Dominant party rule, elections, and cabinetinstability in African autocracies

Alex M. KroegerLecturer of Political Science

University of California, Mercedakroeger2@ucmerced.edu

Abstract

I draw on the authoritarian institutions literature to explain the role of dominantparties in constraining the ability of autocrats to reshuffle cabinet ministers. Domi-nant party leaders are constrained in their ability to frequently reshuffle ministers bythe need to maintain credible power-sharing commitments with party elites. Theseconstraints also produce distinct temporal patterns of instability where large reshuf-fles occur following elections. Conversely, personalist leaders face fewer power-sharingconstraints and engage in more extensive cabinet reshuffles at more arbitrary intervals.Military leaders face complex constraints that depend on whether officers or civiliansoccupy cabinet posts and the extent to which leaders are dependent upon civilian min-isters for regime performance and popular support. Empirical analyses using data onthe cabinets of 94 authoritarian leaders from 37 African countries between 1976 and2010 support the theoretical expectations for dominant party and personalist leaders;but are inconclusive for military leaders.

Keywords: cabinet, dictatorship, party, regime, Africa

Word count: 11,998

African countries have a reputation for having the largest and most unstable cabinets in

the world. The size and instability of African cabinets has left some scholars to question

whether African ministers serve anything more than a ceremonial role, where ministers cap-

ture patronage but do not play a meaningful role as managers of the bureaucracy (van de

Walle, 2001). Recent scholarship enhances our understanding of why Africa’s cabinets have

grown over time as well as the economic impact of large cabinets (Arriola, 2009; LeVan &

Assenov, 2016; van de Walle, 2001). However, little research systematically analyzes the

instability of African cabinets.1 This is surprising given the number of anecdotal accounts

of cabinet reshuffles being used to undermine elites, the potential for cabinet reshuffles to

spark conflict between elites and leaders, and the detrimental impact of cabinet instability

on bureaucratic capacity. Furthermore, variation in cabinet stability, particularly among

Africa’s authoritarian regimes, is greater than is often recognized. This variation cannot

be explained using the unifying framework of patronage politics that guides most studies of

African autocracies.

Building on the literature on authoritarian institutions, I explain how regime type influ-

ences the ability of leaders to dismiss ministers without destabilizing their regimes. I argue

that leaders of Africa’s dominant party autocracies are more constrained in their ability to

dismiss cabinet ministers than are personalist leaders. These constraints reduce the extent of

minister dismissals by dominant party leaders as dismissals signal that years of loyal service

to the party will not be rewarded with meaningful power-sharing. The need to maintain

credible power-sharing with cabinet ministers also results in more distinct temporal patterns

of cabinet reshuffles by dominant party leaders compared to personalist leaders. Dominant

party leaders use elections as a mechanism to induce regularity in minister turnover, allowing

them to appoint a new group of party elites to cabinet positions while minimizing claims that

they are arbitrarily undermining the authority of ministers who have been dismissed. Con-

versely, personalist leaders are able to engage in more arbitrary patterns of cabinet shuffles

in both election and non-election years. Additionally, I argue that the extent of dismissals by

1

military leaders depends upon whether cabinets are composed of officers or civilians, and, if

the cabinet is composed of civilians, the extent to which the regime is dependent upon civil-

ians for regime performance and popular support. Military leaders with military cabinets

and those who are dependent upon civilian ministers for regime performance and popular

support should behave similarly to dominant party leaders.

I test my argument using an original dataset of cabinet ministers from 37 African au-

tocracies between 1976 and 2010. These data include yearly cabinet observations from 94

leaders of regimes classified as authoritarian by Geddes, Wright, and Frantz (2014). I find

that cabinet stability varies significantly across authoritarian regime types, with dominant

party leaders dismissing fewer cabinet ministers than personalist leaders. Dominant party

leaders also tend to engage in large cabinet reshuffles following elections while personalist

leaders engage in more arbitrary patterns of reshuffles in election and non-election years. The

results for military leaders are ambiguous for minister dismissals; however, military leaders

engage in more post-election horizontal reshuffles than personalist leaders.

This article makes several contributions. First, it qualifies existing accounts of elite

politics, patronage, and political stability in Africa. While I agree with Bratton and van de

Walle (1997) who argue that elite shuffles are used to “regulate and control rent-seeking,

to prevent rivals from developing their own bases, and to demonstrate power” (p. 86), the

ability of leaders to safely engage in reshuffles varies. This has implications for debates over

patronage politics and political stability, and helps to explain why, as Arriola (2009) notes,

patronage politics is frequently cited as a cause of both stability and instability. Although

Africa’s dominant party, personalist, and military leaders rely on patronage politics to build

and maintain coalitions, differences in regime type condition the extent of political instability

within the cabinet. This shows that formal institutions influence the behavior of leaders and

produce meaningful differences in political stability in African autocracies.

Second, I provide new empirical support for theories stressing the institutional roots of

authoritarian regime and leader stability. It is now well established that autocrats use in-

2

stitutions to co-opt elites and establish credible power-sharing commitments (Blaydes, 2010;

Boix & Svolik, 2013; Brownlee, 2007; Gandhi, 2008; Geddes, 2003; Magaloni, 2006, 2008;

Smith, 2005; Svolik, 2012). However, studies linking the presence of particular authoritarian

institutions to increased leader and regime tenure often make untested assumptions about

differences in coalition dynamics between regime elites and leaders. This study helps to

close the gap between the broad support for institutional theories of authoritarian regime

and leader stability and the coalition dynamics posited to explain differences in regime and

leader tenure. Additionally, it adds to existing research on authoritarian elections and elite

shuffles (e.g., Hassan, 2017; Magaloni, 2006, 2008) by highlighting the heterogeneous effects

of elections on cabinet reshuffles across regime types.

Third, this study advances the growing literature on authoritarian cabinets. Although

theories of cabinet durability in parliamentary and presidential democracies have been sub-

ject to significant cross-national evaluation (see Fischer, Dowding, & Dumont, 2012), rela-

tively few scholars have studied authoritarian cabinets cross-nationally. Furthermore, exist-

ing cross-national studies have not considered cabinet instability in the form of dismissals and

horizontal reshuffles (e.g., Arriola, 2009), how cabinet instability varies within autocracies

(e.g., Quiroz Flores, 2009), or whether authoritarian regime type influences leader decisions

to dismiss ministers (e.g., Francois, Rainer, & Trebbi, 2015a; Roberts, 2015). This study

provides a general theoretical framework through which studies of authoritarian cabinets can

be further specified and subject to greater empirical scrutiny.

The article proceeds as follows. First, I review explanations of cabinet stability found in

the literature on patronage politics and discuss how the broader literature on authoritarian

rule applies to cabinet stability in African autocracies. Second, I build the authoritarian

institutions literature to develop hypotheses on the extent and timing of minister dismissals

and horizontal reshuffles. Third, I introduce the cabinet data, the empirical model, present

the results, and discuss several robustness checks. Finally, I explain the implications of the

empirical analyses and offer suggestions for future research.

3

What explains cabinet instability?

Scholars have identified several factors that influence the stability of cabinets in authoritar-

ian regimes. First, particularly in African autocracies, cabinet instability is often attributed

to the practice of patronage politics. Patronage politics, which is often considered the or-

ganizing principle of African politics, is characterized by reciprocal, although necessarily

unequal, relations between leaders and their coalition of clients (Bratton & van de Walle,

1997; Clapham, 1982). The inequality within patronage arrangements arises from the leader’s

ability to condition continued access to state resources on political support. In resource poor

environments, this allows leaders to form broad coalitions of supporters, often spanning mul-

tiple ethnic groups (Arriola, 2009; Francois, Rainer, & Trebbi, 2015b; Lemarchand & Legg,

1972; Herbst, 2000).

As some of the most senior clients of leaders, cabinet ministers often reap substantial

benefits from their positions. Cabinet appointments come with many perks including high

salaries, cars, homes, policy influence, and opportunities for enrichment through corruption.

Access to patronage resources means that ministers can also develop their own patronage

networks (Barkan & Chege, 1989; Bratton & van de Walle, 1997). In his study of Houphouët-

Boiny’s Democratic Party of Côte d’Ivoire (PDCI), Zolberg (1969) argues that “regardless

of his specific duties as a member of the executive, each minister is also a kind of superrep-

resentative who keeps in touch with the country through his clientele of deputies” (p. 283).

Therefore, ministers, while accumulating patronage, are important for broadening the base

of regime support because of their ties to local and regional constituencies.

Ministerial appointments produce a paradox for leaders in patronage-based polities.

Leaders need the support of ministers and the constituencies they mobilize, which provides

incentives for cabinet expansion (Arriola, 2009). At the same time, cabinet appointments

provide elites with patronage resources that can be used to challenge the leader (Roessler,

2011). As Widner (1992) argues, “only a head of state who is exceptionally clever in his abil-

ity to elevate and demote the “barons” with who he allies himself–or keep them guessing–can

4

long maintain power” (pp. 55-6). In this view, cabinet reshuffles are a strategy of political

survival because they prevent ministers from building independent bases of power within

particular ministries that could be used to challenge the leader (Barkey, 1994; Bayart, 2009;

Chabal & Daloz, 1999; Jackson & Rosberg, 1982; Migdal, 1988; van de Walle, 2001).

Minister reshuffles, while undermining potential challengers, are not without costs. Lead-

ers that dismiss ministers and expel them from the ruling coalition risk elite splits and open

conflict. Roessler (2011) theorizes that sacking threatening regime elites, particularly those

of rival ethnic groups, exchanges coup risk for a future risk of civil war. For example, South

Sudan’s Salva Kiir sacked his entire cabinet in July 2013 following internal power struggles

within the South Sudan People’s Liberation Movement, particularly with Prime Minister

Riek Machar. Kiir’s decision sparked civil war between his government and those loyal to

Machar. Attempts to undermine regime elites and personalize power can also produce coup

attempts as remaining regime elites face an increasingly uncertain future (Svolik, 2012; Wig

& Rod, 2016). Recently dismissed ministers, such as former Liberian Minister of Rural De-

velopment Samuel Dokie, have coordinated with remaining regime elites to oust leaders. In

1983, Dokie organized a coup attempt against President Samuel Doe after reportedly being

dismissed for denying a request to transfer $3 million from the Ministry of Rural Develop-

ment to Doe’s personal account (Huband, 1998, p. 32). Mobutu unraveled a similar plot

before it materialized in 1966. In what was known as the “Whitsun plot,” recently dismissed

Premier Evariste Kimba and 3 other ministers were accused of plotting against Mobutu and

were publicly hanged (Dickie & Rake, 1973, p. 576). This is not to say that all ministers

possess the capability to mobilize civil wars or coups against incumbents. Particularly with

coups, mobilization against the incumbent typically requires discontent among members of

the military. Nevertheless, the cases from South Sudan, Liberia, and Zaire demonstrate

that some ministers are capable of mobilizing such actions. Furthermore, as Francois et al.

