The Foundations of Limited Authoritarian Government:

Institutions and Power-sharing in Dictatorships∗

Carles Boix† and Milan Svolik‡

Abstract

Why do some dictatorships establish institutions typically associated with democ-

racy, such as legislatures or political parties? We propose a new theoretical model of

authoritarian power-sharing and institutions in dictatorships. We argue that political

institutions in dictatorships enhance the stability of power-sharing, and therefore the

survival of these regimes. However, authoritarian power-sharing through institutions

is feasible only when it is backed by the crude but credible threat of a rebellion by

the ruler’s allies. Whereas the allies’ political opportunities – rather than a contin-

gent coordination of beliefs among them – determine the credibility of their rebellion,

institutions alleviate the commitment and monitoring problems caused by the secrecy

in authoritarian governance. Our theory generates new predictions about the empiri-

cal relationship between political institutions, leader tenure, and the concentration of

power in dictatorships, which we assess using data on the existence of legislatures in

dictatorships.∗We would like to thank Bruce Bueno de Mesquita, Barbara Geddes, Peter Lorentzen, Frances Rosenbluth,

Bonnie Weir, the participants at seminars and conferences at Princeton, Berkeley, University of Illinois atUrbana-Champaign, and the APSA, ISA and MPSA annual conventions for helpful comments.

†Department of Politics and Woodrow Wilson School of Public and International Affairs, PrincetonUniversity. Email: cboix@princeton.edu.

‡Department of Political Science, University of Illinois at Urbana-Champaign. Email: msvolik@uiuc.edu.

1 Introduction

Tyranny, or the unconstrained rule of a polity by one person, has long been treated as the

standard, almost stereotypical type of dictatorship. The classical literature on

dictatorships mainly investigated personal autocracies and the mechanisms employed by

dictators to govern and secure the acquiescence of their subjects (Xenophon/Strauss 1961;

Machiavelli 1513/1985). Similarly, the postwar literature on dictatorships focused on the

phenomenon of totalitarianism and on the means by which the totalitarian leader and his

party exercised absolute control over society (Arendt 1973; Friedrich and Brzezinski 1965;

Linz 1975, 2000; Neumann 1957).1 The formal literature which replaced that descriptive

body of work did not abandon the basic point of departure of the traditional research on

autocracies; dictatorships continue to be modeled as regimes in which a single tyrant

governs and is not subject to any external constraint or influence (Haber 2007; Kuran 1991;

Tullock 1987; Wintrobe 1998).

Yet for all their historical and theoretical importance, single-ruler dictatorships

constitute a minority of the universe of authoritarian regimes. Figure 1 displays the

proportion of dictatorships with legislatures as well as the percentage of those with at least

one political party since 1950. Only during the mid seventies did this proportion fall to less

than 60 percent.2 The proportion of dictatorships with at least one political party was even

higher and fluctuated between 80 and 90 percent from 1950 to 1990.3

1Linz (1975, 2000) is an exception in that he also examines non-totalitarian regimes.2Our data on legislatures and parties in dictatorships are from Przeworski et al. (2000) and Keefer (2002).

The fact that Przeworski and his colleagues do not report data on oil-exporting countries explains most ofthe difference with Keefers results.

3Even in regimes without any of those institutions, the leadership often maintains a smaller institution-alized body, such as a ruling council or a politburo, which sustains regularized political interaction that mayserve to restrain the tyrannical tendencies of any single ruler. Cf. Suny (1998) on the Soviet Union andMacFarquhar (1997) on China.

1

Figure 1 about here.

In this paper, we develop a new theory of institutions and power-sharing in

dictatorships. We define the conditions under which dictatorships establish governing

institutions such as parties, committees of notables, and even parliaments. The theory also

allows us to show the ways in which those institutions become a key functional component

of authoritarian regimes: they enhance the stability of power-sharing in these regimes and

thus contribute decisively to their survival.

We start with the observation that most dictators do not directly control enough

resources to govern alone. Dictators therefore seek the support of a set of notables and

share power with them. However, an important commitment problem complicates

power-sharing in dictatorships: There is no independent authority that can guarantee that

the spoils of joint rule will be divided as the dictator and his allies have agreed, since that

would imply surrendering the very powers that they wish to monopolize. In other words,

the power-sharing agreement constitutes the only foundation of political authority within

this polity. Hence the central dilemma of any non-tyrannical autocracy is to establish a

mechanism that allows the dictator and the ruling elite to credibly commit to joint rule.

Power-sharing in authoritarian regimes is ultimately sustained by the ability of the

ruler’s allies to credibly threaten a rebellion that would replace the dictator if he violates

the power-sharing agreement. (A violation of the agreement happens when the dictator

refuses to share the spoils of joint rule as agreed or when he exploits his privileged access

to information about government revenue and misrepresenting the amount of benefits to be

shared.) But, because the threat of an allies’ rebellion (which is the only punishment

device available to the dictator’s allies) is so crude in nature, the authoritarian elite has

very strong incentives to establish political institutions that will alleviate the moral hazard

2

problems they face. More precisely, we argue that the regular, institutionalized interaction

between the ruler and his allies within some party or government bodies with respected

participation and decision rules – as is the case of some politburos, governing councils, and

legislatures – reduces the suspicions that the ruler’s allies may have about the dictator

complying with the power-sharing agreement. In turn, the latter will be able to maintain a

more stable ruling coalition under less favorable circumstances than would be possible

without those institutions. Hence political institutions enhance the survival of

authoritarian regimes.

Still, political institutions only play a conditional role in dictatorships. They facilitate

authoritarian power-sharing only when the crude threat of an allies’ rebellion against the

dictator is credible. We show that rulers will abandon institutions when the distribution of

power within the ruling coalition shifts markedly in favor of the dictator.

Although theoretical research on institutions in dictatorships has been scant, several

scholars have recently examined the role of legislatures (Gandhi and Przeworski 2007;

Wright 2008), parties (Brownlee 2007; Magaloni 2006; Smith 2005), and elections (Levitsky

and Way 2003; Lust-Okar 2006) in authoritarian regimes. This body of work has made

important contributions to the study of dictatorships. However, this literature remains

incomplete in two important ways. First, with some notable exceptions, the existing

research focuses narrowly on the analysis of particular cases. More importantly, although it

concludes that institutions in dictatorships facilitate authoritarian governance, the existing

literature has not clearly identified how they do so, why the same results could not be

accomplished without them, and why they are adopted in some cases but not others. In

this paper, we develop a theory of institutions in dictatorships that provides a

comprehensive answer to this set of questions.

3

Notably, we depart from two predominant explanations of the role of political

institutions in dictatorships. The first is that autocrats adopt seemingly democratic

institutions in order to broaden their basis of support by coopting opposition to the regime

(Gandhi and Przeworski 2006). In contrast, we argue that political institutions in

dictatorships lead to durable ruling coalitions by reducing the moral hazard problem of

power-sharing, whether it is among those who already support the ruler or between the

ruler and newly recruited supporters.

The second view argues that institutions facilitate the maintenance of norms of

collective action among the dictator’s allies or opposition (North and Weingast 1989;

Myerson 2008). While such a fortuitous coordination of beliefs is possible in principle, it is

unlikely to be the key determinant of whether a rebellion against a dictator occurs or

succeeds, given how high the stakes are. Instead, we argue that the key determinants of

successful collective action in dictatorships are political facts, such as the distribution of

power or the benefits from supporting the dictator, rather than a focal coordination of

beliefs. Our model of an allies’ rebellion builds on the global games methodology (Carlsson

and van Damme 1993; Morris and Shin 2003) and we obtain a unique equilibrium by

assuming that allies do not have common knowledge of the benefit from joining a

challenger. Our paper thus contributes to the literature on collective action problems in

regime and leaderships change. Chwe (2001) and Medina (2007) develop alternative

approaches to collective action with a motivation similar to ours, whereas Dewan and

Myatt (2007) apply the global games methodology to the study of party leadership.

In fact, an important advantage of our approach to the collective action problem of

rebelling against a dictator is that it yields testable empirical predictions about the

relationship between political institutions in dictatorships, leader tenure, and the

4

concentration of power in dictatorships. Although the difficulty of measuring whether

ruling councils, parties, or legislatures indeed constrain the dictator is a serious

impediment to any large-N research design, our model leads to several predictions

consistent with existing large-N empirical findings. We predict that institutionalized ruling

coalitions, and as a consequence their leaders, will be more durable and less susceptible to

economic downturns than coalitions and leaders in dictatorships without such institutions.

