but most observers would not. In contrast, the

20 years between the 1946 and the 1966 Kurdish uprisings in Iraq are uncontroversially two distinct

conflicts. How much time does it take before we can call a conflict over, and hence code renewed

27

conflict between the same parties over the same incompatibility as a new onset? If every lull in

fighting is counted as the end of conflict, we would greatly exaggerate the number of conflict

terminations. To address this problem more fully, we estimated robustness tests that exclude states

with recent recurring conflicts. Rather than set a specific cut-off for what "recent" means (as we did

in our main specification), we instead re-estimate Model 1 for each value of a measure of time since

the last conflict, in days. In other words, these models iteratively increase the threshold of conflict

recentness and exclude countries from the sample that do not meet that threshold. Thus, our sample

size iteratively decreases until it includes only states that never experienced civil conflict between

1946 and 2010. Figures A3-1, A3-2, and A3-3 in the Appendix report the key results from these

tests, which strongly support our theory.

Missing Data. As noted above, there are missing values for some of the underlying indicators

in the powersharing indices. In an alternate specification reported in Appendix 3, we employ an

alternative strategy to address this issue. The results of these robustness tests support our theory.

Measures of Electoral Democracy. Electoral Democracy is one of the most important control

variables in our models, but also one that is particularly difficult to measure. In our main

specifications we primarily used the Alvarez et al. (1996) measures, supplemented by the Geddes et

al. (2013) data. To test the robustness of our results, we estimated additional models that use either

the Alvarez et al. (1996) data alone or the Geddes et al. (2013) data alone. These samples are

smaller than those of our main models because these measures are not available for all of the same

polities, but the results are consistent with our main findings. See Table A3-2 in Appendix 3.

5. Conclusions.

Powersharing, particularly as manifest in institutions that protect the rights and security of the

population at large, provides a pathway to peace. Constraining powersharing institutions, in contrast

to inclusive and dispersive types of powersharing, embody a credible commitment to peace by

28

ensuring protection from repression and making it difficult to mobilize ordinary citizens against the

state. Constraints also lower the stakes of gaining power, thus increasing the likelihood of

cooperation from elites who are left out of the ruling coalition. Mechanisms that serve to coopt rival

elites and guarantee their personal security may work to get fighting parties to sign a peace

agreement, but cannot assure credible commitment to peace over time in environments of changing

levels of relative power and military capacity. Inclusive powersharing is inherently elite-focused and

does not adequately address the commitment problem. Ensuring against the monopolization of

power, constraining powersharing institutions work by raising the costs of mobilizing a rebel force,

thereby creating the conditions for the credible commitment to peace. Mechanisms that work more

broadly through the civil society in general more effectively address the commitment problem.

Drawing on new data, we have examined the pacific properties of different types of powersharing

institutions both for countries that have previously experienced civil war and for those which have

not. We thereby address a fundamental bias affecting previous analyses of powersharing and peace.

Moreover, these data have been precisely dated, allowing us to create a country-day data set that

supplements the more conventional country-year format. By precisely dating institutional change (as

well as conflict onset and termination) we can establish temporal precedent with a much higher

degree of accuracy than is possible with conventional data.

Our analysis shows that only constraining powersharing institutions are statistically

significantly related to reducing the risk of armed civil conflict onset for all states in the world,

including reducing the risk of the recurrence of civil conflict in post-conflict environments. Inclusive

and dispersive institutions are not statistically associated with peace. Constraining powersharing

institutions that limit the stakes of government control and guarantee security for the masses play a

critical role in preventing the onset and recurrence of armed conflict. Protection against government

repression is a key pathway through which powersharing leads to peace.

29

References Cited

1. Alvarez Michael, Antonio Cheibub José, Limongi Fernando, Przeworski Adam. 1996.

“Classifying Political Regimes.” Studies in Comparative Political Development 31:1-37.

2. Anderson, Christopher J., André Blais, Shaun Bowler, Todd Donovan, and Ola Listhaug. 2005.

Loser’s Consent. Elections and Democratic Legitimacy. Oxford: Oxford University Press.

3. Andeweg, Rudy B. 2000. "Consociational Democracy." Annual Review of Political Science 3:

509–536.

4. Besley, Timothy and Torsten Persson. 2009. "Repression or Civil War?" American Economic

Review 99(2): 292-297.

5. Binningsbø, Helga Malmin. 2013. "Power Sharing, Peace and Democracy: Any Obvious

Relationships?" International Area Studies Review 16(1): 89-112.

6. Binningsbø, Helga Malmin, and Kendra Dupuy. 2009. "Using power-sharing to win a war: The

implementation of the Lomé Agreement in Sierra Leone." Africa Spectrum: 87-107.

7. Brancati, Dawn. 2006. "Decentralization: Fueling the Fire or Dampening the Flames of Ethnic

Conflict and Secessionism." International Organization 60(3):651-685.

8. Buhaug, Halvard and Jan Ketil Rød. 2006. "Local determinants of African civil wars,

1970-2001." Political Geography 25: 315-335.

9. Campbell, John. 2013. Nigeria: Dancing on the Brink. Rowman & Littlefield Publishers.

10. Cederman, Lars-Erik, and Luc Girardin. 2007. "Beyond fractionalization: Mapping ethnicity

onto nationalist insurgencies." American Political Science Review 101 (01): 173-185.

11. Cederman, Lars-Erik, Nils B. Weidmann, and Kristian Skrede Gleditsch. 2011. "Horizontal

Inequalities and Ethnonationalist Civil War: A Global Comparison." American Political Science

Review 105(3): 478-495.

12. Chapman, Thomas and Philip G. Roeder. 2007. "Partition as a Solution to Wars of Nationalism:

The Importance of Institutions." American Political Science Review 101(4): 677-691.

13. Collier, Paul. 2000. "Rebellion as Quasi-Criminal Activity." Journal of Conflict Resolution

44(6): 839—853.

14. Conrad, Courtenay. 2011. "Constrained Concessions: Beneficent Dictatorial Responses to the

Domestic Political Opposition." International Studies Quarterly 55(4): 1167-87.