(2015a) state, military mobilization against the incumbent rarely takes place “without the

complicity of important civilian insiders like ministers” (pp. 2-3). Thus, the decision to

5

dismiss potential rivals from the cabinet non-trivial.2

Leader tenure provides a second explanation for cabinet instability. Leaders that have

survived longer in office may be less vulnerable to violent reprisals when undermining regime

elites. Scholarship on leader survival shows that authoritarian leaders face high initial risks of

being overthrown but become more secure in office as their tenure increases (Bienen & van de

Walle, 1989; Bueno de Mesquita, Smith, Siverson, & Morrow, 2003). Similarly, Svolik (2012)

argues that established autocrats who have survived the initial high risk period of their tenure

have fundamentally different relationships with regime elites than contested autocrats who

have recently entered office. Contested autocrats must share power with regime elites as they

are vulnerable to “allies rebellions.” Established autocrats have consolidated their authority

and can no longer be threatened by allies rebellions. As a result, “key administrators or

military commanders are periodically purged, publicly humiliated, rotated across posts, or

dismissed and later reappointed” (Svolik, 2012, p. 79). Cabinet reshuffles may, therefore, be

a function of leader tenure.

Third, cabinet instability may be explained by the vulnerability of particular authoritar-

ian regime types to elite splits. Leaders of regimes that are more vulnerable to elite splits

may limit cabinet change to prevent threatening rifts from emerging within their regimes.3

Geddes (2003) argues that the propensity of elite splits varies across dominant party, person-

alist, and military regimes. Dominant party regimes are defined as those where the ruling

party has “some influence over policy, control[s] most access to political power and govern-

ment jobs, and ha[s] functioning local-level organizations” (Geddes, 2003, p. 72). Examples

of African regimes coded as dominant party autocracies include the Socialist Party of Senegal

(PS) regime led by Leopold Senghor and Abdou Diouf and the Zimbabwe African National

Union-Patriotic Front (ZANU-PF) regime led by Robert Mugabe. Personalist regimes are

those where an individual leader controls personnel appointments and policy. The regimes of

Blaise Compaoré in Burkina Faso and Yahya Jammeh in Gambia are coded as personalist.4

Military regimes are those where a group of military officers influences policy decisions and

6

controls access to political power. For example, Geddes codes Justin Lekhanya’s regime in

Lesotho and Ibrahim Babangida’s regime in Nigeria as military.

Geddes expects the risk of elite splits to be low in dominant party and personalist regimes

and high in military regimes. Dominant party regimes, despite relying on broad coalitions

that often include rival factions, are able to minimize elite splits through their dominance

over the political system. Rival factions are discouraged from challenging the dominant

faction because doing so risks regime breakdown and thus the benefits from holding office in

a system monopolized by a single party (Geddes, 2003, p. 63). Similarly, Brownlee (2007)

argues that dominant party regimes “harness elites together” by reassuring “power holders

that their immediate and long-term interests are best served by remaining within the party

organization” (p. 39). Conflict between rival factions is thus prevented by mutual interests

in the continuation of the regime.

Elites splits are discouraged in personalist regimes by the leader’s control over patronage

and the security apparatus, as well as the exclusion of rival factions. As Geddes (2003)

argues, “As long as the dictatorship is able to supply some benefits and has a sufficiently

competent repressive apparatus to keep the probability of successful plotting reasonably

low, they [regime elites] will remain loyal” (p. 63). Similarly, Bueno de Mesquita et al.

(2003) argue that authoritarian regimes with small winning coalitions, which include many

personalist regimes described by Geddes, encourage loyalty between regime elites and the

leader. Although elites in small winning coalition systems have access to substantial private

resources, they are discouraged from using those resources to challenge the incumbent be-

cause the likelihood of successfully challenging the leader and being included in a subsequent

small coalition system is small.

Conversely, Geddes argues that military regimes are vulnerable to elite splits that emerge

over personal rivalries or policy differences. When rival factions emerge within military

regimes, soft-line factions within the regime will voluntarily exit politics and return to the

barracks in an attempt to preserve military unity (Geddes, 2003). Since returning to the

7

barracks allows the majority of officers to continue their military careers and experience

improved post-tenure fates (Debs, 2016; Geddes, 2003), other factions follow the first move of

soft-line factions and return to the barracks as well. The vulnerability to elite splits, interests

in maintaining unity within the military, and improved post-tenure fates for voluntarily

returning to the barracks thus explains the short tenure of military regimes (Debs, 2016;

Geddes, 2003; Kim & Kroeger, 2017).

On its own, Geddes’s discussion of elite splits suggests that both personalist and domi-

nant party leaders should face lower risks of elite splits after cabinet reshuffles than military

leaders. However, Geddes (2003, pp. 53, 60) only mentions that personalist leaders en-

gage in frequent elite shuffles. Personalist leaders are expected to utilize elite shuffles to

undermine challenges from rival factions and capture greater rents for the majority faction.

Conversely, elite shuffles in dominant party regimes are mentioned only tangentially. For in-

stance, Geddes’s (2003) game theoretic account of politics in dominant party regimes shows

that excluding rival factions is risky “because exclusion gives the minority an incentive to

try to unseat the majority” (p. 59). While Geddes’s theoretical logic may be sound, its

application to the stability of authoritarian cabinets leaves an additional puzzle. The dis-

tinction between personalist and dominant party regimes by Geddes (2003) and Geddes et

al. (2014) is mainly based upon whether the leader or the party controls policy and po-

litical appointments. However, leaders of Africa’s dominant party and personalist regimes

retain formal control over cabinet appointments and dismissals. Even in Botswana, where

the Botswana Democratic Party (BDP) is highly institutionalized, the president has the

authority to appoint and dismiss cabinet ministers. In other highly institutionalized domi-

nant party regimes such as Tanzania’s Chama Cha Mapinduzi (CCM) regime, the president

makes cabinet appointments after consultation with the prime minister. Nevertheless, the

Tanzanian president can revoke appointments without the approval of CCM leaders.

The lack of formal control over cabinet appointments and dismissals along with Geddes’s

expectation that personalist and dominant party regimes are relatively resistant to elite

8

splits produces ambiguity over the expected patterns of cabinet shuffles in Africa’s dominant

party regimes. This contrasts with military regimes, where Geddes’s logic suggests that

threatening rivalries should emerge when leaders frequently shuffle officers out of cabinet

positions. In the next section, I build on studies of authoritarian institutions to explain the

relative constraints faced by dominant party, personalist, and military leaders when shuffling

ministers and how these constraints influence patterns of minister dismissals and horizontal

reshuffles.

Regime Type, elections, and cabinet instability

My explanation of cabinet stability in African autocracies focuses on the different power-

sharing dynamics between leaders and cabinet ministers in dominant party, personalist,

and military regimes. Rather than expecting cabinets in African autocracies to be similarly

unstable as in the literature on patronage politics, or that the relative resistance to elite splits

determines the extent of elite shuffling, I explain why dominant party leaders face greater

constraints on their ability to shuffle cabinet ministers than personalist leaders, and provide

several countervailing expectations for the constraints faced by Africa’s military leaders.

Furthermore, I expect more distinct temporal patterns in cabinet reshuffles in dominant

party regimes, with leaders shuffling ministers following elections to establish regular patterns

of cabinet turnover that reward new elites and serve as a bulwark against claims of failed

power-sharing.

Scholars of authoritarian regimes have begun to stress not only the importance of power-

sharing institutions, but also the credibility of power-sharing commitments between leaders

and elites (Magaloni, 2008; Svolik, 2012). Credible power-sharing commitments are most

important for dominant party leaders as their rule depends on their ability to co-opt broad

coalitions of elites. Elites, particularly those from rival factions, have few reasons to acquiesce

to co-optation into a regime that does not offer them increased patronage opportunities

and policy influence. Magaloni (2008) argues that parties solve this commitment problem

9

when they are expected to persist into the future and can “(a) control access to power

positions, spoils, and privileges; and (b) deliver on the promise to promote those who join

the organization” (pp. 723-4). Additionally, both Magaloni and Svolik emphasize that

credible power-sharing requires that elites can credibly threaten rebellion against leaders

who renege on power-sharing commitments.

Rather than focusing on power-sharing between leaders and elites broadly as do Magaloni

and Svolik, I focus specifically on the relationships between leaders and cabinet ministers.

As with Geddes’s definition of dominant party regimes, Magaloni (2008) suggests that the

party must control ministerial appointments and dismissals for power-sharing to be credible

within the cabinet. While this is not the case in the African regimes studied here, I argue

that dominant parties still constrain the leader’s ability to arbitrarily dismiss ministers. To

make power-sharing credible, dominant party regimes must establish a system of regular pro-

motion in exchange for party service. Party members must also be rewarded with meaningful

increases in patronage and policy influence as they rise through the ranks. Without a regular

system of promotion and increasing benefits, current and prospective members of the party

have few incentives to incur the “sunk political costs” at the lower levels of the party (Svolik,

2012, p. 163). The promise of promotion, policy influence, and patronage is particularly

important at the higher levels of the party as these individuals have invested the most time

and effort in party service. Few party elites can expect to be selected as the next leader, but

those that have risen through the party ranks are likely to expect cabinet appointments, or

other prestigious positions in parastatals or party committees, for their years of loyal service.

Because of the importance and prestige associated with cabinet appointments, leaders that

rapidly dismiss ministers send a public signal that political promotions at the highest level

do not provide meaningful power-sharing. This undermines the party at the upper and lower

levels. High ranking members realize that they can no longer expect meaningful political

promotion while lower ranking members, who often engage in the most costly party service,

begin to question whether their investment in the party will be rewarded. Thus, party mem-

10

bership begins to lose its appeal, creating splits among party factions and leading some to

challenge the incumbent by supporting rebellion or forming their own political parties.