This claim is supported by Gandhi and Przeworski (2007), who find that dictators with

single parties survive in office longer than those without such parties.

Our model also predicts that institutions will collapse when changes in the distribution

of power favor the dictator at the expense of his allies, and that institutionalized

power-sharing may be impossible when too much power is concentrated in the hands of the

dictator. Historical evidence from the Soviet Union (Suny 1998) and China (MacFarquhar

1997) on the collapse of “collective leadership” and the elimination of the Communist

Parties as independent political forces during the periods of Stalin’s and Mao’s rule is

consistent with this prediction. In some regimes, the discovery of natural resources or an

increase in the price of those resources may lead to such power shifts. In fact, existing

research indicates that those events result in the collapse of existing institutional

constraints on dictators (Friedman 2006; Karl 1997).

The paper is organized as follows. Section 2 develops our theoretical model of

institutions and power-sharing in dictatorships. Sections 3 and 4 examine the empirical

implications of our theoretical model on the emergence of institutions in dictatorships and

their consequences on leader tenure by employing two research strategies. Section 3

exploits the long-run variation of two cases: Mexico (from its convulsive nineteenth-century

politics to the stable PRI regime) and the Soviet Union. Section 4 turns to the statistical

5

analysis of the panel of dictatorships after 1950. Section 5 concludes.

2 The Theoretical Model

We now explore the incentives within authoritarian regimes in which the dictator rules in

coalition with a set of individuals. In most autocracies the dictator does not have the

resources to govern alone and must rely on the support of allies in order to hold power.

However, this agreement to share power and govern according to some agreed-upon set of

rules is bedeviled by an important commitment problem.

To obtain higher rents, to lower the probability of rebellions among ambitious notables

or to build up a more cohesive country that can be then employ to expand abroad, the

dictator will strategize to get rid of or at least subject with a much firmer grasp the class of

notables. The leader may try to check, imprison or kill all the notables at the same time

and then proclaim himself absolute ruler or monarch. Alternatively, he may use more

gradual tactics. He may get rid of a few notables at a time without the rest taking notice,

proceed to absorb the fortunes and power of those he just eliminated and use his growing

power to further purge more notables. A few sequential rounds of such type of “salami”

tactics will transform him into a tyrant.4 These intra-elite tensions are exacerbated by the

secrecy typically associated with authoritarian governance. If the allies of the dictator have

limited or no information about actual government revenue, they will have to rely on the

ruler’s report of revenue when they claim their promised share in the spoils of government.

4In turn, one or several local notables may as well try to upset the existing balance of power. They mayattempt to accumulate resources, organize particular networks of supporters or reshape the class of notablesitself to constrain the dictator or to fire him. Conversely, one or several allies may as well try to upsetthe existing balance of power. They may attempt to accumulate resources, organize particular networks ofsupporters or reshape the coalition of allies to constrain the dictator or get rid of him. We do not examinethis possibility in the paper.

6

Because the allies anticipate that the ruler will be tempted to understate government

revenue, the possibility of maintaining the autocratic deal will become even lower.

A verbal pact among autocrats, i.e. an orally given promise to respect the position or

status of everyone and to consult everyone informally to decide over any issue is not

sufficient to sustain an autocratic deal over the medium run. A mere written agreement -

in the form of a contract accepted and signed by everyone is not enough either. As in any

pact, the autocratic deal needs to rely on some external guarantees, that is, on some

guarantees that go beyond the strict promise to behave well and keep the agreement.

The autocratic “pact” will only last over a relatively prolonged period of time if some

public body or institution confirms the nature of the deal made among autocrats and

guarantees its maintenance. This body (or organization) cannot be a third party

independent from the autocratic elite to whom the latter entrusts the enforcement of the

pact since in that instance the sovereignty of the state would be actually lying in its hands

and not under the control of the governing elite. The institution (or a set of institutions)

which embodies (and preserves) the pact must be some structure where the members of the

elite are represented or in which they participate and which reflects the nature of the pact

among the members of the elite. Through this institutional structure, the ruling elite

receives the right type of information about the resources of the dictator and the internal

flows of income within the elite and, therefore, more generally, about the current balance of

power and the possible attempts made to alter the latter. The allies of the dictator make

sure that the leader does not develop strategies to shift the distribution of power, assets

and status. Similarly, every member of the ruling coalition observes the nature, size and

stability of the existing factions in the country. In doing so, he verifies that no section of

the ruling class is too loyal to the leader or, in other words, too ”monarchical.” The very

7

routine of meeting in a committee, party congress or assembly serves as a yardstick to

measure the intentions of the leader. Any attempt by the national leader to block or not

convene his allies is a signal that he is indeed intent on disrupting the old balance of power

and should therefore trigger an immediate backlash from them.

2.1 Allies’ Rebellion as a Collective Action Problem

In order to investigate when and how institutions in dictatorships facilitate authoritarian

power-sharing, we now develop a simple formal model. Our formal argument proceeds in

three steps. First, we examine the collective action problem that the ruler’s allies must

overcome in order to replace the dictator. This allows us to identify the circumstances

under which the threat of an allies’ rebellion is credible and, in turn, the circumstances

under which power-sharing will be feasible without institutions. Finally, we demonstrate

that institutions that improve the transparency among the ruler and his allies allow for the

formation of more stable ruling coalitions and under less favorable circumstances than

would be possible without such institutions.

Consider an authoritarian polity in which political power is controlled by a ruler and a

continuum of notables.5 The ruler controls a share λ of the total power, while the notables

control the rest, 1− λ. In order to stay in office, the ruler needs to maintain a ruling

coalition that commands at least a κ majority of the total power; we call κ the effective

power threshold.6 When λ > κ, the ruler controls a sufficient amount of power in order to

5The assumption that notables are atomless players simplifies the analysis below, but our results also holdin a setting with a finite number of allies. In substantive terms, this assumption states that the notablesenjoy significant influence locally, but the power of any single notable is of little consequence at the nationallevel.

6Although our concept of the ruling coalition is close to Bueno de Mesquita et al.’s (2003) winningcoalition, there is an important difference between the two: whereas Bueno de Mesquita et al. measure thesize of the winning coalition in individuals, our metric is the amount of power held by the ruling coalition.Thus we capture a distinctive feature of dictatorships: the existence of large differences in power across

8

rule alone. But when λ < κ, the ruler needs to recruit some allies from among the notables

in order to stay in office. In that case, we assume that the ruler recruits allies with the

minimum joint power required for him to stay in office, µ = κ− λ. Furthermore, we assume

that an alternative coalition of notables that would not include the ruler but would survive

in office is also feasible, λ < 1− κ. Thus λ ∈ (0, 1− κ). Figure 2 illustrates this model of

an authoritarian polity.

Figure 2 about here.

When the ruler recruits allies, he promises each ally a share β > 0 of the total benefits

from joint rule per unit of power the ally holds. For instance, when β = 1, then the share

of benefits paid to the allies is µβ = µ. That is, the share of benefits paid to the allies

corresponds to the power held by them. These benefits may take the form of government

revenue, bureaucratic appointments, or policy choices. The allies may value these benefits

because of pecuniary or ideological reasons, or because they allow them to compensate

their followers or cultivate their local political influence.

However, the benefits from joint rule may differ across periods as a result of variation in

administrative costs, economic performance, or political turmoil. In order to keep our

analysis as simple as possible, we assume that benefits are 1 with probability π (good times)

and they are 0 with probability 1− π (a crisis). Thus when the ruler keeps his promise,

each ally receives the payoff β with probability π and the payoff 0 with probability 1− π.

We assume that power-sharing between the ruler and the allies is politically desirable:

In expectation, the ruler maintains µ allies and still keeps a positive share of total benefits,

π − µβ > 0. On the other hand, our assumptions about the payoffs to the allies imply that,

individuals.

9

as long as the ruler keeps his promise to pay a β fraction of benefits to each ally, each ally

receives a nonnegative payoff in any period as well.

In order to understand when power-sharing between the ruler and his allies will

succeed, we start by examining a central feature of authoritarian politics: the only

punishment that the allies may use to deter the ruler from reneging on his promise to

compensate them for their support is to replace him with a challenger. While any single

notable is too weak to compel the ruler to comply with their power-sharing arrangement, a

collection of notables may be able to credibly threaten to join a challenger. The credibility

of that threat in turn determines the terms of any power-sharing agreement that the ruler

will abide by in the first place.