15. Conrad, Courtenay. 2014. "Divergent incentives for dictators: Domestic institutions and

30

(international promises not to) torture." Journal of Conflict Resolution 58(1): 34–67.

16. Davenport, Christian. 1995. "Multi-Dimensional Threat Perception and State Repression: An

Inquiry into Why States Apply Negative Sanctions." American Journal of Political Science 39:683-

713.

17. Davenport, Christian. 2007. "State Repression and Political Order." Annual Review of Political

Science 10:1-23.

18. Downes, Alexander B. 2004. "The Problem with Negotiated Settlements to Ethnic Civil Wars."

Security Studies 13(4): 230-79.

19. Fearon, James D. 1995. "Rationalist Explanations for War." International Organization 49(3):

379-414.

20. Fearon, James D. 1998. "Commitment Problems and the Spread of Ethnic Conflict." In The

International Spread of Ethnic Conflict, ed. David A. Lake and Donald Rothchild. Princeton, NJ:

Princeton University Press.

21. Fearon, James D. 2004. "Why Do Some Civil Wars Last So Much Longer than Others?" Journal

of Peace Research 41(3): 275-301.

22. Fearon, James D. 2011a. "Governance and Civil War Onset." World Bank Development Report.

23. Fearon, James D. 2011b. "Self-Enforcing Democracy." The Quarterly Journal of

Economics 126(4): 1661-1708.

24. Fearon, James D., and David D. Laitin. 2003. "Ethnicity, Insurgency, and Civil War." American

Political Science Review 97(1): 75-90.

25. Findley, Michael G. and Joseph K. Young. 2011. "Terrorism, Democracy, and Credible

Commitments." International Studies Quarterly 55(2): 357-378

26. Fortna, Virginia Page. 2003. "Scraps of Paper? Agreements and the Durability of Peace."

International Organization 57(1): 337-372.

27. Fortna, Virginia Page. 2004. "Does Peacekeeping Keep Peace? International Intervention and the

Duration of Peace after Civil War." International Studies Quarterly 48(2): 269-292.

28. Fortna, Virginia Page. 2008. Does Peacekeeping Work? Shaping Belligerents' Choices after

Civil War. Princeton, NJ: Princeton University Press.

29. Frantz, Erica, and Andrea Kendall-Taylor. 2014. "A Dictator’s Toolkit: Understanding how Co-

optation Affects Repression in Autocracies." Journal of Peace Research, forthcoming.

31

30. Gandhi, Jennifer and Adam Przeworski. 2006. "Cooperation, Cooptation, and Rebellion under

Dictatorships," Economics and Politics 18(1): 1-26.

31. Gates, Scott. 2002. "Recruitment and Allegiance: The Microfoundations of Rebellion" Journal of

Conflict Resolution 46(1): 111-130.

32. Gates, Scott, Håvard Hegre, Mark Jones, and Håvard Strand. 2006. "Institutional Inconsistency

and Political Instability: Polity Duration, 1800–2000," American Journal of Political Science 50(4):

261-282.

33. Geddes, Barbara, Joseph Wright, and Erica Frantz. 2013. “New Data on Autocratic Breakdown

and Regime Transitions.” Working Paper.

34. Gleditsch, Kristian Skrede. 2002. "Expanded Trade and GDP Data." Journal of Conflict

Resolution 46(5): 712-724.

35. 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(5): 615-

637.

36. Harbom, Lotta, Stina Högbladh, and Peter Wallensteen. 2006. "Armed Conflict and Peace

Agreements." Journal of Peace Research 43(5).

37. Hartzell, Caroline A. and Matthew Hoddie. 2003. "Institutionalizing Peace: Power Sharing and

Post-Civil War Conflict Management." American Journal of Political Science 47(2): 318–332.

38. Hartzell, Caroline A. and Matthew Hoddie. 2007. Crafting Peace. Powersharing Institutions and

the Negotiated Settlement of Civil Wars. University Park, PA: Pennsylvania State University Press.

39. Hegre, Håvard, Tanja Ellingsen, Scott Gates, and Nils Petter Gleditsch. 2001. "Toward a

Democratic Civil Peace? Democracy, Political Change, and Civil War, 1816–1992." American

Political Science Review 95(1): 33–48.

40. Honaker, James, Gary King, and Matthew Blackwell. 2011. "Amelia II: A Program for Missing

Data (Special Vol.)." Journal of Statistical Software 45(7):1– 47.

41. Horowitz, Donald L. 2003. "The Cracked Foundations of the Right to Secede." Journal of

Democracy 14(2): 5–17.

42. Humphreys, Macartan, and Jeremy M. Weinstein. 2008. "Who Fights? The Determinants of

Participation in Civil War." American Journal of Political Science 52(2): 436-455.

43. Jarstad, Anna. 2001. Changing the game: Consociational Theory and Ethnic Quotas in Cyprus

32

and New Zealand (Doctoral dissertation, Uppsala University).

44. Jarstad, Anna and Desiree Nilsson. 2008. "From Words to Deeds: The Implementation of

Powersharing Pacts in Peace Accords." Conflict Management and Peace Science 25(3): 206-223.

45. Kalyvas, Stathis, and Matthew Kocher. 2007. “How ‘Free’ is Free Riding in Civil Wars?

Violence, Insurgency and the Collective Action Problem.” World Politics 59(1): 177–216

46. Keith, Linda Camp. 2002. "Judicial Independence and Human Rights Protection Around the

World." Judicature 85 (4):195–200.

47. Kreutz, Joakim. 2010. "How and When Armed Conflicts Ends: Introducing the UCDP Conflict

Termination Dataset." Journal of Peace Research 47(2): 243-250.

48. Lake, David A. 2003. "International Relations Theory and Internal Conflict: Insights from the

Interstices." International Studies Review 5(4): 81-89.

49. Lichbach, Mark Irving. 1987. "Deterrence or Escalation? The Puzzle of Aggregate Studies of

Repression and Dissent." Journal of Conflict Resolution 31:266-297.

50. Lijphart, Arend, 1977. Democracy in Plural Societies: A Comparative Exploration. New Haven:

Yale University Press.