Conversely, personalist leaders face lower risks of compromising their regimes by rapidly

shuffling ministers. Unlike dominant party regimes, personalist regimes do not depend upon

broad coalitions and are less reliant on the service of lower ranking regime members. Person-

alist leaders, having often gained their positions through intense struggles (Geddes, 2003),

implement a variety of coup-proofing measures to prevent successful rebellion against their

rule, including the recruitment and promotion of soldiers based on family, ethnic, or reli-

gious ties (Quinlivan, 1999). For example, most Togolese officers serving under Gnassingbe

Eyadema were coethnics of the Kabye ethnic group and many senior commanders were

from Eyadema’s home village (Decalo, 1989; Pilster & Böhmelt, 2011). Coup-proofing can

also involve counterbalancing military forces by creating rival factions or paramilitary units

(Quinlivan, 1999). Mobutu engaged in counterbalancing in 1974 when he replaced the uni-

fied General Staff with four separate departments under different leadership (Emizet, 2000).

Similarly, Blaise Compaoré established a separate Regiment of Presidential Security (RSP),

which served as an “army within an army” that was better equipped than the regular armed

forces (Chouli, 2015, p. 327).5 Furthermore, most elites will refrain from challenging per-

sonalist leaders so long as the leader maintains control over patronage distribution (Bratton

& van de Walle, 1997; Ulfelder, 2005). This allows personalist leaders to rapidly dismiss

ministers without facing an organized response.

The constraints on Africa’s military leaders are less clear. As Geddes describes, military

regimes are reliant upon power-sharing between the leader and a group of officers. While

this suggests that military regime cabinets should be relatively stable compared to those in

personalist regimes, it is not clear that this logic holds among Africa’s military regimes. This

is because the cabinets of Africa’s military regimes are often quickly “civilianized” (Anene,

1997), with military officers serving in the cabinet being replaced by civilian ministers.

Anene (1997, p. 63) argues that civilian cabinet appointments serve to improve government

11

effectiveness, address demands for civilian rule, and build bases of support separate from

the military.6 This “civilianization” process produces cabinet instability in the early years of

military regimes as civilians replace military officers. However, it is unclear whether military

leaders face the same power-sharing constraints with civilian ministers as they do with high-

ranking officers. For instance, civilian ministers are less able to punish military leaders who

engage in frequent cabinet reshuffles than are military ministers, which could result in more

extensive and frequent cabinet reshuffles. At the same time, Anene’s (1997) argument that

civilian ministers improve government effectiveness, accommodate demands for civilian rule,

and build support outside the military suggests that military leaders may be dependent upon

power-sharing with civilian ministers as well, resulting in fewer cabinet reshuffles than in

personalist regimes. Although existing theory is ambiguous about the stability of cabinets

in Africa’s military regimes, empirical analyses can help to explain the extent to which

Africa’s military leaders are constrained by power-sharing commitments with ministers. If

the lesser ability of civilian ministers to punish military leaders is most important, patterns

of cabinet reshuffles by Africa’s military leaders should be similar to those of personalist

leaders. Alternatively, if Africa’s military leaders are dependent upon civilian ministers for

regime performance and popular support, they should engage in patterns of cabinet reshuffles

that are more consistent with dominant party leaders.

The discussion above produces several testable hypotheses. First, despite arguments that

dominant party regimes face a low risk of elite splits, I argue that dominant party leaders

dismiss fewer ministers than personalist leaders. Given the theoretical ambiguity over the

reshuffling behavior of Africa’s military leaders, I seek only to test whether they behave

similarly to either dominant party and personalist leaders, following the countervailing logics

described above. These arguments produce Hypothesis 1:

Hypothesis 1: Dominant party leaders dismiss fewer ministers than personalist

and military leaders.

While Hypothesis 1 examines differences in rates of minister dismissals across author-

12

itarian regime types,7 it does not predict the extent to which leaders horizontally shuffle

ministers between ministerial portfolios. This distinction is important as dismissing min-

isters often provides a different signal to regime elites than does horizontally reshuffling

ministers. Horizontal reshuffles provide a middle ground between dismissing problematic

ministers and allowing them to continue developing power and autonomy within a particular

portfolio, and provide a mechanism to reduce ministerial moral hazard. (Huber & Martinez-

Gallardo, 2008; Indridason & Kam, 2008). I expect all authoritarian leaders to use horizontal

reshuffles to reduce ministerial moral hazard and that overall rates of horizontal reshuffles

are similar across regime types. While the relative lack of power-sharing constraints faced

by personalist leaders may be expected to result in more horizontal reshuffles, I argue that

this simply allows personalist leaders to engage in more dismissals instead, with horizontal

reshuffles reserved to check the power and autonomy of close allies who remain in the cabinet

for longer periods of time. This leads to Hypothesis 2.

Hypothesis 2: Dominant party, military, and personalist leaders engage in

horizontal reshuffles at similar overall rates.

Finally, maintaining credible power-sharing with ministers in dominant party regimes

should produce more distinct temporal patterns of dismissals and horizontal reshuffles. I

expect dominant party leaders to engage in major cabinet reshuffles predominantly following

elections. Magaloni (2006) argues that elections in dominant party autocracies establish

regular mechanisms of power-sharing among party elites. They also encourage party unity

by providing public signals of regime strength, provide the leader with information on regime

supporters and opponents, and encourage the opposition to operate within the existing set

of institutions rather than seeking change through violent means (Magaloni, 2006, pp. 8–9).

Regularity in cabinet turnover is important for the credibility of power-sharing as it mitigates

claims that the leader is arbitrarily dismissing ministers to undermine their authority. By

dismissing and horizontally reshuffling ministers at regular intervals, dominant party leaders

establish norms of turnover that allow them to reward a new group of party elites with cabinet

13

appointments while also minimizing impressions that power-sharing commitments have been

breached. Furthermore, in party regimes where ministers must be selected from members of

the legislature, seats can be lost in free or manipulated elections, partially reducing the need

for overt dismissals.

Many of Africa’s personalist regimes also hold elections, but the lesser importance of

credible power-sharing results in more arbitrary patterns of minister reshuffles in both elec-

tion and non-election years. As in Hypothesis 1, the extent to which military leaders increase

dismissals and horizontal reshuffles following elections depends upon the power-sharing con-

straints they face vis-á-vis civilian ministers. If military leaders are dependent upon power-

sharing with civilian ministers for the performance of and support for the regime, they will

engage in larger numbers of dismissals and horizontal reshuffles following elections. However,

if military leaders are less dependent upon civilian ministers, we will see relatively smaller

increases in dismissals and horizontal reshuffles following elections. This leads to Hypothesis

3.

Hypothesis 3: Dominant party leaders increase dismissals and horizontal reshuf-

fles to a greater extent following elections than personalist or military leaders.

Data and methods

To test these hypotheses, I collected data on the composition of cabinets for 94 leaders

in 37 African countries from 1976 to 2010 using volumes of Africa South of the Sahara

(Europa Publications Limited, 1975–2010).8 The unit of analysis is the leader-year, with

leaders identified using the Archigos data (Goemans, Gleditsch, & Chiozza, 2009).9 For each

leader-year in the sample, I code the names of cabinet ministers and the portfolios they

held. This includes all individuals listed as cabinet members by Africa South of the Sahara,

with the exception of the leader.10 Taking the leader-year as the unit of analysis means that

cabinet changes occurring because of leadership changes are excluded from analyses. Cabinet

14

instability caused by leader turnover is consequential for governance, particularly given the

prevalence of coups in Africa, but it does not speak to the constraining role of dominant

party institutions. Therefore, the hypotheses are directed only at dismissals and horizontal

reshuffles that occur within the tenure of leaders. A complete list of leaders, leader-years,

and regime type codings is included in the Supplemental Appendix.

Dependent variables

The first dependent variable measures the number of ministers who leave the cabinet in each

leader-year. The coding of this variable does not distinguish between minister dismissals,

deaths in office, or voluntary resignations because of limited data on the specific conditions

surrounding minister exits. While this only provides a proxy of minister dismissals, there are

four primary reasons to believe that this coding does not unduly bias the results in favor of

the hypotheses outlined above. First, minister deaths are likely to be randomly distributed

across dominant party, personalist, and military regimes. Second, there are few reasons to

expect more voluntary resignations in either dominant party, military, or personalist regimes.

Personalist regimes with small winning coalitions provide ministers with strong incentives

to remain loyal to the leader (Bueno de Mesquita et al., 2003), even though personalist

leaders need not remain loyal to their ministers. In dominant party regimes, the dominant

strategy of ministers, including those from rival factions, is to remain in office (Geddes,

2003). While the theoretical logic is more ambiguous for ministers in military regimes, there

are few reasons to expect higher rates of voluntary resignations compared with dominant

party or personalist regimes. Third, voluntary resignations often occur when ministers defect

to the opposition and seek to challenge candidates from the incumbent regime in legislative

or executive elections. This requires multiparty competition that is not limited to a single

authoritarian regime type. Fourth, voluntary resignations often occur because ministers are

unsatisfied with the level of power-sharing or to preempt dismissals or worse from the leader.

Dickie and Rake (1973) describe such a situation near the beginning of Hastings Banda’s

15

tenure in Malawi, stating that “His determination to go his own way cost him dearly. He lost

much needed talent when seven ministers went into exile after charges of plotting” (Dickie

& Rake, 1973, p. 259). This suggests that many voluntary resignations are the outcome of

failed power-sharing that my theory seeks to highlight. Therefore, the number of ministers

exiting the cabinet each year represents a reasonable proxy of minister dismissals.

The second dependent variable measures the number of ministers shuffled horizontally to

new ministerial portfolios each year. For example, I code Zimbabwe’s Emmerson Mnangagwa

as being horizontally reshuffled in 2009 when he left his previous post as Minister of Rural

Housing and Amenities to be reappointed as Minister of Defense. Horizontal reshuffles are

expected to reduce ministerial moral hazard without severely breaching power-sharing com-

mitments. Additional information on the coding of both minister dismissals and horizontal

reshuffles is provided in the Supplemental Appendix.