Hence our first step is to examine the credibility of the allies’ threat to abandon the

ruler in favor of a challenger - the threat of an allies’ rebellion. The choice of the word

“rebellion” in our model should not be taken too literally. In our use of the term allies’

rebellion, we are motivated by the recognition of the right to a barons’ rebellion in Clause

61 of the Magna Charta of 1215 in the case of the King’s transgression against the Charter.

But most modern rebellions are called coups or plots or even revolutions (as in the 1968

July Revolution that brought the Baath Party to power in Iraq). Below we examine the

central role that the participants’ costs, benefits, and the expectation of the success or

failure of such collective endeavors play in their credibility as threats against the ruler’s

opportunism.

In a rebellion, each ally may either support the ruler or rebel against the ruler by

joining a challenger. As long as the ruler remains in power, each ally who supports the

ruler receives her share of benefits b ≥ 0 from the ruler. If the ruler keeps his promise and

shares β with the allies as agreed than b = β. However, the ruler may also renege, in which

10

case b = 0. Alternatively, if a rebellion is staged and succeeds, the allies who joined the

challenger will enjoy the share of benefits promised by the challenger, θ. On the other

hand, those allies who supported the ruler will loose any benefits and receive the payoff

zero. However, if a rebellion fails, an ally who joined the challenger will receive the payoff

θ − r, where r > 0 represents the ruler’s punishment of those who participated in a failed

rebellion. Thus while rebeling entails the risk of a lower payoff in the case of a failure,

supporting the ruler is also risky since a rebellion may succeed.7

Whether a rebellion succeeds depends on the proportion of allies that join the

challenger, which I denote ρ. The rebellion succeeds when ρ > ρ∗ and fails otherwise.

These payoffs are summarized in Figure 3.

Figure 3 about here.

We assume that all aspects of this setting except for the payoff from a successful

rebellion θ are common knowledge. More precisely, each ally privately observes an

imperfect signal si of the payoff from a successful rebellion θ. The signal si is distributed

uniformly on the interval [θ − ε, θ + ε], and the realizations of si are independent across

allies. We think of ε > 0 as “small.” In turn, we can say that each ally’s signal si comes

with a small, idiosyncratic noise. These difference in allies’ information about the payoff

from a successful rebellion may correspond to different estimates of the credibility of the

challenger’s offer by each ally. Alternatively, the allies may learn about the challenger’s

offer via separate private channels and thus have correlated but marginally different

information about it. For expositional simplicity, we assume that θ has a uniform prior

7In many instances, r will be so large that it will overwhelm any positive value of θ in the payoff from afailed rebellion θ − r. This may be the case when the punishment from a failed rebellion is a certain deathor imprisonment. But there may also be circumstances when the quantities θ and r will be sufficiently highand low, respectively, so that an ally prefers to be the “second in command” if the rebels continue to existin opposition or exile after a failed rebellion to supporting the current leader for a very small b.

11

density on the interval [1− σ, 1 + σ], σ > 0. In other words, before observing the signal si,

each ally assigns the same probability to any value of θ in the interval [1− σ, 1 + σ].8

When should an ally join a rebellion against the ruler? To answer this question,

consider first an alternative, simpler setting in which the benefit from a successful rebellion

θ is public information and thus common knowledge among the allies. If θ < 0, each ally

prefers supporting the ruler to rebeling, regardless of how many allies plan to rebel.

Alternatively, if θ > b + r, rebeling strictly dominates supporting the ruler.

When the payoff from a successful rebellion θ is in the interval [0, b + r], however, this

model resembles a multi-person Stag Hunt. That is, supporting the ruler is an ally’s

optimal choice whenever at most ρ∗ allies rebel, and rebeling is her optimal choice as long

as more than ρ∗ allies rebel. Thus whether a rebellion succeeds is unrelated to the benefit

from supporting the ruler b, the cost of a failed rebellion r, or the participation threshold

ρ∗; it depends exclusively on what each ally believes about the intended actions of the

others.

This multiplicity of equilibria disappears in the present setting where each ally observes

only an imperfect signal si of the payoff from a successful rebellion θ. Given our

assumptions about the distribution of si, each ally has an unbiased estimate of θ. More

precisely, after ally i observes the signal si, she believes that θ is distributed uniformly on

the interval [si − ε, si + ε], and her expectation of θ is si. However, she does not know the

signals s∼i that other allies observed, and in turn, the true value of θ is not common

knowledge.

Suppose therefore that each ally follows a threshold strategy according to which she

8These simplifying distributional assumptions are inconsequential as long as the support of θ containsthe interval [0, b + r] and ε is small relative to σ. The mean of the prior distribution of θ can be interpretedas the long-run average payoff from being a member of the ruling coalition.

12

rebels when her signal si exceeds some threshold s∗ and supports the ruler otherwise. Then

in equilibrium, each ally must be indifferent between supporting and rebeling against the

ruler whenever si = s∗. When si = s∗, ally i’s expected payoff from supporting the ruler is

Pr(ρ ≤ ρ∗)b + [1− Pr(ρ ≤ ρ∗)]0 = Pr(ρ ≤ ρ∗)b.

On the other hand, her expected payoff from rebeling is

Pr(ρ ≤ ρ∗)(s∗ − r) + [1− Pr(ρ ≤ ρ∗)]s∗ = s∗ − rPr(ρ ≤ ρ∗),

given that the expectation of θ is si. Therefore, for an ally who observes the signal si = s∗,

we have

Pr(ρ ≤ ρ∗) =s∗

b + r. (1)

In order to find the threshold signal s∗, we need to compute the equilibrium probability

that a rebellion will fail, Pr(ρ ≤ ρ∗), for an ally with the signal si = s∗. Given the threshold

strategy, the proportion of allies ρ that rebel corresponds to the proportion of allies with

the signal si > s∗. Given some payoff from a successful rebellion θ, this proportion is

ρ =θ + ε− s∗

2ε.

In turn, ρ ≤ ρ∗ when

θ + ε− s∗

2ε≤ ρ∗,

or equivalently, when

θ ≤ s∗ + 2ερ∗ − ε. (2)

13

Thus we have

Pr(ρ ≤ ρ∗) = Pr(θ ≤ s∗ + 2ερ∗ − ε) =s∗ + 2ερ∗ − ε− (s∗ − ε)

2ε= ρ∗.

In other words, an ally with the signal si = s∗ believes that the proportion of allies that

will rebel is distributed uniformly,

Pr(ρ ≤ ρ∗) = ρ∗. (3)

Substituting (3) into (1), we see that, in equilibrium, the allies follow a threshold strategy,

with the threshold signal

s∗ = ρ∗(r + b). (4)

In effect, the signal si coordinates allies’ beliefs about the likelihood of a successful

rebellion. Importantly, this equilibrium is unique and thus requires no additional

assumptions about the formation of allies’ beliefs.9

The equilibrium threshold (4) implies a very simple and intuitive relationship between

the likelihood of a successful rebellion and the key elements in our political setting. In

order for an ally to rebel, the imperfect signal si of the payoff from a rebellion θ must be

higher: i) when a greater proportion of allies is required for a successful rebellion (high ρ∗),

ii) when the payoff from supporting the ruler is high (high b), and iii) when the cost of a

failed rebellion is high (high r).

The equilibrium threshold (4) also implies that the punishment of those who

participated in a failed rebellion and benefits are substitutes from the ruler’s point of view.

9That is, our setting satisfies the general conditions for a unique equilibrium in a global game, as discussedin Morris and Shin (2003).

14

The choice of the two policies may therefore depend on the cost of repression relative to

that of economic benefits, which may vary across regimes (e.g. military vs. civilian

dictatorships) or circumstances (e.g. economies at different stages of development.) In fact,

Wintrobe (1998) assumes that repression and benefits (loyalty in his terminology) are

substitutes; we derive this relation within a fully strategic model of an allies’ rebellion.

When should the ruler expect a rebellion to succeed? The ruler must form an

expectation about the likely success of a rebellion without observing the signal si. The

threshold signal s∗ implies that there is a threshold benefit from a successful rebellion θ∗,

such that a rebellion succeeds for any θ > θ∗. Using (2) and (4), we have

θ∗ = ρ∗(r + b) + 2ερ∗ − ε for θ∗ ∈ [1− σ, 1 + σ].