51. Lijphart, Arend. 1985. Power-Sharing in South Africa. Berkeley: Institute of International

Studies, University of California.

52. Lupu, Yonatan. 2013. "Best Evidence: The Role of Information in Domestic Judicial

Enforcement of International Human Rights Agreements." International Organization 67(3): 469-

503.

53. Magaloni, Beatriz. 2008. "Credible Powersharing and the Longevity of Authoritarian Rule."

Comparative Political Studies 41(4-5): 715-741.

54. Mattes, Michaela and Burcu Savun. 2009. "Fostering Peace after Civil War: Commitment

Problems and Agreement Design." International Studies Quarterly 53(3): 737–759.

55. Moore, Will H. 1998. "Repression and Dissent: Substitution, Context and Timing." American

Journal of Political Science 45(3):851–73.

56. Moore, Will H. 2000. "The Repression of Dissent: A Substitution Model of Government

Coercion." Journal of Conflict Resolution 44(1):107–27.

57. Mukherjee, Bumba. 2006. "Why Political Powersharing Agreements Lead to Enduring Peaceful

Resolution of Some Civil Wars, But Not Others?" International Studies Quarterly 50(2): 479–504.

33

58. Murshed, S. Mansoob. 2009. Explaining Civil War. A Rational Choice Approach. Cheltenham,

UK: Edward Elger.

59. Murshed, S. Mansoob. 2011. "The Clash of Civilizations and the Interaction between Fear and

Hatred." International Area Studies Review 14(1): 31-48.

60. North, Douglass C. 1993. "Institutions and Credible Commitment." Journal of Institutional and

Theoretical Economics (JITE)/Zeitschrift für die gesamte Staatswissenschaft: 11-23.

61. North, Douglass C., and Barry R. Weingast. 1989. "Constitutions and Commitment: the

Evolution of Institutions Governing Public Choice in Seventeenth-Century England."The Journal of

Economic History 49(4): 803-832.

62. North, Douglass C., William Summerhill and Barry R. Weingast. 2000. "Order, Disorder and

Economic Change." In Governing for Prosperity, Bruce Bueno de Mesquita and Hilton Root, eds.

New Haven: Yale University Press.

63. O'Donnell, Guillermo. 1979. Modernization and Bureaucratic-Authoritarianism: Studies in

South American Politics. Berkeley, CA: Institute of International Studies.

64. O'Donnell, Guillermo. 1998. "Horizontal Accountability in New Democracies." Journal of

Democracy 9(3): 112-126.

65. Olson, Mancur. 1965. The Logic of Collective Action: Public Goods and the Theory of Groups.

Cambridge: Harvard University Press.

66. Philpott, Daniel. 2007. "Explaining the political ambivalence of religion." American Political

Science Review, 101(3), 505-525.

67. Popkin, Samuel. 1979. The Rational Peasant: The Political Economy of Rural Society in

Vietnam. Berkeley, CA: University of California Press.

68. Powell, Robert. 2004. "Bargaining and Learning While Fighting." American Political Science

Review 48(2): 344-361.

69. Powell, Robert. 2006. "War as a Commitment Problem." International Organization 60 (1):

169–203.

70. Powell, Emilia Justyna, and Jeffrey K. Staton. 2009. "Domestic Judicial Institutions and Human

Rights Treaty Violation." International Studies Quarterly 53 (1):149–74.

71. Raknerud, Arvid and Håvard Hegre. 1997. "The Hazard of War: Reassessing the Evidence for

the Democratic Peace." Journal of Peace Research 34(4): 385-404.

34

72. Roeder, Philip G. 2005. “Power dividing as an alternative to ethnic power sharing,” Sustainable

Peace: Power and Democracy after Civil Wars, Roeder, P. G., and Rothchild, D. S. (Eds.). Ithaca:

Cornell University Press, pp. 51-82.

73. Sisk, Timothy D. 1996. Power Sharing and International Mediation in Ethnic Conflicts. New

York: Carnegie Corporation.

74. Skocpol, Theda. 1979. States and Social Revolutions. Cambridge: Cambridge University Press.

75. Tilly, Charles. 2003. The Politics of Collective Violence. Cambridge: Cambridge University

Press.

76. Toft, Monica Duffy. 2007. "Getting religion? The puzzling case of Islam and civil

war."International Security 31(4): 97-131.

77. Tsebelis, George. 1990. Nested Games: Rational Choice in Comparative Politics. Berkeley and

Los Angeles: University of California Press.

78. Tsebelis, George. 2002. Veto Players: How Political Institutions Work. Princeton University

Press.

79. Tull, Dennis M. and Andreas Mehler. 2005. "The Hidden Costs of Powersharing: Reproducing

Insurgent Violence in Africa." African Affairs 104(416): 375–398.

80. Vreeland, James. 2008. "The Effect of Political Regime on Civil War: Unpacking Anocracy."

Journal of Conflict Resolution 52(3): 401-425.

81. Walter, Barbara F. 2002. Committing to Peace: The Successful Settlement of Civil Wars.

Princeton: Princeton University Press.

82. Walter, Barbara F. 2011. "Conflict Relapse and the Sustainability of Post-Conflict Peace."

World Bank Development Report.

83. Weingast, Barry R. 1995. "The Economic Role of Political Institutions: Market-Preserving

Federalism and Economic Development." Journal of Law, Economics & Organization 11(1): 1-31.

84. Wintrobe, Ronald. 1998. The Political Economy of Dictatorship. Cambridge, UK: Cambridge

University Press.

85. Wucherpfenning, Julian. 2013. “The Strategic Logic of Power-Sharing after Civil War.”

manuscript, ETH Zurich.

35

Appendix 1: Statistical Model

Analyses of binary variables, such as conflict onset, often apply a logit or probit regression model.

One limitation of such models is that they assume the probability of civil war, when all explanatory

variables are accounted for, to be constant over time (Raknerud and Hegre 1997). This can perhaps

be remedied by the estimation of dummy variables designed to capture the fixed effects of time, but

because these temporal dummy variables are of no theoretical interest, it is more efficient not to

estimate them. Raknerud and Hegre (1997) show that a semi-parametric Cox model is unaffected by

temporal variations in global conflict propensity and is therefore better suited to our purposes. This

model, when estimated on fixed duration observations, is equivalent to conditional logistic

regression, which is the model we estimate.