Independent variables

The main model specifications contain two variables of interest. The first is authoritarian

regime type. Following the coding of Geddes et al. (2014), I include dummy variables for

dominant party regimes and military regimes, leaving personalist regimes as the reference

category. In addition to dominant party, military, and personalist regimes, Geddes, Wright,

and Frantz also code hybrids of these pure types. I follow their convention of coding hybrid

regimes as the first regime type listed in the main analyses. Thus, party-personalist or party-

military hybrids are coded as dominant party regimes. However, I also estimate additional

models disaggregating hybrid regimes which are discussed below.

The second independent variable of interest is an indicator of whether or not an executive

or legislative election was held before the cabinet measurement in year t. This requires

special attention as the month in which cabinet composition is recorded by Africa South of

the Sahara varies by country, leader, and year. The variable Election takes a value of one if

there was an executive election, legislative election, or both after the cabinet measurement

16

at time t−1 but before the cabinet measurement at time t. Data on elections come from the

NELDA dataset (Hyde & Marinov, 2012). Elections are expected to induce cabinet changes

in authoritarian regimes just as they do in democracies, even if they are driven by different

motivations. However, increases in dismissals and horizontal reshuffles following elections

are expected to be greater in dominant party regimes where power-sharing commitments

discourage arbitrary reshuffles in non-election years. Elections occur in 20% of dominant

party leader-years, 20% of personalist leader-years, and 11% of military leader-years in the

sample.

Controls

I control for a variety of factors that may also influence cabinet instability. First, a leader’s

ability to co-opt clients is influenced by access to resource rents (Arriola, 2009; Smith, 2005).

Leaders with access to resource rents can more easily buy the support of political opponents,

perhaps reducing the need for credible power-sharing commitments and producing greater

cabinet instability. I control for access to resource rents using data on total resource rents

as a percentage of GDP from the World Bank’s World Development Indicators.11 Because

of missing data, the variable Resource Rents is coded 1 if the median of available data on

total resource rents for a particular country is greater than the median total resource rents

for available observations across all countries and 0 for all other leader-years. Similarly, I

also control for Ln(GDP per Capita) and GDP Growth as additional measures of the resource

constraints faced by leaders.

Second, cabinet instability is common following failed coup attempts as leaders seek to

purge coup supporters and critics from the regime. Cameroon’s Paul Biya, while initially

retaining most of former President Ahmadou Ahidjo’s cabinet, shuffled ministers extensively

after uncovering a coup plot in 1983 hatched by Ahidjo, Maj. Ibrahim Oumharou, and Capt.

Ahmadou Saleh and again after an actual coup attempt by Ahidjo and Saleh materialized

in 1984. The dismissals targeted Ahidjo allies, particularly ministers from Ahidjo’s base in

17

Cameroon’s North and Extreme North provinces. In total, Biya dismissed 10 ministers from

the North and Extreme North provinces following the failed coup attempt. The variable

Coup Attempt represents an indicator of whether a failed coup attempt took place after the

cabinet measurement at time t − 1 but before the cabinet measurement at time t, and is

coded using data from Powell and Thyne (2011).

Third, scholars have stressed the importance of ethnic balancing within African cabi-

nets. As Francois et al. (2015b) show, cabinet appointments are allocated proportionately

to ethnic group size in African states. For example, Kenneth Kaunda carefully balanced his

cabinet by appointing individuals roughly in proportion to their group’s share of the Zam-

bian population (Posner, 2005).12 However, this ethnic balancing is argued to be a source

of instability. Ethnically diverse cabinets may increase cabinet instability by producing dis-

trust between leaders and ministers from rival ethnic groups. Roessler (2011) argues that

reciprocal maneuvering of leaders and ethnic elites, both trying to consolidate their positions

within the regime, undermines power-sharing commitments and results in rival ethnic elites

being excluded from the regime. Additionally, ethnic diversity provides a greater number of

possible minimal winning coalitions, giving leaders the ability to marginalize certain groups

and form an alternative coalition (Posner, 2005). I include the variable Senior Partners

from Version 3 of the Ethnic Power Relations dataset (Wimmer, Cederman, & Min, 2009)

to measure the extent of ethnic power-sharing at high levels within the regime. This vari-

able captures the extent of formal or informal power sharing in senior positions within the

executive and is measured as the percentage of the population represented by coethnics in

senior positions.

Fourth, I account for the effect of leader tenure on cabinet instability. Scholars of par-

liamentary democracies theorize that leaders face adverse selection problems when making

cabinet appointments (Huber & Martinez-Gallardo, 2008). After imperfect appointments

are made, leaders observe the true abilities of their appointees and shuffle them to portfolios

that better suit their skill set or remove them from the cabinet entirely. Although authoritar-

18

ian leaders are likely to have different minister selection criteria (see Quiroz Flores & Smith,

2011), selection problems may still produce instability early in the tenure of leaders. Also,

the hazard of losing office declines over time for authoritarian leaders (Bueno de Mesquita

et al., 2003; Bienen & van de Walle, 1989; Svolik, 2012), which can produce more frequent

reshuffles early in the tenure of leaders as they seek to prevent ministers from capitalizing on

the initial weakness of their grasp on power. I control for duration dependence in a flexible

way by including a cubic polynomial of leader tenure in all models (Carter & Signorino,

2010).13

Finally, because the dependent variable does not distinguish between dismissals and

voluntary resignations, I control for the presence of multiparty elections in the legislature

using the Lexical Index of Electoral Democracy (LIED) dataset (Skaaning, Gerring, & Bar-

tusevičius, 2015). This variable addresses the possibility that ministers are more willing

to voluntarily exit the cabinet when they can legally challenge the incumbent through an

opposition party.

Estimation method

Since both dependent variables represent discrete numbers of ministers dismissed or hori-

zontally reshuffled each year, count models provide an appropriate estimation framework. I

choose the more flexible negative binomial distribution over the Poisson as both dependent

variables show signs of overdispersion. The natural logarithm of cabinet size is included as

an offset variable in all models to account for differing levels of exposure to dismissals and

horizontal reshuffles across leader-years.14 For instance, 41 ministers faced the possibility of

dismissal or horizontal reshuffle in Laurent Gbagbo’s 2006 cabinet in Côte d’Ivoire, but only

14 ministers faced these actions in Moussa Traore’s 1980 cabinet in Mali.

Additionally, the panel structure of the data presents the potential for leader-specific ef-

fects, non-constant variance across leaders, and autocorrelation within the tenure of leaders.

Fixed effects are commonly used to control for unit-specific effects in panel data settings;

19

however, the lack of variation in the regime type variables within the tenure of individual

leaders prevents their use. Non-constant error variance and autocorrelation are addressed us-

ing standard errors clustered by leader. Further robustness checks allowing for leader-specific

random effects and modeling first order autocorrelation directly through the generalized es-

timating equations (GEE) framework are discussed below and presented in Tables A3 and

A4 in the Supplemental Appendix. These alternative modeling choices do not alter the main

findings.

Results

Table 1 reports the results of the negative binomial regressions. Models 1 and 2 estimate

the effects of regime type and elections on minister dismissals. In Model 1, the effects of

the regime type and election variables are modeled independently. As expected, the Party

Regime coefficient is negative and significant, indicating that dominant party leaders dismiss

fewer ministers than personalist leaders. The Military Regime coefficient is negative but

only significant at the 10% level. Additionally, an F-test of the equality of the Party Regime

and Military Regime coefficients reveals no significant difference (F = 1.09, p = 0.30).

Therefore, dominant party leaders dismiss fewer ministers than personalist leaders, but the

results for military leaders are ambiguous. Finally, the Election coefficient is positive and

significant, indicating that minister dismissals increase following elections.

Model 2 tests Hypothesis 3 by interacting the Party Regime and Military Regime

variables with Election to determine whether post-election increases in dismissals differ

across regime types. The Party Regime coefficient remains negative and significant, showing

that dominant party leaders dismiss fewer ministers in non-election years than personalist

leaders. The Military Regime coefficient is negative but not significant at any conventional

level, suggesting no differences in non-election year shuffles by military and personalist regime

leaders. The positive and significant coefficient for Election shows that personalist leaders

20

Table 1: Minister dismissals and horizontal reshuffles

Dependent variable:Dismissals Horizontal Moves

(1) (2) (3) (4)Party Regime −0.392∗∗∗ −0.483∗∗∗ −0.188 −0.262

(0.099) (0.112) (0.149) (0.172)Military Regime −0.232∗ −0.234 0.024 −0.044

(0.139) (0.145) (0.135) (0.144)Election 0.538∗∗∗ 0.340∗∗∗ 0.555∗∗∗ 0.363∗∗∗

(0.072) (0.083) (0.083) (0.102)Party X Election 0.430∗∗∗ 0.339∗∗

(0.133) (0.169)Military X Election −0.163 0.415∗∗

(0.193) (0.163)Multiparty −0.219∗∗∗ −0.219∗∗∗ −0.248∗∗ −0.248∗∗

(0.079) (0.081) (0.102) (0.102)Resource Rents 0.053 0.053 −0.048 −0.048

(0.078) (0.079) (0.120) (0.120)Ln(GDP per Capita) −0.113∗∗ −0.118∗∗ 0.130∗ 0.125∗

(0.056) (0.058) (0.069) (0.068)GDP Growth −0.015 −0.024 −0.417 −0.412

(0.270) (0.276) (0.282) (0.283)Coup Attempt 0.306∗ 0.292∗ 0.174 0.191

(0.156) (0.153) (0.174) (0.176)Senior Partners −0.332∗∗ −0.351∗∗ 0.647∗∗∗ 0.646∗∗∗

(0.152) (0.157) (0.195) (0.195)Constant −0.517 −0.473 −2.956∗∗∗ −2.903∗∗∗

(0.347) (0.357) (0.480) (0.476)Leader Tenure Polynomial? Yes Yes Yes YesObservations 947 947 947 947Log Likelihood −2,631 −2,627 −2,115 −2,113θ 1.294∗∗∗ 1.310∗∗∗ 1.295∗∗∗ 1.306∗∗∗

Akaike Inf. Crit. 5,287 5,283 4,257 4,256

Note: Pooled negative binomial regressions with standard errors clustered byleader. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

21

increase dismissals following elections. Consistent with Hypothesis 3, however, the Party

X Election coefficient is positive and significant, which demonstrates that dominant party

leaders increase dismissals following elections to a greater extent than personalist leaders.