Thus the ruler expects that the allies’ rebellion succeeds when θ > θ∗ and fails otherwise.

Then the probability of a successful rebellion is

Pr(θ > θ∗) =

0 if θ∗ < 1− σ,

1+σ−θ∗2σ

if θ∗ ∈ [1− σ, 1 + σ],

1 if θ∗ > 1 + σ.

(5)

Like the threshold signal s∗, the threshold benefit from a successful rebellion θ∗ also

depends on the key elements in our political setting in an intuitive way: a large payoff to

the allies b, a large proportion of allies that is required for a successful rebellion ρ∗, and

high cost of a failed rebellion r raise the equilibrium threshold benefit from a successful

rebellion θ∗ and thus lower the probability of its success. In other words, the ruler knows

that a rebellion is more likely to succeed if he pays his allies poorly, when a small fraction

15

of them must defect to the challenger in order for a rebellion to succeed, or when the

punishment for those who participate in a failed rebellion is lenient.

In fact, recall that the ruler recruits allies with the minimum joint power that satisfies

the effective power threshold, µ = κ− λ. In order for the notables to form a coalition that

excludes the ruler and commands a κ majority of the total power, 2κ− 1 notables must

abandon the ruler. Thus the fraction of allies required for a successful rebellion is

ρ∗ =2κ− 1

κ− λ.

Note that ρ∗ is increasing in the ruler’s relative power λ. In other words, weak rulers –

rulers who need to maintain a large coalition of allies in order to stay in power – are more

vulnerable to a rebellion, because the defection of a smaller fraction of allies will be

sufficient for a successful rebellion.

Proposition 1. In a unique Bayesian Nash equilibrium, each ally supports the ruler if

si ≤ s∗ and rebels if si > s∗, where s∗ = ρ∗(r + b). An allies’ rebellion succeeds if θ > θ∗,

where θ∗ = ρ∗(r + b) + 2ερ∗ − ε and ρ∗ = (2κ− 1)/(κ− λ).

2.2 Authoritarian Power-Sharing without Institutions

Above, we have established how the credibility of the threat of a rebellion depends on the

key factors in our political setting: the payoff from supporting the ruler, the number of

allies that the ruler needs in order to stay in power, and the punishment for those who

participate in a failed rebellion. We can now examine how the credibility of this threat

affects the possibility and the terms of a power-sharing agreement between the ruler and

the allies.

16

The timing of actions in this extensive game is as follows. In period t = 0, the ruler and

the allies form a power-sharing agreement according to which the ruler pays allies a β share

of the total benefits from joint rule in each period. The timing of actions in any period

t ≥ 1 is depicted in Figure 4. First, nature determines the size of benefits. Then the ruler

privately observes the size of those benefits, reports it (and possibly lies) to allies, and pays

them. Then the allies observe the ruler’s report and their compensation but not the size of

benefits. Finally, if a rebellion is staged, each ally observes a signal of her payoff under the

challenger and either supports the ruler or rebels against him. If the rebellion succeeds, the

game ends and a new power-sharing agreement forms between the challenger and his allies.

On the other hand, if the rebellion fails, the same power-sharing agreement remains in

place but the rebellious allies are replaced by new ones from among the notables. We study

a Markov Perfect Equilibrium in which the allies condition their actions in any period t ≥ 1

on the ruler’s announcement of the size of total benefits, the compensation they receive,

and, if a rebellion is staged, the challenger’s offer.

Recall that a rebellion is the only punishment that the allies can threaten the ruler

with. In order to compel the ruler to share power as agreed, it must fulfil two objectives.

First, it must discourage the ruler from reneging on his promise to pay allies a fraction β of

the benefits. But the same threat must also deter the ruler from lying about the size of

benefits. In fact, either type of defection hurts the allies equally and yields the same benefit

to the ruler. If the ruler reneges on his promise to pay allies β, the allies receive the payoff

0. If the ruler lies about the size of benefits – when he claims that the size of benefits was 0

when it was 1 – than the allies receive the payoff 0 as well. Thus the two types of defection

are indistinguishable when the size of benefits is observed only by the ruler.

Consider first how the allies may use the threat of a rebellion in order to deter the ruler

17

from reneging on his promise to share with each of them a β fraction of the total benefits.

The allies may threaten to rebel in any period in which they receive a payoff other than β.

But note that the threat of a rebellion does not imply that each ally will unconditionally

join any challenger. Once the ruler defects, each ally considers the challenger’s offer (based

on her signal si) and decides whether to rebel. This is what we mean when we say that

“allies rebel.”10

But the allies also fear that the ruler will lie about the size of benefits in order to avoid

paying the allies. For instance, the ruler may claim that the government is in a fiscal crisis

or that the bureaucratic appointments promised to the allies have to be delayed for some

reason, simply because he does not want to share β with each ally. Given the lack of any

subtler instruments that would discourage lying, the allies must also threaten to stage a

rebellion after the ruler claims that the size of benefits is 0. In turn, the threat of a

rebellion will outweigh the immediate benefit to the ruler from lying when

1− µβ + δV ≥ 1 + (1− φρ∗)δV, (6)

where δ ∈ (0, 1) is a discount factor, φρ∗ is the probability of a successful rebellion when

each ally receives the payoff 0 and the proportion of allies required for a successful rebellion

is ρ∗, and V is the ruler’s expected discounted payoff given the allies’ threat,

V = π(1− µβ + δV ) + (1− π)(1− φρ∗)δV =π(1− µβ)

1− δ[1− φρ∗(1− π)].

Since the two types of defection – not sharing benefits and lying about their size – are

10We can check that once a rebellion is in place, the ruler pays allies 0 and that each ally considers thechallenger’s offer. Alternatively, no ally has an incentive to consider the challenger’s offer when a rebellionis not in place, as long as other’s do not. Thus the ruler’s and allies’ actions constitute a Bayesian Nashequilibrium in each period, both during a rebellion and when a rebellion is not in place.

18

indistinguishable, and the same punishment – an allies’ rebellion – is used to discourage

both, the incentive constraint (6) also describes the incentives that will discourage the ruler

from not sharing benefits as agreed.11 Solving (6) for δ, we see that the ruler will comply

with the power-sharing agreement as long as

δ ≥ µβ

µβ + φρ∗(π − µβ). (7)

Recall that π − µβ > 0. Thus the greater the probability that a rebellion succeeds and

the lower the payoff to each ally β, the greater the range of discount factors under which

power-sharing is possible.12 But the likelihood of fiscal crises also affects the feasibility of

power-sharing: the more likely such crises are, the harder it is to share power.

2.3 Political Institutions and Authoritarian Power-Sharing

The discussion above highlights the limits to authoritarian power-sharing when the threat

of an allies’ rebellion is the sole deterrent against the ruler’s opportunism: Even though the

threat of a rebellion may be sufficient to compel the ruler to share benefits as agreed, it

does so at the price of potentially collapsing in each period with a crisis, which happens

with the probability (1− π)φρ∗ . Thus both the ruler and the allies would prefer to

eliminate any asymmetries of information among each other.

This is precisely what an institutionalized power-sharing agreement may accomplish.

Both the ruler and the allies will benefit from establishing institutional mechanisms that

reveal the size of benefits from joint rule to all parties in each period. Such mechanisms

11Both constraints bind only during good times, when total benefits are 1.12The effect of ruler’s initial power λ is ambiguous: higher λ implies lower µ and thus a lower total

expenditures on allies, which lowers the attractiveness of a defection. But higher λ also implies higher ρ andthus decreases the credibility of a rebellion and thus raises the attractiveness of a defection.

19

could be, for instance, periodic reviews of government spending and revenue, or the

consultation of major policies by a council of allies or their representatives. These

institutional mechanisms will reduce the opportunities for the ruler to lie about the size of

benefits from joint rule and, in turn, reduce the suspicion among his allies that he is doing

so. To simplify the analysis, we assume that institutions completely reveal the size of

benefits to the allies.13 Lying about the size of benefits is, in turn, observable to the allies.

Once such an institutionalized power-sharing agreement is in place, the threat of an

allies’ rebellion will serve to deter the ruler from circumventing those institutions and from

the downright refusal to share benefits as agreed, both of which are now observable to the

allies. The ruler will comply with the power-sharing agreement as long as inequality (7) is

satisfied, but now the ruler’s expected discounted payoff is

V = π(1− µβ) + δV =π(1− µβ)

1− δ.