In our analysis, the time of observation is a day in which one country experiences a civil war

onset, and where all other observations are censored. Being censored in this setting means we know

that these regimes did not experience armed conflict onset on that particular day, but that we do not

know whether they will have an onset in the future. Given that we know there is an onset in one

country at a point in time t, the probability of that war occurring in country A is given by:

Pr(onset in country A | onset happens at t)=

∑ ∑

∑

∈ =

=

tRi

p

jjij

p

jjAj

tX

tX

1

1

)(exp

)(exp

β

β (2)

where R t is the set of all countries at risk of experiencing an onset at time t; p is the number of

explanatory variables; XjA is an explanatory variable j observed for each country A; and βj is the

corresponding coefficient

1

Appendix 2: Powersharing Indicators

The definition of each powersharing indicator is given in Table A2. Additional details are provided

in Appendix 4.

2

Table A2: Powersharing Indicators Inclusive Powersharing Mandated Grand Coalition or Unity Government

Binary. 1 if there is a constitutional or treaty provision requiring representation by all major parties in the cabinet or if they are all represented in a government of national unity.

Mutual Veto Binary. 1 if there is a minority veto over a particular area of policy.

Reserved Seats Share of seats in the lower house that are reserved for ethnic/religious minorities.

Inclusive Military Binary. 1 if it is required that all major groups or all regions be represented in the military or its officer corps.

Reserved Executive Positions

Binary. 1 if particular executive positions are reserved for specific groups.

Dispersive Powersharing Subnational Education Authority

Binary. 1 if states/regions share or have control over education policy.

Subnational Tax Authority

Binary. 1 if states/regions can levy their own taxes.

Subnational Police Authority

Binary. 1 if subnational governments have control of local police/paramilitary forces in their area.

Constituency Alignment

Binary. 1 if the states/provinces are the constituencies of a majority of legislators in the upper (or only) house.

State Elections_1 Binary. 1 if state/provincial legislatures are elected. State Elections_2 Binary. 1 if state/provincial executives are elected.

Constraining Powersharing Religion Protected (Freedom from Discrimination)

Binary. 1 if constitution/peace treaty guarantees freedom from religious discrimination.

Religion Protected (Freedom to Practice)

Binary. 1 if constitution/peace treaty guarantees freedom of religious practice.

Military Legislator Ban

Binary. 1 if there is a ban on military officers serving in the legislature.

Ethnic Party Ban Binary. 1 if there is a ban on religious or ethnic parties. Judicial Constitution

Binary. 1 if the role of the judiciary is specified in the constitution.

Judicial Review Binary. 1 if the judicial branch has the power to declare the actions of the legislature AND executive unconstitutional.

Judicial Tenure_1 Binary. 1 if tenure of supreme court justices is greater than 6 years.

Judicial Tenure_2 Binary. 1 if the tenure of supreme court justices is lifelong or until a mandatory retirement age.

3

Appendix 3: Robustness Tests

This Appendix sets forth the results of the robustness tests discussed in the text.

Time Since Conflict. Figures A3-1, A3-2, and A3-3 report the key results from robustness tests in

which we expand the amount of elapsed time required before we treat a conflict as recurring. For

each of the iterative regression models, the figures report the coefficients of the three powersharing

indicators with 95% confidence intervals. The results strongly support our theory. Regardless of the

threshold we choose for conflict recentness, constraining powersharing institutions have a significant

and negative relationship with conflict onset. Interestingly, the magnitude of the coefficient

increases slightly as we increase the threshold, indicating that the effects of constraining

powersharing institutions may be more important in societies that have not experienced very recent

conflicts. By contrast, the coefficients of dispersive and inclusive institutions remain non-significant

regardless of the threshold. However, it is worth noting that the coefficient of dispersive institutions

changes from being consistently negative in the larger samples to consistently positive in the smaller

samples. It may be the case that in states with long-past conflicts, dispersive institutions eventually

foster new conflicts. Although our findings on this point are tentative, they do lend support to

arguments made by Chapman and Roeder (2007) and others that, over time, grants of regional and

other sub-national autonomy create instability and lead to future conflicts. Finally, we note that the

coefficient of inclusive institutions oscillates between positive and negative, hovering around zero.

The results suggest, therefore, that these institutions have little effect on the probability of conflict

onset, regardless of the recentness of past conflicts.

4

Figure A3-1 - Constraining Powersharing Institutions

5

Figure A3-2 - Dispersive Powersharing Institutions

6

Figure A3-3 - Inclusive Powersharing Institutions

7

Missing Data. When coders looked for evidence of the existence of a political institution, but coded

the data as missing, it may be more likely than not that the given institution did not exist. As a

robustness test, we therefore re-code all missing indicator data as zero, re-run the factor analysis

used to measure the powersharing indicators, and re-run our statistical models analysis using these

data. The results of these models, reported in Table A3-1, are substantially similar to those reported

above.

8

Table A3-1: Risk of Civil Conflict -- Missing Powersharing Indicators Re-Coded as Zeroes

(1) (2) All States Post-Conflict

States Constraining Powersharing -0.202** -0.238** (0.0821) (0.108) Dispersive Powersharing 0.0238 0.174 (0.0881) (0.107) Inclusive Powersharing 0.129** 0.0242 (0.0579) (0.0674) Economic Growth -2.239*** -0.730 (0.759) (0.875) Electoral Democracy 0.0221 -0.124 (0.164) (0.199) Population (logged) 0.396*** 0.202*** (0.0448) (0.0668) GDP per Capita (logged) -0.417*** -0.292*** (0.0704) (0.0861) Ethno-Linguistic Fractionalization 1.014*** 0.501 (0.258) (0.336) Temporal Proximity to Prior Change in Institutions 0.378* 0.223 (0.227) (0.270) Interregnum -0.711 -1.587 (0.597) (1.020) Prior Conflict Characteristics