The Military X Election coefficient is negative but not significant at any conventional

level, indicating that post-election dismissal increases by military and personalist leaders

are similar. As in Model 1, F-tests fail to reject the hypothesis that dominant party and

military regimes have the same effect in non-election years (F = 2.35, p = 0.13) or that

elections have the same effect in dominant party and military regimes (F = 2.39, p = 0.12).

Taken together, these results show that the hypothesized differences between dominant party

and personalist leaders are supported, but the differences between military leaders and either

dominant party or personalist leaders are ambiguous.

Models 3 and 4 in Table 1 estimate the effects of regime type and elections on horizon-

tal reshuffles. The coefficients for Party Regime and Military Regime are not significant

in Model 3. This provides support for Hypothesis 2 which predicts similar overall rates

of horizontal reshuffles by dominant party, personalist, and military leaders. As expected,

the Election coefficient is positive and significant for horizontal shuffles. Model 4 interacts

Party Regime and Military Regime with Election to test whether horizontal reshuffles

differ across regime types in election and non-election years. The Election coefficient, now

representing the effect of elections in personalist regimes, is reduced in magnitude, but re-

mains positive and significant. The coefficients for the Party X Election and Military

X Election terms are positive and significant, showing that the post-election increases in

horizontal reshuffles of dominant party and military leaders are greater than those of per-

sonalist leaders. However, an F-test shows that the effects of elections in dominant party

regimes and military regimes do not differ from each other (F = 0.16, p = 0.69). Thus, while

the dismissal behavior of military leaders is ambiguous, the horizontal reshuffles of military

leaders resemble those of dominant party leaders.

Since the substantive effects of the negative binomial coefficients are difficult to interpret,

22

●●

●

●

●

●

Military Party Personal

No Yes No Yes No Yes

0123456789

1011

Election Year?

Dis

mis

sals

●

●

●

●●

●

Military Party Personal

No Yes No Yes No Yes

0123456789

1011

Election Year?

Hor

izon

tal M

oves

Figure 1: Predicted dismissals and horizontal reshuffles with 95% confidence intervals.Predictions calculated using Models 2 and 4 in Table 1 with all variables set to their respectiveregime type means with the exception of Multiparty and Resource Rents which are set to1 and Coup Attempt which is set to 0 for all regime type categories.

I present the results graphically in two ways. First, Figure 1 plots the predicted number of

minister dismissals and horizontal reshuffles for military, party, and personalist leaders during

election and non-election years. The predictions are calculated holding the control variables

at their sample means for the particular regime type, with the exception of Multiparty

which takes a value of 1 for all regime categories and Coup Attempt and Resource Rents

which take a value of 0 for all regime categories. The plot of predicted dismissals in the

top panel of Figure 1 follows the same general patterns shown in Table 1. Dominant party

leaders are predicted to dismiss 3.2 ministers in non-election years compared to 6.3 ministers

for personalist leaders. In election years, dominant party leaders are predicted to dismiss

7 ministers compared to 8.8 ministers for personalist leaders. Therefore, dominant party

23

leaders clearly have more stable cabinets in non-election years and tend to refrain from

larger numbers of dismissals until after elections. Moreover, as Table 1 demonstrates, the

differences between military regimes and either party or personalist regimes are less clear in

Figure 1. Military leaders are predicted to dismiss 4.4 ministers in non-election years and 5.3

ministers in election years, but the wide confidence intervals for the predictions make them

statistically indistinguishable from the predictions for either party or personalist leaders.15

●

●

●

●

●

●

●

●

●

Party Regime

Military Regime

Election

Multiparty

Coup Attempt

Senior Partners

High Rents

Ln(GDP per Capita)

GDP Growth

−4 −3 −2 −1 0 1 2 3 4

(a) Dismissals

●

●

●

●

●

●

●

●

●

Party Regime

Military Regime

Election

Multiparty

Coup Attempt

Senior Partners

High Rents

Ln(GDP per Capita)

GDP Growth

−4 −3 −2 −1 0 1 2 3 4

(b) Horizontal moves

Figure 2: Average marginal effects with 95% confidence intervals calculated using Models2 and 4 from Table 1.

Second, I also present the average marginal effects of each variable in Figure 2 following

the recommendation of Hanmer and Kalkan (2013). Average marginal effects (AMEs) are

calculated by holding each additional covariate at its observed value rather than the values

chosen to calculate the predicted counts in Figure 1. This means that no AMEs are calculated

for the Party X Election or Military X Election interactions. Figure 2 shows that the

AME of Party Regime is negative and significant for dismissals. This provides additional

support for Hypothesis 1. The AME of Military Regime is also negative and significant for

dismissals, providing some support for the hypothesis that the dependence of military leaders

on civilian ministers for regime performance and popular support produces greater power-

sharing constraints than are found in personalist regimes.16 As the right panel of Figure 2

shows, the AMEs for Party Regime and Military Regime are both indistinguishable from

24

zero for horizontal reshuffles, consistent with Hypothesis 2.

Estimates for the control variables also demonstrate interesting dynamics, particularly

for the Multiparty and Senior Partners variables. The Multiparty coefficient is negative

and significant in all model specifications for dismissals and horizontal reshuffles, as are the

average marginal effects displayed in Figure 2. Multiparty competition, even in Africa’s

authoritarian regimes, has thus brought greater cabinet stability. The coefficient for Senior

Partners are negative and significant for dismissals and positive and significant for horizontal

reshuffles, as are the average marginal effects displayed in Figure 2. This shows that greater

ethnic diversity in the executive branch reduces the number of minister dismissals, but

increases the rate at which leaders horizontally reshuffle ministers. These results provide an

interesting extension of research by Roessler (2011) who finds that commitment problems

between African leaders and rival ethnic groups results in exclusion from the regime. While

this may be the case broadly, diversity among senior regime elites appears to restrict the

leader’s ability to dismiss ministers. However, following Roessler’s argument, mutual distrust

between the leader and rival ethnic elites within the cabinet may result in more horizontal

reshuffles as the leader attempts to maintain a diverse coalition while still attempting to

undermine the authority of non-coethnic elites.

Robustness checks

I estimate several additional models to examine the robustness of the results to a number

of different modeling and specification choices. One concern is that the estimates in Table

1 are being driven by unobserved leader-specific effects and not differences in constraints

across regime types. Although leader fixed effects cannot be estimated because regime type

does not vary across the tenure of leaders, I adopt three alternative approaches to assess the

impact of individual leaders on the main findings.

First, I split the data into dominant party, personalist, and military samples and estimate

the effect of elections in each sample while including leader fixed effects. If the estimates

25

in Table 1 are not biased by leader specific effects, the Election coefficient in models of

dismissals should be largest in the dominant party sample. As Table 2 shows, the Election

coefficient is approximately three times larger in the dominant party sample than in the

personalist sample. Similar to the results in Table 1, the Election coefficient for the military

regime sample in Model 3 is positive but is only significant at the 10% level. These findings

suggest that the estimates of post-election dismissals for dominant party, personalist, and

military leaders are not biased by leader specific effects. The split sample estimate for

horizontal reshuffles in Table 2 are also consistent with the findings in Table 1. The Election

coefficient is larger in the dominant party sample than the personalist sample, and is largest

in the military regime sample. Additionally, the split sample estimates reveal differences in

the effects of control variables not modeled in Table 1. This is particularly the case with

the Coup Attempt variable, which is positive and significant only for the personalist leaders.

This finding provides indirect evidence of the power-sharing constraints faced by dominant

party and military leaders.

Second, I estimate negative binomial models with leader random effects. The estimates,

presented in Table A3 of the Supplemental Appendix, remain very similar to those in Table

1.. Third, I examine the sensitivity of the findings of Models 2 and 4 in Table 1 to the

exclusion of single leaders and countries from the dataset. Figures A2 and A3 plot the

density of coefficient estimates and p-values obtained for the Party Regime and Party x

Election coefficients in Table 1 Model 2 when dropping each individual leader and country

from the dataset and reestimating the model. The results remain consistent across the

different subsets of leaders and countries.

Another concern given the panel structure of the data is autocorrelation. I address

this through the generalized estimating equations (GEE) framework (see Hilbe, 2011). The

estimates from GEE models with an AR(1) working correlation structure are presented

in Table A4 Supplemental Appendix. Coefficients for the regime type variables and their

interactions with elections remain very similar to those in Table 1.

26

Dependent variable:Dismissals Horizontal Moves

Party Personal Military Party Personal Military(1) (2) (3) (4) (5) (6)

Election 0.989∗∗∗ 0.334∗∗∗ 0.392∗ 0.777∗∗∗ 0.533∗∗∗ 1.090∗∗∗

(0.124) (0.095) (0.208) (0.127) (0.120) (0.245)Coup Attempt −0.169 0.483∗∗ 0.153 0.098 0.042 0.376

(0.345) (0.244) (0.570) (0.392) (0.190) (0.401)Senior Partners −0.325 0.623 −0.875 −0.192 −0.060 −0.665

(0.257) (0.614) (1.477) (0.291) (1.124) (1.150)Ln(GDP per Capita) −0.896∗ −0.431∗∗∗ −1.056 0.448∗∗ 0.678∗∗ 0.814

(0.468) (0.127) (1.144) (0.219) (0.327) (0.710)GDP Growth 0.430 0.311 −1.255 −0.519 −0.036 −0.525

(0.697) (0.215) (1.660) (0.394) (0.423) (1.255)Constant 4.966 1.568∗∗ 3.881 −5.726∗∗∗ −6.557∗∗∗ −77.059∗∗∗

(3.260) (0.793) (6.575) (1.569) (2.005) (7.693)Cubic Splines? Yes Yes Yes Yes Yes YesObservations 431 425 91 431 425 91Log Likelihood −1,048 −1,242 −236 −907 −915 −178θ 1.388∗∗∗ 2.064∗∗∗ 2.619∗∗∗ 1.568∗∗∗ 2.418∗∗∗ 10.071Akaike Inf. Crit. 2,195 2,572 522 1,913 1,917 406