Thus the ruler complies with an institutionalized power-sharing agreement when

δ ≥ µβ

µβ + πφρ∗(1− µβ). (8)

Importantly, the range of discount factors under which power-sharing is possible is

always greater when power-sharing is institutionalized than when it is not. When we

rewrite the threshold discount factors under power-sharing with institutions, δI from (8),

13The intuition in the more realistic case when institutions reveal the size of benefits via an imperfect butcorrelated signal is very similar to the argument here.

20

and without institutions, δ∼I from (7), as

δI =1

1 + πφρ∗

(1

µβ− 1

) and δ∼I =1

1 + πφρ∗

(1

µβ− 1

π

)

we see that the latter is larger than the former, δI < δ∼I . This result is intuitive: When

power-sharing is institutionalized, the allies no longer need to stage a rebellion every time

the ruler claims there is a crisis. In turn, the expected payoff from power-sharing is greater,

which reduces the ruler’s temptation to renege on it. Thus institutions allow for

power-sharing when it otherwise would not be possible.

Proposition 2. In a Markov Perfect equilibrium, the ruler complies with an

institutionalized power-sharing agreement and allies support the ruler as long as δ ≥ δI ,

whereas the ruler honors a power-sharing agreement that is not institutionalized and allies

support the ruler as long as δ ≥ δ∼I . Moreover, δ∼I > δI .

The implications of this result are sharpest when we consider how a change in a key

factor in our political setting – the distribution of power between the ruler and the allies –

affects the feasibility of power-sharing. Consider what happens when the distribution of

power changes in favor of the ruler: once the ruler controls more power λ, the fraction of

allies who must rebel in order to remove him ρ∗ increases, which according to (5) lowers

the likelihood of a successful rebellion. As a result, defection is now more attractive to the

ruler. In other words, both δ∼I and δI are increasing with a positive change in the ruler’s

power 4λ, via a decline in the likelihood of a successful rebellion φρ∗ .14

14Both δ∼I and δI are also decreasing in π: When the expected payoff from power-sharing increasesbecause crises are less likely, power-sharing is easier to sustain.

21

2.4 A Numerical Illustration

These results may be best illustrated with the help of a numerical example. Suppose that

κ = 5/9 and λ = 1/9, then µ = 4/9. Furthermore, suppose π = 0.6, β = 1, r = 2, ε = 0.1

and σ = 2. For these parameters, Figure 5 illustrates how a change in the ruler’s power 4λ

affects the feasibility of power-sharing. We plot the threshold discount factors under

power-sharing with and without institutions (solid and dashed line respectively) against

changes in the ruler’s power 4λ. We see that an increase in the ruler’s power reduces the

range of discount factors under which power-sharing is feasible, both with and without

institutions. However there is also a range power shifts after which power-sharing will

collapse without institutions, but will survive if it is institutionalized.

Yet as Figure 5 indicates, institutions may not save a power-sharing agreement once the

ruler’s power grows too large. This result may explain why the discovery of natural

resources or an increase in the price of those resources are frequently followed by the

collapse of the existing institutional constraints on dictator’s (Friedman 2006; Karl 1997):

the discovery of natural resources or an increase in the price of those resources advance the

power of the ruler relative to that of the allies, which according to our model narrows the

range of discount factors under which power-sharing will be feasible, and may result in the

breakdown of power-sharing even if it is institutionalized.

Figure 5 about here.

What happens when power-sharing breaks down? When power shifts substantially in

favor of the ruler or when the expected benefits from joint rule decline sharply,

power-sharing may no longer be feasible without renegotiating the share of benefits β that

the allies receive in return for their support. If such a renegotiation does not succeed, the

22

ruler violates the power-sharing agreement in every period while the allies support the ruler

in any period only because a more attractive challenger is not presently available and, in

fact, abandon the ruler as soon as such a challenger appears. Thus our model predicts that

these polities will be unstable, with frequent leadership changes.

To summarize, we show that the terms and the stability of power-sharing between the

ruler and the allies depend on the credibility of the allies’ threat of a rebellion. This threat

is credible when the ruler needs to maintain a large number of allies in order to stay in

office, when the cost of a failed rebellion is low, and when the challenger’s offer is high

relative to the benefits that allies receive from the ruler. Institutions expand the range of

circumstances under which power-sharing is feasible by eliminating asymmetries of

information between the ruler and the allies and allow for power-sharing when it otherwise

would not be possible. Yet even institutionalized power-sharing agreements may collapse

under unfavorable circumstances, in particular, if the ruler’s power grows too large or if the

expected benefits from joint rule decline.

3 Historical Evidence: Authoritarian Institutions in

Mexico and the Soviet Union

[UNDER CONSTRUCTION.]

23

4 Empirical Analysis: Institutionalized Dictatorships

and Leader Tenure in a Panel of Dictatorships

In this section, we test several key theoretical propositions and insights derived from our

historical analysis by employing the panel of dictatorships after 1950. We first investigate

the proposition that dictators introduce and maintain political institutions, such as

legislatures, party committees or ruling councils, in order to facilitate power-sharing

between the leader and the allies, but the existence of these institutions is conditional on a

permissive balance of power between the ruler and his allies. We then test our claim that,

by reducing asymmetries of information among the ruler and his allies, these institutions

reduce the need for allies rebellions and thus result in longer leader tenures.

To assess these claims, we exploit the available large-N data on legislatures [and parties]

in dictatorships. Our results, which confirm our theoretical expectations, come with two

caveats. First, a legislature [and a party] is [are] only one of several institutions which may

serve to reduce asymmetries of information between the ruler and his allies and thus lead

to more stable authoritarian power-sharing. Still, we think it is appropriate to employ

them as a measure of institutionalized dictatorships because they are a good indicator of

strong power-sharing agreements [as opposed to other institutional structures that may be

less formalized and therefore less resistant to power shifts]. Second, large-N data

approximate but do not capture actual behavior. That is, they do not track the extent to

which the regular interaction within the institution actually reduces asymmetries of

information between the ruler and his allies or the extent to which the formal rules

concerning membership, procedures, and decision making are followed.

24

4.1 Covariates of Institutionalized Dictatorships

For the purposes of this paper, we define as a dictatorship any regime where at least one of

the two following conditions is not met: free and competitive legislative elections and an

executive that is accountable to citizens, either directly via elections in presidential systems

or indirectly via the legislature in parliamentary systems.15

In order to determine whether a country has a legislature, we use Banks’s (2001) dataset

for the period from 1950 to 1995 and complement it with Keefer’s “Database of Political

Institutions” (Keefer 2002) through 1999. Whenever there are discrepancies between the

two datasets, we reconcile them by employing direct written sources. The data includes

4,234 country-years: 3,109 country-years (or 73.4 percent) show an elected legislature; 276

country-years have unelected legislatures (i.e. legislative bodies appointed by the ruler);

and 849 country-years have no legislature. Given the distribution of the data, we classify a

country as having a legislature when it has an independently elected legislative body.

Measuring the underlying distribution of power of any given ruler is a daunting, if not

impossible enterprise. Accordingly, we have decided to employ a set of tentative proxies

that may capture the likelihood or the ease with which a dictator may rule alone:

1. Level of development, measured as the log of per capita income: Dictators should

have a harder time controlling its subjects in developed societies, where human

capital is more widespread and where thriving markets are based on widespread sets

of formal institutions, social norms and generalized trust.

2. Log of population: Our expectation is that if, other things being equal, ruling alone is

easier in small countries, legislatures should be positively correlated with size of the

country. Naturally, the reverse could be also true: if collective action problems are

15The definition and the coding is taken from Boix and Rosato (2001).

25

harder to solve as population increases, dictators may govern alone more often in

large countries.

3. Ethnic fractionalization. If ethnic heterogeneity increases the costs of coordination

across citizens, we should see fewer legislatures in fractionalized countries.

4. Production structure: Dictators will be in less need of allies in economies with a

production structure that can be easily controlled and exploited by the state. The

extreme example is a country with a single natural resource located at a unique

location and easily susceptible to extraction. The ruler who controls such a resource

would be able to pay off a set of subordinates who would substitute for allies. In our

analysis, we include two measures of production structure: the oil share of exports (a

dummy variable that equals 1 if oil accounts for a third or more of total exports) and

an index of export concentration (the Hirsch-Herfindhal index of concentration based

on 239 three-digit standard international trade classification categories of exports as

estimated by UNCTAD) which varies from 0.045 (a highly diversified economy) to 1

(an economy that exports only one product).