• Intensity (0.279) 7.25e-05*** • Duration (1.68e-05) -0.475 • Third Party Intervention (0.342) 0.234 • Disarmament (0.300) 0.341 • Peace Agreement (0.376) 0.141 • Peacekeeping (0.315) Observations 40,640 18,793 Ll -1139 -758.2 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

9

Table A3-2: Risk of Civil Conflict -- Alternative Electoral Democracy Measures

Alvarez et al. (1996) data Geddes et al. (2013) data

All States Post-Conflict States

All States Post-Conflict States

Constraining Powersharing -0.190** -0.202** -0.199** -0.212** (0.0778) (0.100) (0.0800) (0.0998) Dispersive Powersharing -0.113 -0.0489 -2.141 -0.0540 (0.961) (0.180) (2.835) (0.207) Inclusive Powersharing 0.108* 0.0359 0.120* 0.0271 (0.0616) (0.0709) (0.0692) (0.0744) Economic Growth -2.334*** -0.882 -2.290*** -1.108 (0.757) (0.886) (0.796) (0.916) Electoral Democracy 0.00750 -0.0553 0.131 0.0491 (0.170) (0.210) (0.174) (0.214) Population (logged) 0.400*** 0.258*** 0.399*** 0.246*** (0.0397) (0.0585) (0.0449) (0.0599) GDP per Capita (logged) -0.407*** -0.291*** -0.419*** -0.304*** (0.0698) (0.0883) (0.0714) (0.0870) Ethno-Linguistic Fractionalization 0.983*** 0.382 1.027*** 0.455 (0.261) (0.338) (0.266) (0.343) Temporal Proximity to Prior 0.347 0.197 0.253 0.145 Change in Regime (0.228) (0.271) (0.237) (0.278) Interregnum -0.716 -1.642 -0.666 -1.576 (0.598) (1.022) (0.600) (1.024) Prior Conflict Characteristics 0.290 • Intensity (0.282) (0.289) 6.97e-05*** 7.13e-05*** • Duration (1.70e-05) (1.69e-05) -0.585* -0.622 • Third Party Intervention (0.349) (0.380) 0.103 0.0853 • Disarmament (0.306) (0.329) 0.514 0.516 • Peace Agreement (0.377) (0.429) 0.128 0.155 • Peacekeeping (0.316) (0.325) Observations 39,499 18,125 35,424 17,747 ll -1108 -731.2 -1063 -724.8 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

10

Appendix 4: The Inclusion, Dispersion, and Constraint Data Set , 1975-2010

This Appendix provides additional information regarding the coding and sources for the

powersharing indicators used in this paper. This Appendix also describes the factor analytic

procedure used to create the indices of inclusive, dispersive, and constraining powersharing.

Section 1.The Data Set

The Inclusion, Dispersion, and Constraints (IDC) data set consists of 19 indicators of powersharing

and 24 related variables across 180 countries7 from 1975-2010. By disaggregating powersharing

into its constituent institutions, we can test nuanced theories about the ways in which component

institutions affect outcomes. Our approach also allows us to use factor analysis to test whether these

component institutions actually cluster in the way predicted by our theoretical typology.

Section 1.1 De Jure vs. De Facto Institutions

Especially in the conflict-torn societies in which powersharing may be most pressing, formal

institutions do not always describe the way in which political decisions are made on the ground. And

yet formal institutions, even when they are not fully functional, can be critical political points of

expectation and ambition. One of the challenges of institutional analysis in such contexts is therefore

how to capture both the formal and the informal rules, especially when these diverge. Most of the

variables coded in this project measure the nature of de jure institutions, with a particular focus on

constitutional and treaty-based provisions. We take this approach for two reasons. First, information

about de jure institutions can be more consistently sourced and more objectively coded than de facto

arrangements. Second, we are concerned theoretically with the effect of the formal rules of the game.

We believe that formal rules continue to matter even when they are being violated. Even when a

country is de facto ruled by various militia groups –Lebanon in the mid-to-late 1980s for example –

7 We code all independent countries with populations over 250,000.

11

the constitution and other written rules still matter because they establish, at the very least, a focal

point in the bargaining over what the rules would be if everyone put down their guns, and this in turn

informs groups’ decisions to put down the guns or not.

Our focus on de jure indictors contrasts with the Ethnic Power Relations (EPR) data set,

which defines powersharing as “any arrangement that divides executive power among leaders who

claim to represent particular ethnic groups” (Wimmer et al. 2009). Thus, EPR is a de facto analog to

our de jure measure of inclusive powersharing. There exist cases, such as Switzerland, where there is

de facto powersharing but no formal inclusive institutions. Contrasting Switzerland with Lebanon

illustrates what we expect is a general pattern across these two types of data: de facto powersharing

can exist in the absence of formal rules primarily in “easy” cases where the risk of violent conflict is

not high, while the cases in which de jure powersharing fails to be enforced are often those

characterized by violence.

However, implementation matters and, therefore, information on the (non)implementation of

mandated powersharing institutions is critical in determining whether the failure of an institution to

promote a desired outcome is attributable to its ineffectiveness or to the fact that it was not

implemented. Therefore, we also include some de facto indicators, particularly those that tell us

whether core inclusive powersharing institutions, such as may be included in the national

constitution, are in fact implemented.

Section 1.2. Measuring Powersharing

We structure our discussion of individual indicators on the basis of the types of powersharing with

which they are associated, and focus heavily on the 19 indicators of powersharing that are at the

theoretical core of the project. However, the 24 additional variables coded in the project capture

important institutions of the institutional characteristic that may condition the effect of powersharing

or constitute subjects of study in their own right. For an exhaustive account of all variables in the

12

IDC data set we direct readers to the official codebook, which is available online. [URL

REDACTED FOR REVIEW.]

Section 1.2.1. Inclusive Powersharing

The inclusive powersharing variables in this study cover two of Lijphart’s (1968, 1977) components

of consociationalism: grand coalitions and mutual veto. They also include the reservation of seats or

executive positions for specific minority groups to ensure their inclusion in central-government

decision-making.