Table 2: Unconditional fixed effects negative binomial regressions with standard errorsclustered by leader. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

I also examine the robustness of the regime type coding in two ways. First, the analyses

in Table 1 follow the approach of Geddes et al. (2014) by coding party–personalist or party–

military hybrids as dominant party regimes.17 Importantly, this decision should bias the

results against the hypotheses above since Geddes’s (2003, p. 73) coding gives less weight to

the party coding where: 1) party membership is highly urban with little grassroots organi-

zation; 2) the politburo serves as a rubber stamp for the leader and the leader plays a large

role in the selection of its members; 3) the government is dominated by one particular group

in heterogeneous societies; and 4) nepotism is common in high offices. As expected, the

estimates in Table A5 of the Supplemental Appendix show that the Pure Party and Pure

Party X Election coefficients increase in magnitude in the dismissal models after including

27

an additional hybrid party variable. However, the coefficients for Hybrid Party and Hybrid

Party X Election are not significant. This suggests that the organization, authority, and

make-up of dominant party regimes is critical for the ability of the party to constrain the

leader.18 Second, I recode the Party Regime variable to distinguish between leaders who

founded the party regime and those who became the leader of a preexisting party regime. If

the non-founding leaders of party regimes are similarly or more constrained than founding

party leaders, the theoretical argument above gains additional support. The estimates in

Table A10 show that both founding and non-founding party regime leaders follow the pat-

tern of Party Regime in Table 1; however, the coefficients for the non-founding party leader

and its interaction with elections are larger than those for founding party leaders. This also

eases concerns that the regime type coding is endogenous to cabinet reshuffles.

Next, the main findings may be driven by dominant party regimes being more democratic

than military or personalist regimes. For example, Botswana is coded as a dominant party

dictatorship, has relatively little cabinet instability, and is considered democratic by most

African politics scholars. While the estimates in Table 1 control for multiparty regimes, I

also control for Polity 2 scores (Marshall, Gurr, & Jaggers, 2013) in Table A6. Controlling

for the level of democracy produces coefficients that are nearly identical to those in Table

1. This increases confidence that the results are not merely due to presence of countries like

Botswana or Tanzania that score high on the Polity index, have relatively stable cabinets,

but are labeled as dominant party dictatorships by Geddes et al. (2014). Similarly, Roberts

(2015) shows, the presidential or parliamentary structure of authoritarian regimes is conse-

quential for their overall durability and cabinet stability. While most African autocracies in

the sample are presidential systems, countries such as Botswana and Ethiopia under Meles

Zenawi have parliamentary systems. Table A7 shows that parliamentary systems do reduce

both dismissals and horizontal reshuffles of cabinet ministers. However, estimates for the

variables of interest remain very similar.

Finally, I estimate models using a restricted sample of cabinet ministers that excludes

28

junior, assistant, and deputy ministers following the coding of Arriola (2009) and conduct

an extreme bounds analysis on the coefficients of Models 2 and 4 of Table 1. The estimates

presented in Table A8 show that the main findings are not driven by my more inclusive

definition of the cabinet. The extreme bounds analysis (Table A12 and Figure A4 shows

that the findings for the variables of interest are robust to alternative model specifications.

Conclusion

This article addresses gaps in the research on patronage politics and authoritarian institu-

tions to explain variation in cabinet instability between dominant party, personalist, and

military dictatorships in Africa. Although the literature on patronage politics thoroughly

discusses the motivations driving African autocrats to frequently reshuffle their cabinets,

it fails to explain the constraining role of dominant party regimes. The literature on au-

thoritarian institutions, while describing the ability of dominant party regimes to establish

credible power-sharing between the leader and party elites, is reliant on studies of leader

and regime tenure for empirical support. This study breaks new ground by examining how

dominant party regimes constrain the ability of leaders to reshuffle cabinets, thus providing

a more direct look at elite power-sharing at the highest level.

I show that dominant party regimes influence the extent and timing of minister dismissals

in African dictatorships. Dominant party leaders dismiss far fewer ministers than do per-

sonalist leaders. Also, dominant party leaders tend to engage in large numbers of dismissals

and horizontal reshuffles only after elections to establish regular patterns of cabinet change

and increase the credibility of power-sharing. Conversely, personalist leaders face fewer con-

straints and engage in more arbitrary patterns of dismissals and horizontal reshuffles that

vary less between election and non-election years. Patterns of dismissals in Africa’s military

regimes exhibit greater variation and are indistinguishable from either personalist or domi-

nant party regimes. Military leaders do appear to engage in greater numbers of horizontal

29

reshuffles following elections, but this should be interpreted with caution as there are few

elections in military regimes.

These findings broadly support the literature on authoritarian power-sharing, but also

add important nuance. I demonstrate that dominant party leaders need not relinquish con-

trol over cabinet appointments and dismissals to the party for credible power-sharing to take

place. Instead, the structure of dominant party regimes constrains the ability of leaders to

shuffle ministers arbitrarily. I also explain that power-sharing dynamics in military regimes

become more complex as civilians are appointed to senior positions within the regime. Addi-

tionally, I show that explanations of elite shuffles rooted in the patronage politics literature

are primarily applicable to personalist regimes. While future work is needed to assess the

generalizability of these findings beyond sub Saharan Africa, the theory is sufficiently general

to be applied to other regions.

Finally, this research highlights two particularly important areas of future research. First,

more research is necessary to understand how regime type influences the risk of dismissal or

reappointment for individual ministers and how individual minister characteristics influence

such risks. Recent work has begun to examine the influence of gender on the tenure of

individual ministers (Arriola & Johnson, 2014; Escobar-Lemmon & Taylor-Robinson, 2014;

Krook & O’Brien, 2012), but more work is needed on the effects of ethnic identity and port-

folio prestige on minister tenure. For instance, further investigation of the divergent effects

of ethnic diversity in the executive on dismissals and horizontal reshuffles can provide an

important addition to Roessler’s (2011) work on ethnic exclusion in African states. Second,

future work can build on recent studies of party system development in Africa LeBas, 2011;

Riedl, 2014, e.g., to better understand the conditions under which ministers and other elites

defect to build opposition parties or join existing opposition parties.

30

Notes

1Francois et al. (2015b, 2015a) are exceptions.2Shuffling ministers is also costly in terms of bureaucratic capacity and efficiency. Fre-

quently shuffling ministers undermines ministers’ ability to implement government policies,

thus undermining state capacity (Besley & Perrson, 2010). Instability in office also discour-

ages the formation of ties and communication among bureaucrats, which is detrimental to

policy implementation (Rauch & Evans, 2000). This is not to say that autocrats desire an

effective bureaucracy. Quiroz Flores and Smith (2011) explain that autocrats prefer bu-

reaucrats that are loyal but mediocre in their performance. Nevertheless, rapidly shuffling

ministers creates a difficult environment for policy implementation.3Alternatively, elite splits can also spur cabinet instability by increasing voluntary resig-

nations.4The mere existence of a political party associated with the leader does not constitute a

dominant party regime according Geddes (2003). Compaoré, Jammeh, and other personalist

regime leaders establish political parties. However, the lack of party control over political

appointments and policy results in these regimes being classified as personalist.5The RSP helped Compaoré maintain power for nearly three decades, but were also

instrumental in his demise (see Chouli, 2015).6Geddes (2006) makes a similar argument about the development of political parties in

authoritarian regimes, arguing that they serve to reduce reliance on the military for survival.7Each empirical model includes ln(cabinet size) as an offset variable. The use of an offset

variable accounts for the fact that 10 dismissals in a 20 minister cabinet is different from 10

dismissals in a 40 minister cabinet.8Cases were selected based upon their coding as authoritarian by Geddes et al. (2014),

their inclusion in Arriola’s (2009) study of cabinet size, and their status as independent

countries by 1990. Geddes et al. (2014) code authoritarian regimes with a population of

greater than 1 million in 2009. The only exceptions to these rules are Somalia and Swaziland.

31

Data was not collected for Somalia because of its coding as a warlord regime from 1991–2010.

Data was not included for Swaziland because it is the only monarchical regime between 1976

and 2010.9Leaders listed in Archigos are cross-checked with those listed in Africa South of the

Sahara. In total, there are 947 leader-years in the dataset. The mean tenure of the leaders

is 11.9 years.10Arriola’s (2009) cabinet coding only includes ministers of full ministerial rank, excluding

assistant ministers, deputy ministers, junior ministers, and the like. Arriola bases this deci-

sion on the inconsistent reporting of such cabinet members across countries in Africa South

of the Sahara. Although this makes sense given Arriola’s focus on cabinet size, I find the

exclusion of lower ranking ministers to be unwarranted when studying minister dismissals

and horizontal reshuffles. Within the tenure of individual leaders, Africa South of the Sahara

provides consistent coverage of cabinets. When ministers of less than full ministerial rank are

listed, they appear consistently throughout the tenure of the particular leader. Furthermore,

all empirical analyses control for cabinet size and robustness checks conducted with only full

ministers (Table A8 of the Supplemental Appendix) produce the same substantive results.11This variable represents a combination of rents from forestry, minerals, and oil.12Despite this ethnic balancing, Posner (2005) notes that Bemba’s occupied a dispropor-

tionate share of prestigious cabinet portfolios such as Defense, Home Affairs, Foreign Affairs,

and Finance.13Models controlling for a cubic polynomial of regime tenure are presented in Table A11.

The findings remain the same.14The offset variable coefficient, ln(cabinet size), is assumed to be 1.15Confidence intervals are particularly wide for military leaders’ election year dismissals

since their are only 10 elections in military regimes in the dataset.16Importantly, the negative binomial coefficients and AMEs are testing different hypothe-

ses. The former test whether the effect of the independent variable differs from zero while

32

the latter test whether the average of all marginal effects for a variable when all covariates

are held at their observed values differs from zero. As Greene (2009) states, “Arguably, the

inference should be based about θk (the coefficient), not δk (the AME), since in the latter

case, one is testing a hypothesis about all the coefficients, not just the one of interest” (p.

487).17Of 431 leader-years coded dominant party in Table 1, 84 are party hybrids. For instance,

Gabon is coded as a party–personalist regime under the rule of Omar Bongo.18Alternatively, the small number of hybrid party regimes may explain the imprecise esti-

mates for Hybrid Party and Hybrid Party X Election coefficients. Models distinguishing

between party-personalist and party-military hybrids also yield insignificant coefficients.