5. Foreign support: We hypothesize that foreign support may substitute for internal

support. Legislatures should be less frequent in those dictatorships that can rely on

foreign support. We proxy foreign support by creating a dummy variable for the Cold

War period (1950-90). During the Cold War, a considerable number of dictators

received support from one of the two superpowers. After 1990, the strategic value of

vast parts of the Third World declined and the United States and Russia withdrew

economic and military support from many of their former clients. In recent work,

Fearon and Laitin (2007) and Balcells and Kalyvas (2008) find that the end of the

26

Cold War and the related decline in foreign support put an end to numerous civil

conflicts across the world. This logic explains the growth of legislatures in

authoritarian regimes after 1990./footnoteIt may also explain for the introduction of

parties and party systems in many authoritarian regimes after 1990. Until the late

1980s less than a third of all legislatures had more than one party. By the mid 1990s

this proportion had just reversed. We also think that it is an alternative to current

theories which, bracketing the assumption of rational self-interested behavior among

dictators, prefer to emphasize the diffusion of certain legitimacy norms to explain

broader authoritarian regimes (see, for example, Levitsky and Way 2003).

6. Office entry: We distinguish among the following entries into office: an independence

war, a civil war, a coup d’etat, dynastic succession, a revolt, elections and foreign

intervention. We hypothesize that successful independence and civil wars strengthen

the position of leaders vis-a-vis their allies and are thus associated with a lesser need

of power-sharing.

We estimate the effect of our independent variables on both the likelihood of the

adoption of a legislature and its maintenance once in place using the dynamic probit

model.16 The dynamic probit model estimates two sets of coefficients, β and α. The β

coefficients capture the unconditional effect of an independent variable on the presence of a

legislature and thus estimate the probability that a legislature will be created when none

exists. The α coefficients estimate the effect of an independent variable on the presence of

a legislature conditional on that legislature being already in place. Hence the sum of the

two coefficients indicates the probability that a legislature will remain in place once it

already exists.

16See Amemiya (1985: chapter 11) for the estimation and properties of the dynamic probit model.

27

Table 1 about here.

Table 1 displays the estimates of four different models. Model 1 includes the economic

and social variables (per capita income, population, ethnic fractionalization and oil

exports) and the Cold War effect. Model 2 adds the index of export concentration.

Because of data limitations, the universe of observations falls from almost 3,700 in Model 1

to about 2,000 in Model 2. Models 3 and 4 add the political variables on the type of entry.

In each we report the β coefficient in the first column and the sum of the β and the α

coefficients in the second column of each model. For each coefficient we report the

corresponding p-value.17

Population and ethnic fractionalization have no impact on either the introduction or

maintenance of legislatures: the estimates are both negligible in size and not statistically

significant (except for population in Model 3). The level of development has no impact on

the emergence of legislatures (except in Model 2). It seems to stabilize existing legislatures

- the β and α joint coefficient is statistically significant in Models 1 and 3. It ceases to have

any effect when we include a control for export concentration. From a substantive point of

view, however, this stabilizing effect is minor: with all the other covariates held at their

median values, increasing per capita income from the bottom sample quartile to the top

quartile raises the probability that a legislature will continue to exist by about 1 percent.

Furthermore, oil exports do not affect the emergence or survival of legislatures.

However, the effect of export concentration on the existence of legislatures is significant

and important in size. The annual probability that a dictator governing a highly diversified

economy (with a concentration index of 0.1) will set up a legislature is 20 percent (with all

the other variables held at their median values). The chances a dictator does so in a

17See Epstein and OHalloran (2005) for the estimation of the p-value for the joint β and α coefficients.

28

single-product economy (with a concentration index of 1) drop to 1 out of 25.

Table 2 about here.

The impact of the international system is also substantial. The β coefficient of the

post-Cold War dummy, which captures the unconditional effect of the end of the Cold War

on the existence of legislatures, is substantively and statistically significant. Table 2 shows

the simulated effect of the end of the Cold War for different levels of economic

concentration with all the other variables at their median values. In single-export

economies, the annual probability of establishing a legislature almost doubles from 3.4 to

7.8 percent. In highly diversified economies, this probability increases from 23 to 34

percent.

Models 3 and 4 assess whether the type of leader entry affects the existence of

legislatures. No type of entry is statistically significant as an explanatory variable for the

creation of legislature except for revolts in Model 4. Dictators who come to power through

civil wars and coups are less likely to maintain legislatures – the decline in the annual

probability is between 7 to 11 percent.

4.2 Institutions and Leader Survival

Our argument in section 2 implies that power-sharing between the ruler and the allies will

be more stable when supported by political institutions. In terms of the observable pattern

of leadership change in authoritarian regimes, this proposition implies that i) political

institutions should be associated with longer tenures tenures, and ii) leadership change will

occur within these institutions rather than outside them. As above, we will use the

available data on the existence of legislatures to examine these hypotheses.

29

Table 3 lists the absolute and relative frequencies of the different ways in which leaders

leave office in dictatorships with and without legislatures. We see that the coup is the most

frequent type of leader exit in both dictatorships with and without legislatures. Coups are

followed by natural causes and elections in dictatorships with legislatures and revolts and

natural causes in dictatorships without legislatures. The relative frequency of coups and

revolts however differs greatly between dictatorships with and without legislatures:

Dictators without legislatures are about three times more likely to be removed in a coup or

a revolt than dictators with legislatures, and only about half as likely to leave office due to

natural causes. This pattern is consistent with our theoretical analysis, which implies that

allies’ rebellions, which are observed primarily as coups and sometimes as revolts in our

data, should occur more often in dictatorships without institutions.

Dictatorships with and without legislatures also differ in terms of the mean duration of

leader tenures, as predicted by our theoretical model. Leaders in dictatorships with

legislatures survive in office for an average of 8.47 years, whereas the corresponding figure is

6.06 years in dictatorships without legislatures, and this difference is statistically significant

at all conventional significance levels. However, this difference in the mean duration of

leader tenures could be the result of the different circumstances under which dictatorships

with and without legislatures are observed. For instance, dictatorships without legislatures

may disproportionately emerge in poor countries and a low level of development may be

associated with shorter leader tenures. We therefore examine the effect of legislatures on

the likelihood of coups, revolts, and natural deaths, while controlling for the effect of other

covariates that may plausibly affect these outcomes. We use annual GDP per capita,

economic growth and fuel exports in order to control for economic factors that may effect

leader tenure. We also control for the type of dictatorship (civilian, military, monarchy),

30

for levels of cultural fractionalization (linguistic, ethnic, religious), whether the leader

presided over a communist country (communist leaders tend to stay in office longer on

average), and for the age of the leader. All time-varying covariates are lagged by one year.

The results of this (competing risks) survival analysis are summarized in Table 4.

Consider first exits due to natural causes, which we use as a benchmark to which other

types of leader exit can be compared. If our data are reliable, then the key factor

associated with exits due to natural causes should be the age of the leader, which is in fact

the case. But note that the coefficients for growth and fuel exports are on the margin of

the 10% significance level, indicating that there may be a spurious association between

these variables and leader exits due to natural causes.

Now consider the effect of the existence of a legislature on the likelihood of coups, the

empirical counterpart to allies’ rebellions in our theoretical model. Consistent with our

model, the existence of a legislature has a large positive and statistically significant effect

on the survival of leaders: at the median level of the remaining covariates, the existence of

a legislature reduces the hazard of a coup about eightfold. This result is robust to the

exclusion of any controls that reduce the size of our sample (see columns labeled “partial”)

and to alternative parameterizations of the hazard (loglogistic, lognormal, generalized

gamma). Furthermore, coefficient estimates on the remaining covariates are very sensible:

leaders in civilian and religiously homogenous dictatorships are less likely to be removed in

a coup than those in military and religiously fractionalized dictatorships, whereas the risk

of a coup grows with the age of a dictator.18 Surprisingly, we do not observe any

statistically significant relationship between the level of economic development, economic

18A likelihood ratio test indicates that the baseline category of monarchy can be excluded from theestimation without a loss of fit. Thus it is the difference between military and civilian dictatorships that isstatistically significant.