We distinguish empirically between two types of grand coalitions: those mandated by

constitutions or peace treaties (mandated grand coalition); and de facto grand coalitions in non-

elected governments, i.e. governments of national unity, which are usually installed by peace treaties

(unity).8

A mutual or minority veto provision (mutual veto) is coded as present if there is a

constitutional or treaty provision providing for a minority veto over legislation in a particular policy

area, such as language or cultural policy. Reserved executive positions captures whether any

executive positions are reserved for members of specific groups, such as the arrangement in

Lebanon, where the president is required to be a Maronite, and the prime minister a Sunni.9

Reserved seats is a continuous measure of the proportion of legislative seats (lower house) reserved

for minority groups.

8 Because mandated grand coalitions and unity governments are substitutes, they are combined into a

single measure in the factor analysis. We also collect data on de facto grand coalitions by seat share,

but these measures are not included in our index of inclusive powersharing.

9 The speaker of the parliament is required to be Shi’a.

13

For our de jure indicators of mandated grand coalitions, reserved legislative seats, and reserved

executive positions we also make a de facto assessment of whether these institutions are

implemented. We code implementation as binary variables (gcimp, resimp, resseatsimp), and then

create an additional binary variable, violation, which takes a value of 1 any time one or more of

these institutions is mandated but is not enforced.

Section 1.2.2. Dispersive Powersharing

Dispersive powersharing refers to institutions that distribute authority among a variety of decision

makers in a territorial pattern.10 The variables in our data that capture dispersive powersharing are

grouped along three dimensions: (1) the powers allocated to sub-national governments; (2) the

accountability of subnational governments to citizens; (3) the representation of sub-national

constituencies in the central government.11

To assess the powers allocated to sub-national governments, we measure whether

state/provincial governments have the authority to levy their own taxes (Subnational Tax Authority),

whether they have shared or sole control over education policy, (Subnational Educational

Authority), which serves as a proxy for non-fiscal domestic policy more broadly, and whether

state/provincial or municipal governments control any police or paramilitary forces (Subnational

Police Authority), which captures decentralization of the legitimate use of force. To measure the

accountability of sub-national governments to citizens, we record whether executives and

legislatures are directly elected at both the state/provincial level and the municipal level. To measure

10 Lijphart’s notion of segmental autonomy would fit here.

11 The theoretical logic of these measures is discussed in greater detail in [REDACTED]. In

developing this coding scheme we are indebted to prior work by Brancati (2006) and Beck et al.

(2001).

14

the representation of sub-national constituencies in the central government, we code a dummy

variable for whether states/provinces form the constituencies of more than half the members in the

upper house of the legislature.

Our coding of sub-national elections (State Elections) and state/provincial representation in

the upper house of the legislature (Constituency Alignment) are based directly on variables first

coded by Beck et al. (2001) in the Database of Political Institutions (DPI).12 We both extend the

coverage of these variables forward in time, and fill in a large number of previously missing values.

Our codings are included in the 2010 release of DPI.

Section 1.2.3. Constraining Powersharing

Constraining powersharing takes several forms. Those that we capture here are the constitutional

protection of religious freedom; the barring of the military from engagement in elected politics; the

banning of ethnic or religious parties; and the existence of an effective judicial check on the

authority of elected officials.

We code whether the constitution (or peace treaty) assures the freedom of religious practice

and/or the freedom from discrimination on the basis of religious affiliation. The variables Religion

Protected (Discrimination) and Religion Protected (Practice) take a value of 1 if freedom from

religious discrimination and freedom of religious practice (respectively) are guaranteed by the

constitution. We also code whether members of the military are banned from serving in the

legislature (Military Legislator Ban), and whether there is a ban on ethnic, religious, and/or regional

parties (Ethnic Party Ban).

The presence of an effective judicial check on the authority of elected officials is assessed by

whether the supreme court has the power to void actions of the legislature and executive that violate

12 The DPI variable names are “state” and “stconst” respectively.

15

the constitution (Judicial Review), the tenure of justices13 (Judicial Tenure), and whether the powers

of the judiciary are enumerated in the constitution (Judicial Constitution).14

Section 1.3 Data Structure

Time series cross sectional data sets on governance, such as Polity IV and DPI, almost always

employ a country-year unit of analysis. This allows for easy integration with other governance data

sets and with economic and political indicators, such as the World Development Indicators, which

also use this format. The IDC data set follows suit with variables coded as of January 1st of a given

country-year.15

However, the core powersharing variables in our study are coded more precisely: we code

transitions from one value to another of these variables down to the day. Precise temporal coding is

important for any causal analysis.

Section 1.4 Summary Statistics

Pairwise correlations between the component indicators of powersharing are given in Table 2. Table

3 provides descriptive statistics on all indicators.

13 Six years or less; more than six years but less than lifelong; or lifelong or until a mandated

retirement age.

14 On this point, we draw on La Porta et al. (2004).

15 In addition to the country-year format, we have also developed a version of the data set in which

the unit of analysis is the period of institutional consistency, or polity. Rather than coding all breaks

between one unit of analysis and the next as occurring on January 1 of each year, a break in the

polity data set occurs at any time the core political institutions governing the country change.

Therefore, polities vary in length, with some lasting only a few weeks, and some stretching across

our entire timespan.

16

Table 2: Pairwise Correlations

17

Section 2. Factor Analysis

Two strengths of our data are that institutions are coded in a highly disaggregated fashion, and that

the data set includes a very large number of country-years (over 5,000). The nineteen indicators of

powersharing we code each represent a specific means through which states allocate power to, and

share power between, competing social groups. Our disaggregated approach allows direct analysis of

which institutions generally serve as complements or substitutes for one another, and allows for the

study of complex interactions that may exist between institutional arrangements.

Beginning with Lijphart (1968, 1977), studies of powersharing have assumed that there exists a set

of observable variables, i.e., specific policies and institutions, associated with one or more

unobservable or latent variables (i.e., factors) that we can identify as powersharing. The core idea is

18

that, when constitutions or treaties are drafted with the aim of sharing power in a particular manner,

they include a variety of specific features that supplement one another. For example, if the aim is to

ensure the inclusion of minority groups in central government decision-making, there are multiple

institutional arrangements that might be employed to do that. States with the highest levels of

inclusive powersharing will include several of these institutional arrangements together. For

example, since 1998 Fiji has featured reserved legislative seats, reserved executive positions, and a

mandated grand coalition.