33

References

Anene, J. (1997). Military administrative behavior and democratization: Civilian cabinet

appointments in military regimes in sub-Saharan Africa. Journal of Public Policy,

17 (1), 63–80.

Arriola, L. (2009). Patronage and political stability in Africa. Comparative Political Studies,

42 (10), 1339–1362.

Arriola, L., & Johnson, M. (2014). Ethnic politics and women’s empowerment in Africa:

Ministerial appointments to executive cabinets. American Journal of Political Science,

58 (2), 495–510.

Barkan, J., & Chege, M. (1989). Decentralizing the state: District focus and the politics of

reallocation in Kenya. Journal of Modern African Studies, 27 (3), 431–453.

Barkey, K. (1994). Bandits and bureaucrats: The Ottoman route to state centralization.

Ithica, NY: Cornell University Press.

Bayart, J.-F. (2009). The state in Africa: The politics of the belly (2nd ed.). Cambridge,

UK: Polity.

Besley, T., & Perrson, T. (2010). State capacity, conflict, and development. Econometrica,

78 (1), 1–34.

Bienen, H., & van de Walle, N. (1989). Time and power in Africa. American Political

Science Review, 83 (1), 19-34.

Blaydes, L. (2010). Elections and distributive politics in Egypt. Cambridge, UK: Cambridge

University Press.

Boix, C., & Svolik, M. (2013). The foundations of limited authoritarian government: Insti-

tutions, commitment, and power-sharing in dictatorships. Journal of Politics, 75 (2),

300–316.

Bratton, M., & van de Walle, N. (1997). Democratic experiments in Africa: Regime transi-

tions in comparative perspective. Cambridge, UK: Cambridge University Press.

Brownlee, J. (2007). Authoritarianism in an age of democratization. Cambridge: Cambridge

34

University Press.

Bueno de Mesquita, B., Smith, A., Siverson, R., & Morrow, J. (2003). The logic of political

survival. Cambridge, MA: MIT Press.

Carter, D., & Signorino, C. (2010). Back to the future: Modeling time dependence in binary

data. Political Analysis, 18 (3), 271–292.

Chabal, P., & Daloz, J.-P. (1999). Africa works: Disorder as a political instrument. Ox-

ford,UK: Currey.

Chouli, L. (2015). The popular uprising in Burkina Faso and the transition. Review of

African Political Economy.

Clapham, C. (1982). Private patronage and public power. In C. Clapham (Ed.), (p. 1-35).

New York: St. Martin’s Press.

Debs, A. (2016). Living by the Sword and Dying by the Sword? Leadership Transitions in

and out of Dictatorships. International Studies Quarterly, 60 (1), 73–84.

Decalo, S. (1989). Psychoses of power: African personal dictatorships. Boulder, CO: West-

view.

Dickie, J., & Rake, A. (1973). Who’s who in Africa: The political, military and business

leaders of Africa. London: African Buyer and Trader Publications Ltd.

Emizet, K. (2000). Explaining the rise and fall of military regimes: Civil-military relations

in the Congo. Armed Forces & Society, 26 (2), 203–227.

Escobar-Lemmon, M., & Taylor-Robinson, M. (2014). Women ministers in Latin American

government: When, where, and why? American Journal of Political Science, 49 (4),

829–844.

Europa Publications Limited. (1975–2010). Africa south of the Sahara. London, UK: Author.

Fischer, J., Dowding, K., & Dumont, P. (2012). The duration and durability of cabinet

ministers. International Political Science Review, 33 (5), 505–519.

Francois, P., Rainer, I., & Trebbi, F. (2015a). The dictator’s inner circle. (Working paper)

Francois, P., Rainer, I., & Trebbi, F. (2015b). How is power shared in Africa? Econometrica,

35

83 (2), 465–503.

Gandhi, J. (2008). Political institutions under dictatorship. Cambridge, UK: Cambridge

University Press.

Geddes, B. (2003). Paradigms and sandcastles. Ann Arbor: University of Michigan Press.

Geddes, B. (2006). Why parties and elections in authoritarian regimes. (Unpublished

manuscript)

Geddes, B., Wright, J., & Frantz, E. (2014). Autocratic breakdown and regime transitions:

A new data set. Perspectives on Politics, 12 (2), 313–331.

Goemans, H., Gleditsch, K., & Chiozza, G. (2009). Introducing Archigos: A dataset of

political leaders. Journal of Peace Research, 46 (2), 269-283.

Greene, W. (2009). Discrete choice modeling. In T. Mills & K. Patterson (Eds.), The

handbook of econometrics: Vol. 2, Applied econometrics. London: Palgrave.

Hanmer, M., & Kalkan, K. (2013). Behind the curve: Clarifying the best approach to

calculating predicted probabilities and marginal effects from limited dependent variable

models. American Journal of Political Science, 57 (1), 263–277.

Hassan, M. (2017). The strategic shuffle: Ethnic geography, the internal security apparatus,

and elections in Kenya. American Journal of Political Science, 61 (2), 382–395.

Herbst, J. (2000). States and power in Africa: Comparative lessons in authority and control.

Princeton, NJ: Princeton University Press.

Hilbe, J. (2011). Negative binomial regression. Cambridge, U.K.: Cambridge University

Press.

Huband, M. (1998). The Liberian Civil War. Frank Cass Publishers.

Huber, J., & Martinez-Gallardo, C. (2008). Replacing cabinet ministers: Patterns of min-

isterial instability in parliamentary democracies. American Political Science Review,

102 (2), 169–180.

Hyde, S., & Marinov, N. (2012). Which elections can be lost? Political Analysis, 20 (2),

191–201.

36

Indridason, I., & Kam, C. (2008). Cabinet reshuffles and ministerial drift. British Journal

of Political Science, 38 (4), 624-656.

Jackson, R., & Rosberg, C. (1982). Personal rule in Black Africa. Berkeley, CA: University

of California Press.

Kim, N. K., & Kroeger, A. M. (2017). Regime and leader instability under two forms of

military rule. Comparative Political Studies, 1–35. (OnlineFirst)

Krook, M. L., & O’Brien, D. Z. (2012). All the president’s men? The appointment of female

ministers worldwide. Journal of Politics, 74 (3), 840–855.

LeBas, A. (2011). From protests to parties: Party-building and democratization in Africa.

Oxford, U.K.: Oxford University Press.

Lemarchand, R., & Legg, K. (1972). Political clientelism and development: A preliminary

analysis. Comparative Politics, 4 (2), 149–178.

LeVan, C., & Assenov, A. (2016). Parties or portfolio? The economic consequences of

Africa’s big cabinets. Government and Opposition, 51 , 661–690.

Magaloni, B. (2006). Voting for autocracy: Hegemonic party survival and its demise in

Mexico. New York: Cambridge University Press.

Magaloni, B. (2008). Credible power sharing and the longevity of authoritarian rule. Com-

parative Political Studies, 41 (4–5), 715–741.

Marshall, M., Gurr, T., & Jaggers, K. (2013). Polity IV project: Political regime character-

istics and transitions, 1800-2012. Center for Systemic Peace.

Migdal, J. (1988). Strong societies and weak states: State–society relations and state capa-

bilities in the Third World. Princeton: Princeton University Press.

Pilster, U., & Böhmelt, T. (2011). Coup-proofing and military effectiveness in interstate

wars, 1967–99. Conflict Management and Peace Science, 28 (4), 1–20.

Posner, D. N. (2005). Institutions and ethnic politics in Africa. New York: Cambridge

University Press.

Powell, J., & Thyne, C. (2011). Global instances of coups from 1950 to 2010: A new dataset.

37

Journal of Peace Research, 48 (2), 249-259.

Quinlivan, J. (1999). Coup-proofing: Its practice and consequences in the Middle East.

International Security, 24 (2), 131-165.

Quiroz Flores, A. (2009). The political survival of foreign ministers. Foreign Policy Analysis,

5 (2), 117–133.

Quiroz Flores, A., & Smith, A. (2011). Leader survival and cabinet change. Economics and

Politics, 23 (3), 345–366.

Rauch, J. E., & Evans, P. B. (2000). Bureaucratic structure and bureaucratic performance

in less developed countries. Journal of Public Economics, 75 , 49–71.

Riedl, R. (2014). Authoritarian origins of democratic party systems in Africa. Cambridge

University Press.

Roberts, T. (2015). The durability of presidential and parliamentary based-dictatorships.

Comparative Political Studies, 48 (7), 915–948.

Roessler, P. (2011). The enemy within: Personal rule, coups, and civil war in Africa. World

Politics, 63 (2), 300–346.

Skaaning, S.-E., Gerring, J., & Bartusevičius, H. (2015). A lexical index of electoral democ-

racy. Comparative Political Studies, 48 (12), 1491–1525.

Smith, B. (2005). Life of the party: The origins of regime breakdown and persistence under

single-party rule. World Politics, 57 (3), 421–451.

Svolik, M. (2012). The politics of authoritarian rule. Cambridge, UK: Cambridge University

Press.

Ulfelder, J. (2005). Contentious collective action and the breakdown of authoritarian regimes.

International Political Science Review, 26 (3), 311–334.

van de Walle, N. (2001). African economies and the politics of permanent crisis, 1979–1999.

Cambridge, UK: Cambridge University Press.

Widner, J. A. (1992). The rise of a party-state in Kenya: From Harambee! to Nyayo!

Berkeley: University of California Press.

38

Wig, T., & Rod, E. G. (2016). Cues to coup plotters: Elections as coup triggers in dicta-

torships. Journal of Conflict Resolution, 60 (5), 787–812.

Wimmer, A., Cederman, L., & Min, B. (2009). Ethnic politics and armed conflict: A

configurational analysis of a new global dataset. American Sociological Review, 74 (2),

316–337.

Zolberg, A. (1969). One-party government in the Ivory Coast. Princeton, NJ: Princeton

University Press.

39

Supplemental Appendix for Dominant party rule, elections, andcabinet instability in African autocracies

(Not for publication)

This document contains additional information on data coding as well as additional

robustness checks that are not included in the main text.