31

growth, or fuel exports and the likelihood of coups. Finally, our estimates imply that the

existence of legislatures also reduces the risk of revolts. However, we should be cautious

when interpreting these results as we observe only 25 revolts when using the partial set of

covariates and 8 revolts when using the full set of covariates.

To summarize, our empirical analysis offers preliminary support for two central

implications of our theoretical model. When we use export concentration as a proxy for the

balance of power between the ruler and his allies, we find that single-export dictatorships

are less likely to establish or maintain legislatures. Similar results obtain when we use the

end of the Cold War as a measure of withdrawal of foreign support and the corresponding

shift in the internal balance of power away from the dictator. These findings support our

theoretical claim that dictatorships benefit from establishing political institutions, but only

conditional on a permissive balance of power within the ruling coalition. We also find

support for two implications of our theoretical model that concern leader tenures: we show

that leaders in dictatorships without legislatures are more likely to be removed in a coup or

a revolt and less likely to leave due to natural causes than leaders in dictatorships with

legislatures. Furthermore, leaders in dictatorships with legislatures stay in office longer

than leaders in dictatorships without legislatures, even after controlling for a set of other

factors that may affect leader tenure. These finding supports our theoretical claim that the

existence of institutions stabilizes ruling coalitions in dictatorships.

5 Conclusion

From a historical point of view democracy has always constituted a very exceptional form of

government. Until the last hundred years, republican polities were confined to a few cities

in the classical world and in medieval and modern Europe - and even then their democratic

32

institutions were of the most imperfect sort. A back-of-the-envelope calculation shows that,

since it emerged about 100,000 years ago, close to ninety nine percent of mankind has been

governed by authoritarian rulers - tyrants, monarchs, princes and warlords of all venues.

In this paper, we depart from the existing literature on nondemocratic regimes in two

important ways. First, most of the literature on dictatorships mistakenly treats the rule of

a tyrant as the standard type of autocratic government. We examine data that indicate

that this view is at best unrepresentative of modern dictatorships. While quantitative data

on institutions in pre-modern dictatorships are lacking, Lo’s (1969) depiction of Court

Conferences under the Ming suggests that same may be true about pre-modern

dictatorships:

Even. . . when absolutism was at one of its heights in China. . . Emperors might

hint and threaten and court favorites might coerce and intimidate, but Court

Conferences persisted as the focus in which important issues were resolved.

And when rulers or their favorites planned unpopular courses of action,

officialdom found it possible to block them by collective action (p. 70).

Second, the study of dictatorships is still wedded to a sociological approach committed

to the construction and description of ideal types, that is, of types developed on the basis

of the high frequency of certain particular traits. By contrast, we start with the

observation that any successful dictatorship must resolve two political conflicts: The

conflict between those in power and those excluded from power, but equally importantly,

also the conflict among those in power.

We therefore investigate how the distribution of power within an authoritarian polity

affects the possibility for power-sharing. We argue that a fundamental problem of

authoritarian rule is the need to share power in an environment where any defection must

33

be backed by a credible threat of violence. This is because, in a dictatorship, any

power-sharing agreement is the very foundation of political authority. Moreover, the only

punishment available the dictator’s allies is to replace him with a challenger. We study

how the credibility of that threat shapes the possibilities for power-sharing and find that

the crude nature of this threat creates strong incentives to establish institutions that would

alleviate moral hazard problems in authoritarian governance. Yet even these institutions

survive only as long as any shifts in the balance of power between the dictator and his

allies do not undermine the credibility of the allies’ threat to replace the dictator.

References

Arendt, Hannah. 1973. The Origins of Totalitarianism. Harvest Books.

Banks, Arthur S. 2001. “Cross National Time Series Data Archive.”

Boix, Carles and Sebastian Rosato. 2001. “A complete dataset of political regimes, 1800-

1999.”

Brownlee, Jason. 2007. Authoritarianism in an Age of Democratization. New York: Cam-

bridge University Press.

Carlsson, Hans and Eric van Damme. 1993. “Global Games and Equilibrium Selection.”

Econometrica 61:989–1018.

Chwe, Michael. 2001. Rational ritual: Culture, coordination, and common knowledge. Prince-

ton, NJ: Princeton University Press.

Dewan, Torun and David P. Myatt. 2007. “Leading the Party: Coordination, Direction, and

Communication.” American Political Science Review 101(4):827–845.

Friedman, Thomas L. 2006. “The First Law of Petropolitics.” Foreign Policy (154):28–36.

Friedrich, Karl and Zbigniew Brzezinski. 1965. Totalitarian Dictatorship and Autocracy.

34

Cambridge, MA: Harvard University Press.

Gandhi, Jennifer and Adam Przeworski. 2006. “Cooperation, Cooptation, and Rebellion

Under Dictatorships.” Economics & Politics 18(1):1–26.

Gandhi, Jennifer and Adam Przeworski. 2007. “Authoritarian Institutions and the Survival

of Autocrats.” Comparative Political Studies 40(11):1279–1301.

Haber, Stephen H. 2007. “Authoritarian Regimes.” In Weingast, Barry R. and Donald

Wittman (Eds.), “The Oxford Handbook of Political Economy,” Oxford University Press.

Karl, Terry Lynn. 1997. The Paradox of Plenty: Oil Booms and Petro-States. Berkeley:

University of California Press.

Keefer, Philip. 2002. DPI2000: Database of Political Institutions. Washington, DC: Devel-

opment Research Group, The World Bank.

Kuran, Timur. 1991. “Now Out of Never: The Element of Surprise in the East European

Revolution of 1989.” World Politics 44:7–48.

Levitsky, Steven and Lucan A. Way. 2003. “Autocracy by Democratic Rules: The Dynamics

of Competitive Autoritarianism in the Post-Cold War Era.” Unpublished manuscript.

Linz, Juan. 1975. “Totalitarian and Authoritarian Regimes.” In Greenstein, Fred and Nelson

Polsby (Eds.), “Handbook of Political Science,” vol. 3. Reading, Mass.: Addison Wesley

Publishing Company, pp. 191–357.

Linz, Juan J. 2000. Totalitarian and Authoritarian Regimes. Boulder: Lynne Rienner.

Lo, Jung-pang. 1969. “Policy formulation and decision-making.” In Hucker, Charles O.

(Ed.), “Chinese government in Ming times,” pp. 41–72.

Lust-Okar, Ellen. 2006. “Elections under authoritarianism: Preliminary lessons from Jor-

dan.” Democratization 13(3):456–71.

MacFarquhar, Roderick. 1997. The Politics of China: The Eras of Mao and Deng. New

35

York: Cambridge University Press.

Machiavelli, Nicolo. 1513/1985. The Prince. Chicago: The University of Chicago Press.

Magaloni, Beatriz. 2006. Voting for autocracy: Hegemonic party surival and its demise in

Mexico. New York: Cambridge University Press.

Medina, Luis Fernando. 2007. A Unified Theory of Collective Action and Social Change.

Ann Arbor: University of Michigan Press.

Bueno de Mesquita, Bruce, Alastair Smith, Randolph M. Siverson, and James D. Morrow.

2003. The Logic of Political Survival. Cambridge: MIT Press.

Morris, Stephen and Hyun Song Shin. 2003. Advances in Economics and Econometrics,

chap. Global Games: Theory and Applications. Cambridge University Press, pp. 56–114.

Myerson, Roger B. 2008. “The Autocrat’s Credibility Problem and Foundations of the

Constitutional State.” American Political Science Review 102(1):125–139.

Neumann, Franz. 1957. The Democratic and Authoritarian State. New York: Free Press.

North, Douglass C. and Barry R. Weingast. 1989. “Constitutions and Commitment: The

Evolution of Institutions Governing Public Choice in Seventeenth-Century England.”

Journal of Economic History 49(4):803–832.

Przeworski, Adam, Michael E. Alvarez, Jose Antonio Cheibub, and Fernando Limongi. 2000.

Democracy and Development: Political Institutions and Well-Being in the World, 1950-

1990. New York: Cambridge University Press.

Smith, Benjamin. 2005. “Life of the Party: The Origins of Regime Breakdown and Persis-

tence Under Single-Party Rule.” World Politics 57(3):421–51.

Suny, Ronald Grigor. 1998. The Soviet experiment: Russia, the USSR, and the successor

states. New York: Oxford University Press.

Svolik, Milan and Seden Akcinaroglu. 2006. “Government Change in Authoritarian

36

Regimes.”