We have already introduced the indicators according to the type of powersharing with which each is

theoretically associated. Our argument predicts that these 19 indicators cluster empirically around

three latent variables that match the types of powersharing with which they are theoretically

associated – e.g. all the indicators of constraining powersharing are predicted to load on the same

factor (latent variable).

This expectation is based on the assumption that indicators of the same type of powersharing

complement or supplement one another, rather than acting as direct substitutes. If two institutional

arrangements operate as substitutes, they will not load on the same latent variable. Therefore, before

entering indicators into the factor analysis, we combine indicators that are believed to be direct

substitutes. We also separate categorical variables into individual dummy variables. Thus, two

indicators of grand coalition government, Mandated Grand Coalition and Unity Government, are

combined into a new variable Mandated Grand Coalition or Unity Government.16 Conversely, State

16 In this context, we omit measures of grand coalition government that are based solely on

legislative seat shares because these de facto grand coalitions are qualitatively distinct phenomena

that are often associated with particular types of party systems, but not with the kind of binding,

inclusive agreements that interest us here.

19

Elections and Judicial Tenure are each divided into two dummy variables – one dummy variable for

each level of the original ordinal variables.

The scree plot presented in Figure 1 is based on factor analysis on 19 component indicators of

powersharing. If one tracks the factors from right to left, a clear discontinuity divides the first three

factors from those that follow. Only the first three factors have eigenvalues greater than one.

The manner in which the factor analysis is conducted does not specify which indicator is expected to

be associated with which factor. It is the correlations between the observed variables that determine

which observed variables are associated with each latent variable. Following the initial factor

analysis, we implement a varimax rotation of the first three factors, which simplifies the factor

structure and eases interpretation.

Table 4 presents the factor loadings for the first four factors in our analysis: the first three factors are

each labeled according to the type of powersharing they represent. We include the fourth factor in

the table for comparison, though it is not of substantive interest. As predicted by our theory, and as

20

shown in the scree plot (Figure 1), the first three factors explain most of the correlation between the

individual measures of powersharing we assess.

Table 4: Factor Loadings

Variable Constraining Powersharing

Dispersive Powersharing

Inclusive Powersharing

Factor 4 Uniqueness

Subnational Education Authority 0.20 0.60 0.03 0.01 0.54 Subnational Tax Authority 0.23 0.61 0.10 0.18 0.51 Subnational Police Authority 0.07 0.53 0.00 0.06 0.66 Constituency Alignment 0.14 0.58 0.11 0.08 0.61 State Elections_1 0.13 0.62 0.00 0.06 0.53 State Elections_2 0.12 0.62 -0.08 -0.02 0.52 Religion Protected (Discrimination) 0.55 0.07 -0.13 0.27 0.52 Religion Protected (Practice) 0.59 0.04 -0.05 0.29 0.50 Military Legislator Ban 0.37 -0.09 0.11 0.02 0.79 Ethnic Party Ban 0.10 -0.09 0.01 0.31 0.82 Judicial Constitution 0.73 0.11 0.06 0.01 0.41 Judicial Review 0.50 0.09 0.17 0.29 0.56 Judicial Tenure_1 0.75 0.24 0.00 -0.38 0.24 Judicial Tenure_2 0.54 0.27 0.00 -0.54 0.34 Mandated Grand Coalition or Unity Government 0.02 -0.04 0.12 0.02 0.92 Mutual Veto 0.03 0.23 0.48 -0.05 0.65 Reserved Seats 0.01 -0.01 0.69 -0.04 0.45 Inclusive Military 0.03 0.05 0.52 0.03 0.67 Reserved Executive Positions 0.00 -0.01 0.77 0.00 0.38

*Note: Factor loadings greater than 0.3 are in bold.

The grouping of the indicators around the theoretically appropriate latent variable is remarkably

strong, stable, and clean. Every indicator loads most heavily on the factor (i.e. latent variable) with

which it is theoretically associated. Also importantly, these three theoretically identified latent

variables are the only ones that obtain large eigenvalues. This provides a strong empirical

21

confirmation for the theoretical assertions we make regarding the distinctiveness of these three types

of powersharing institutions.

The weakest associations between any indicators and the latent variable with which they are

theoretically associated are between Grand Coalition or Unity Government and inclusive

powersharing and between Ethnic Party Ban and constraining powersharing. Each of these variables

registers uniqueness scores of above 0.8, higher than any other indicators. This shows that they are

only weakly correlated with our other indicators of powersharing.17 This makes intuitive sense for

mandated grand coalitions and unity governments: while reserved legislative seats, reserved

executive positions, and military inclusiveness mandates are often used in conjunction with one

another, mandated grand coalitions and unity governments more often stand alone as the sole

institutional arrangement through which powersharing is implemented.18 With Ethnic Party Ban this

is less intuitive: such bans are simply not highly correlated with any other powersharing indicator.

Because of their low factor loadings, these two variables are assigned almost no weight in our

indices.

Section 2.1. Building Indices

Once we have confirmed both that three latent factors explain most of the variance in our 19

observed components of powersharing, and that each observed component is empirically associated

with the same factor with which it is theoretically associated, we can construct index measures of

each latent variable. This is critical because, while disaggregated data contains great richness, in

17 See Table 2.

18 Governments of national unity, in particular, often emerge as negotiated settlements to specific

crises, such as the conflict between the Orange Democratic Movement and the Party of National

Unity following the disputed 2007 presidential election in Kenya.

22

applied work based on this data, we also want to assess the effects of higher or lower levels of each

of three types of powersharing, rather than each of 19 specific institutional arrangements. We

therefore create index variables by running factor analysis on the set of indicators associated with a

given latent variable, and then using the factor loadings from the primary factor in that analysis to

weight each indicator in the index.

Section 3. Empirical Patterns in Powersharing

Section 3.1 Powersharing Trends

Figure 2 shows trends in global levels of powersharing from 1975-2010. Constraining powersharing

increased steadily and dramatically throughout the period, while dispersive institutions began

increasing in prevalence beginning in the mid 1980s. These upward trends are matched by the steady

march of democratization during the same period.