Tullock, Gordon. 1987. Autocracy. Boston: Kluwer Academic Publishers.

Wintrobe, Ronald. 1998. The political economy of dictatorship. Cambridge: Cambridge

University Press.

Wright, Joseph. 2008. “Do Authoritarian Institutions Constrain? How Legislatures Impact

Economic Growth and Foreign Aid Effectiveness.” American Journal of Political Science

52(2).

Xenophon/Strauss. 1961. On Tyranny. Chicago: The University of Chicago Press.

37

Table 1: Covariates of legislatures in dictatorships

Model 1 Model 2 Model 3 Model 4

β β + α β β + α β β + α β β + α

Legislaturet−1 -1.123∗ -0.163∗∗∗ 0.683 1.869 -1.107 0.184 0.684 1.652(0.091) (0.231) (0.530) (0.154) (0.190) (0.408) (0.593) (0.316)

Log of GDP p.c.t−1 -0.059 0.221∗∗∗ -0.179∗ 0.026 -0.045 0.159∗ -0.172 0.013(0.351) (0.002) (0.056) (0.154) (0.591) (0.090) (0.129) (0.313)

Log of populationt−1 0.036 0.036 -0.020 -0.021 0.043 0.094∗∗ -0.015 0.052(0.417) (0.374) (0.729) (0.866) (0.346) (0.023) (0.803) (0.655)

Ethnic fract. 0.010 0.082 0.077 0.402 -0.060 -0.209 0.023 0.081(0.962) (0.876) (0.793) (0.194) (0.800) (0.527) (0.941) (0.951)

Oil exports -0.136 -0.280 0.170 0.238 -0.176 -0.029 0.137 0.382(0.372) (0.089) (0.432) (0.430) (0.309) (0.586) (0.575) (0.290)

Post Cold War 0.304∗ 0.286 0.330∗ 0.368 0.283 0.241∗ 0.274 0.439∗∗(0.054) (0.025) (0.065) (0.023) (0.108) (0.095) (0.174) (0.041)

Export conc. -1.034∗∗∗ -0.851 -1.046∗∗ -0.411∗∗(0.009) (0.003) (0.015) (0.034)

Type of entry:

Independence 0.074 0.067 0.063 –(0.880) (0.977) (0.931) –

Civil war -0.372 -0.734∗∗∗ -0.451 -1.388∗∗∗(0.203) (0.005) (0.326) (0.006)

Coup -0.214 -0.860∗∗∗ -0.129 -0.841∗∗∗(0.341) (0.000) (0.678) (0.006)

Revolt 0.266 -0.573 7.842∗∗∗ -1.154∗∗∗(0.689) (0.460) (0.000) (0.000)

Foreign int. 0.270 -0.096 0.092 -0.466(0.455) (0.717) (0.888) (0.681)

Elections 0.462 0.351∗ 0.765∗ 0.049(0.140) (0.077) (0.080) (0.213)

Succession -0.157 -0.444 0.013 -0.788∗∗(0.592) (0.111) (0.973) (0.037)

Consensus 0.104 0.248 0.028 -0.104(0.740) (0.599) (0.944) (0.951)

Observations 3688 2036 3479 1905

Pseudo R2 0.64 0.66 0.66 0.67

Log-likelihood -765.10 -424.66 -669.64 -409.94

Note: Standard errors in parentheses. Significance levels ∗10%, ∗∗5%, ∗∗∗1%.

Tab

le2:

The

end

ofth

eC

old

War

and

the

intr

oduct

ion

ofle

gisl

ature

sin

dic

tato

rship

s

Leg

isla

ture

alre

ady

exis

ts

Con

centr

atio

nin

dex

:0

0.2

0.4

0.6

0.8

1

1950-1

990

0.97

90.

969

0.95

60.

937

0.91

30.

883

1991-1

999

0.99

20.

987

0.98

10.

971

0.95

80.

940

No

legi

slat

ure

yet

Con

centr

atio

nin

dex

:0

0.2

0.4

0.6

0.8

1

1950-1

990

0.23

00.

172

0.12

40.

087

0.05

90.

038

1991-1

999

0.34

10.

269

0.20

50.

152

0.10

80.

074

Sour

ce:

Sim

ulat

ion

base

don

Tab

le1.

All

othe

rva

riab

leat

med

ian

valu

es.

Tab

le3:

Lea

der

exit

indic

tato

rship

sw

ith

and

wit

hou

tle

gisl

ature

s,19

45-2

001a

coup

revolt

fore

ign

transi

tion

ele

ctio

nsb

natu

ral

oth

erc

tota

l

Legis

latu

re68

(19.

71)

11(3

.19)

6(1

.74)

28(8

.12)

52(1

5.07

)50

(14.

49)

139

(37.

68)

345

(100

.00)

No

legis

latu

re11

4(5

8.46

)16

(8.2

1)7

(3.5

9)1

(0.5

1)1

(0.5

1)14

(7.1

8)42

(21.

54)

195

(100

.00)

Sour

ces:

Ban

ks(2

001)

,Sv

olik

and

Akc

inar

oglu

(200

6).

aR

elat

ive

freq

uenc

ies

(per

cent

ages

)in

para

nthe

ses.

bIn

clud

esex

its

due

tote

rmlim

its

and

case

sw

hen

incu

mbe

nts

did

not

cont

est

elec

tion

s.cIn

clud

esex

its

due

toas

sass

inat

ions

,ci

vilw

ars,

inte

rim

term

s,an

dex

its

that

did

not

fitan

yof

the

abov

eca

tego

ries

.

Table 4: A survival analysis of authoritarian leader tenures

Coups Revolts Natural Causes

Partial Full Partial Full Partial Full

GDP per capita 0.025 -0.010 0.029 0.109 -0.023 -0.003(0.025) (0.065) (0.046) (0.124) (0.016) (0.078)

Growth 0.000 -0.007 0.030∗ 0.036 -0.004 -0.044∗(0.010) (0.015) (0.016) (0.027) (0.010) (0.026)

Fuel exports -0.002 -0.003 -0.009(0.004) (0.008) (0.005)

Legislature 1.678∗∗∗ 1.889∗∗∗ 1.153∗∗∗ 1.481∗∗∗ 0.205 0.477(0.166) (0.290) (0.268) (0.536) (0.178) (0.416)

Civilian 1.305∗∗ -0.463 0.151(0.600) (0.730) (0.666)

Military 0.118 0.769 0.037(0.444) (0.883) (0.623)

Ethnic -0.510 0.268 -0.284(0.553) (0.882) (0.824)

Language 0.643 0.323 0.053(0.505) (0.820) (0.805)

Religion -0.871∗ 0.293 0.066(0.511) (0.824) (0.693)

Communist 0.054 -0.132 0.135(0.449) (0.582) (0.523)

Age -0.028∗∗ -0.016 -0.043∗∗(0.012) (0.020) (0.018)

Intercept 2.200∗∗∗ 3.621∗∗∗ 3.453∗∗∗ 3.473∗∗∗ 3.545∗∗∗ 6.312∗∗∗(0.114) (0.726) (0.254) (1.149) (0.189) (1.352)

Shape parameter αa 1.022 1.116 1.549∗∗∗ 2.133∗∗∗ 1.746∗∗∗ 1.340∗(0.059) (0.095) (0.202) (0.560) (0.149) (0.168)

Leaders 574 280 574 280 574 280

Exits 168 87 25 8 61 26

Time at risk 4267 1795 4267 1795 4267 1795

Note: Standard errors in parentheses. Significance levels ∗10%, ∗∗5%, ∗∗∗1%.aWeibull parameterization, hazard increasing for α > 1, constant for α = 1, and decreasing for α < 1.

Fig

ure

1:P

ropor

tion

ofdic

tato

rship

sw

ith

legi

slat

ure

and/o

rpar

ties

,19

51-1

999.

Figure 2: A model of an authoritarian polity

Ally i

ρ ≤ ρ∗ ρ > ρ∗

Support b 0Rebel θ − r θ

Figure 3: Payoffs to ally i

Fig

ure

4:T

he

tim

ing

ofac

tion

sin

per

iods

t≥

1

0.00

0.05

0.10

0.15

0.20

0.25

0.30

DΛ

0.5

0.6

0.7

0.8

0.9

1.0∆

Fig

ure

5:Pow

er-s

har

ing

and

chan

ges

inth

edis

trib

uti

onof

pow

er