23

Inclusive institutions, on the other hand, exhibit a much more stable pattern; they actually

declined slowly in prevalence into the early 1990s, before registering a short burst of proliferation

toward the end of that decade. Yet, the overall trend for inclusive powersharing is quite flat

throughout. It also bears mention that, as a general rule, the indicators associated with inclusive

powersharing, such as mandated grand coalitions and reserved executive positions, are less common

than those associated with constraining or dispersive powersharing. Figure 3 provides histograms of

the three powersharing indices. Figures 4-6 plot the values of the three powersharing indices for a

random sample of country-years.

24

Figure 3: Histograms of Powersharing Indices

25

Figure 4: Constraining Powersharing Institutions: Random Sample

26

Figure 5: Dispersive Powersharing Institutions: Random Sample

27

Figure 6: Inclusive Powersharing Institutions: Random Sample

28

Section 4. A Comparison to Existing Data Sets

Wimmer et al. (2009) introduce the Ethnic Power Relations data set, which defines

powersharing as “Any arrangement that divides executive power among leaders who claim to

represent particular ethnic groups.” This conception of powersharing is a de facto analog to our de

jure measure of inclusive powersharing. There exist cases, such as Switzerland, where there exists

de facto powersharing, but where there exists no formal inclusive institutions. There also exist

cases, such as Lebanon in the mid-late 1980s, where de jure inclusive powersharing institutions are

not in force. These cases illustrate what we expect is a general pattern across these two types of

data: de facto powersharing can exist in the absence of formal rules primarily in “easy” cases where

the risk of violent conflict is not high, while the cases in which de jure powersharing fails to be

enforced are often those characterized by violence.

A primary focus of our research agenda is to analyze the degree to which powersharing

institutions can be implemented as a solution to intra-state conflict. De jure institutions are a policy

lever that can be pulled – de facto powersharing is not.

Section 5. Sources and Data Collection

Coders on this project employed a mix of primary and secondary source material. Primary sources

included constitutions and peace treaties, as well as official government websites and the text of

individual laws.19 Secondary sources included the Political Handbook of the World (various years)

and Europa World Yearbook (various years),20 as well as the website of the International

19 In many cases, coders relied on English translations of these documents. See Melton et al. (2013)

for a discussion of the effects of language and culture on the interpretability of constitutions.

20 Both print and online versions were used for both Europa World Yearbook and the Political

Handbook of the World.

29

Parliamentary Union (Parline, www.ipu.org), Library of Congress country studies

(http://lcweb2.loc.gov/frd/cs/), the World Encyclopedia of Police Forces and Penal Systems (1989;

2006), and Freedom in the World (various years).

The greatest challenge facing coders involved assessing the precise dates on which different

institutional forms were adopted or abandoned. Constitutional provisions are easy in this regard:

there is a date when a constitution enters into force and a date when it is nullified, amended, or

superseded. With some rules, however, it is difficult to identify the precise date at which a law was

first enacted or the date at which it was superseded.

In all cases of uncertainty, coders were directed to employ a “preponderance of the evidence”

standard rather than a “beyond a reasonable doubt” standard. Instances where the necessary

information to make an informed coding is unavailable are coded as missing. Instances where

information is available but the coder was uncertain were flagged for discussion in regular meetings

of the coding team and a group decision was made on how to code the variable. Complicated

institutional arrangements and codings perceived to be potentially controversial are discussed in the

coder notes, which are available online, along with the full codebook, list of sources, and the data set

itself at [URL REDACTED FOR REVIEW].

Section 6. Equations for Index Creation

The weights used to create the indices are described below. The indices are rescaled to a mean of

zero and a standard deviation of one to ease comparison.

Constraining Powersharing = (0.14947*Religion Protected (Discrimination)) + (0.17436*Religion

Protected (Practice)) + (0.06715*Military Legislator Ban) + (0.02426*Ethnic Party Ban) +

(0.23544*Judicial Constitution) + (0.12023*Judicial Review) + (0.39587*Judicial Tenure_1) +

(0.14688*Judicial Tenure Dummy_2)

30

Dispersive Powersharing = (0.20446*Subnational Education Authority) + (0.24015*Subnational Tax

Authority) + (0.16415*Subnational Police Authority) + (0.20277*Constituency Alignment) +

(0.22828*State Elections_1) + (0.21493*State Elections Dummy_2)

Inclusive Powersharing = (0.03115*Grand Coalition or Unity Government) + (0.17175*Mutual

Veto) + (0.30421*Reserved Seats) + (0.18562*Inclusive Military) + (0.43405*Reserved Executive

Positions)

31

Bibliography for Appendix 4

1. Beck, Thorsten, George Clarke, Alberto Groff, Philip Keefer, and Patrick Walsh. 2001. “New

Tools in Comparative Political Economy: The Database of Political Institutions.” The World Bank

Economic Review 15(1): 165–76.

2. Brancati, Dawn. 2006. “Decentralization: Fueling the Fire or Dampening the Flames of

Ethnic Conflict and Secessionism?” International Organization 60(3): 651–85.

3. La Porta, Rafael, Florencio López de Silanes, Cristian Pop-Eleches, and Andrei Shleifer.

2004. “Judicial Checks and Balances.” Journal of Political Economy 112(2): 445–70.

4. Lijphart, Arend. 1968. The Politics of Accommodation. Berkeley, CA: University of

California Press.

5. _____. 1975. “Review Article: The Northern Ireland Problem; Cases, Theories, and

Solutions.” British Journal of Political Science 5(01): 83–106.

6. Melton, James, Zachary Elkins, Tom Ginsburg, and Kalev Leetaru. 2013. On the

Interpretability of Law: Lessons from the Decoding of National Constitutions. British Journal of

Political Science 43 (02): 399-423.

7. Wimmer, Andreas, Lars-Erik Cederman and Brian Min. “Ethnic Politics and Armed Conflict.

A Configurational Analysis of a New Global Dataset.” American Sociological Review 74(2):316-

337, 2009.

32