Coding of cabinet instability

The data on cabinets is taken from yearly volumes of Europa Publications Limited (1975–

2010). This presents several challenges for empirical analyses. First, Europa records the

composition of cabinets once a year, often at irregular intervals. For instance, the composi-

tion of Kenya’s cabinet may be recorded in May during year t and in September during year

t+ 1. My own inspection of the data indicates that cabinet composition is usually measured

after events such as elections or coups. However, reshuffles can happen multiple times each

year. This is something that my coding of cabinet instability cannot capture.

Similarly, the irregular measuring of cabinet composition introduces assumptions into the

coding of dismissals and horizontal reshuffles. A minister is coded as having been dismissed

in year t if they were a member of the cabinet in year t−1 but are not recorded as part of the

cabinet in year t. This coding does not allow for the possibility that the minister actually

left the cabinet in the previous year, creating the potential for measurement error. The

same problem occurs with the coding of horizontal reshuffles. A minister is coded as being

horizontally reshuffled in year t if they are in charge of a different portfolio in year t than

they were in year t − 1. Therefore, the precise timing of minister dismissals and horizontal

reshuffles is not known.

The cabinet composition data also present challenges for the measurement of independent

variables representing discrete events such as elections and coup attempts. Fortunately, both

NELDA and Powell and Thyne (2011) code the dates that elections and coups take place.

This information is used when coding these variables to ensure that the month in which an

40

election or coup took place comes before, or is equal to, the month in which the cabinet was

recorded. For example, if the cabinet was recorded in March of 1995 and a coup attempt

took place in October of 1995, a coup attempt would be recorded in 1996. If the cabinet

was recorded in August 2000 and an election occurred in July of 2000, then 2000 would be

coded as an election year because the election came before the cabinet measurement.

Technical details and code files for the coding of the raw minister data are available from

the author upon request.

41

Table A1: Summary Statistics

Statistic N Mean St. Dev. Min MaxDismissals 947 6.099 6.519 0 48Dismissals Reduced 947 5.823 6.195 0 47Horizontal Shuffles 947 3.166 3.214 0 22Horizontal Reduced 947 3.045 3.072 0 18Promotions 947 1.031 1.270 0 8Demotions 947 0.901 1.135 0 8Party Regime 947 0.455 0.498 0 1Personal Regime 947 0.449 0.498 0 1Military Regime 947 0.096 0.295 0 1Election 947 0.194 0.396 0 1Coup Attempt 947 0.031 0.172 0 1Cabinet Size 947 23.845 8.036 9 66Senior Partners 947 0.169 0.250 0.000 0.940Ln(GDP per Capita) 947 6.750 0.853 4.984 9.330Resource Rents 947 0.548 0.498 0 1GDP Growth 947 0.049 0.112 −0.517 1.273Leader Tenure 947 11.917 8.130 1 42Polity 2 947 −3.765 4.589 −9 9Multiparty 947 0.436 0.496 0 1Parliamentary 947 0.061 0.240 0 1

42

Table A2: Sample of African autocrats

leader country cowcode start end regimeNeto Angola 540 1978 1979 party-basedDos Santos Angola 540 1981 2010 party-basedKerekou Benin 434 1976 1990 personalKhama Botswana 571 1976 1980 party-basedMasire Botswana 571 1982 1997 party-basedMogae Botswana 571 1999 2007 party-basedIan Khama Botswana 571 2010 2010 party-basedLamizana Burkina Faso 439 1977 1980 personalZerbo Burkina Faso 439 1982 1982 militarySankara Burkina Faso 439 1985 1987 personalCampaore Burkina Faso 439 1989 2010 personalMicombero Burundi 516 1976 1976 party-militaryBagaza Burundi 516 1978 1987 party-militaryBuyoya Burundi 516 1989 1992 militaryBuyoya Burundi 516 1997 2002 military-personalAhidjo Cameroon 471 1976 1982 party-personalBiya Cameroon 471 1984 2010 personalBokassa Central African Republic 482 1976 1979 personalDacko Central African Republic 482 1981 1981 personalKolingba Central African Republic 482 1983 1993 military-personalFrancois Bozize Central African Republic 482 2004 2010 personalMalloum Chad 483 1977 1978 militaryHabre Chad 483 1984 1990 personalDeby Chad 483 1992 2010 personalMobutu Congo-Kinshasa 490 1976 1996 personalLaurent Kabila Congo-Kinshasa 490 1998 2000 personalJoseph Kabila Congo-Kinshasa 490 2002 2010 personalNgouabi Congo-Brazzaville 484 1976 1976 party-militaryNguesso Congo-Brazzaville 484 1980 1991 party-militaryNguesso Congo-Brazzaville 484 1999 2010 personalMengistu Marriam Ethiopia 530 1978 1990 military-personalMeles Zenawi Ethiopia 530 1992 2010 party-basedBongo Gabon 481 1976 2009 party-personalJawara Gambia 420 1976 1993 party-basedJammeh Gambia 420 1995 2010 personalAcheampong Ghana 452 1976 1978 militaryRawlings Ghana 452 1983 2000 personalToure Guinea 438 1976 1983 party-basedConte Guinea 438 1985 2008 personalCabral Guinea-Bissau 404 1977 1980 party-basedVieira Guinea-Bissau 404 1982 1998 personalKumba Iala Guinea-Bissau 404 2003 2003 personal

43

Table A2: Sample of African autocrats

leader country cowcode start end regimeHouphouet-Boigny Ivory Coast 437 1976 1993 party-basedKonan Bedie Ivory Coast 437 1995 1999 party-basedLaurent Gbagbo Ivory Coast 437 2002 2010 personalKenyatta Kenya 501 1976 1978 party-basedMoi Kenya 501 1980 2002 party-basedJonathan Lesotho 570 1976 1985 party-basedLekhanya Lesotho 570 1987 1990 militaryRamaema Lesotho 570 1992 1992 militaryTolbert Liberia 450 1976 1979 party-personalDoe Liberia 450 1981 1990 personalTaylor Liberia 450 1998 2002 personalRatsiraka Madagascar 580 1977 1992 personalBanda Malawi 553 1976 1993 personalTraore Mali 432 1976 1990 personalOuld Daddah Mauritania 435 1976 1978 personalOuld Haidalla Mauritania 435 1981 1984 personalSidi Ahmed Taya Mauritania 435 1986 2004 personalOuld Mohamed Vall Mauritania 435 2006 2006 militaryMachel Mozambique 541 1978 1986 party-basedChissano Mozambique 541 1988 2004 party-basedGuebuza Mozambique 541 2006 2010 party-basedNujoma Namibia 565 1992 2004 party-basedPohamba Namibia 565 2006 2010 party-basedKountche Niger 436 1977 1987 military-personalSeibou Niger 436 1989 1991 military-personalMainassara Niger 436 1997 1998 personalObasanjo Nigeria 475 1977 1979 militaryBuhari Nigeria 475 1985 1985 militaryBabangida Nigeria 475 1987 1992 militaryAbacha Nigeria 475 1995 1997 military-personalHabyarimana Rwanda 517 1976 1993 military-personalPaul Kagame Rwanda 517 1995 2010 party-military

44

Table A2: Sample of African autocrats

leader country cowcode start end regimeSenghor Senegal 433 1976 1980 party-basedDiouf Senegal 433 1982 1999 party-basedStevens Sierra Leone 451 1976 1985 party-basedMomoh Sierra Leone 451 1987 1991 party-basedNimeiri Sudan 625 1976 1984 personalAl-Bashir Sudan 625 1990 2010 personalNyerere Tanzania 510 1976 1985 party-basedMwinyi Tanzania 510 1987 1995 party-basedMkapa Tanzania 510 1997 2005 party-basedKikwete Tanzania 510 2007 2010 party-basedEyadema Togo 461 1976 2004 personalFaure Gnassingbe Togo 461 2006 2010 personalAmin Uganda 500 1976 1978 personalObote Uganda 500 1982 1985 personalMuseveni Uganda 500 1987 2010 personalKaunda Zambia 551 1976 1991 party-basedChiluba Zambia 551 1997 2001 party-basedLevy Mwanawasa Zambia 551 2003 2008 party-basedBanda Zambia 551 2010 2010 party-basedMugabe Zimbabwe 552 1982 2010 party-based

45

●

●

●

0

2

4

6

Military Party Personal

Regime Type

Mar

gina

l Effe

ct (

Dis

mis

sals

)

●

●

●

0

2

4

6

Military Party Personal

Regime Type

Mar

gina

l Effe

ct (

Hor

izon

tal M

oves

)

Figure A1: Average marginal effect of elections with 95% confidence intervals

(Calculated using Models 2 and 4 in Table 1)

46

Table A3: Leader Random Effects

Dependent variable:Dismissals Horizontal Moves

(1) (2) (3) (4)Party Regime −0.445∗∗∗ −0.490∗∗∗ −0.278∗∗ −0.279∗∗

(0.101) (0.104) (0.122) (0.127)Election 0.339∗∗∗ 0.345∗∗∗ 0.489∗∗∗ 0.436∗∗∗

(0.106) (0.112) (0.113) (0.120)Party X Election 0.465∗∗∗ 0.462∗∗∗ 0.235 0.288∗

(0.157) (0.161) (0.161) (0.166)Military −0.249∗ −0.003

(0.148) (0.175)Mil. X Election −0.159 0.426

(0.341) (0.345)Multiparty −0.167∗∗ −0.190∗∗ −0.153 −0.150

(0.083) (0.083) (0.093) (0.094)Coup Attempt 0.286 0.280 0.141 0.149

(0.179) (0.179) (0.187) (0.187)Senior Partners −0.274 −0.298∗ 0.626∗∗∗ 0.633∗∗∗

(0.176) (0.174) (0.198) (0.198)Resource Rents 0.018 0.006 −0.062 −0.056

(0.089) (0.088) (0.107) (0.107)Ln(GDP per Capita) −0.163∗∗∗ −0.161∗∗∗ 0.117∗ 0.116∗

(0.061) (0.060) (0.070) (0.070)Leader Tenure −0.003 −0.004 −0.006 −0.006

(0.005) (0.005) (0.005) (0.005)Constant −1.270∗∗∗ −1.200∗∗∗ −2.112∗∗∗ −2.119∗∗∗

(0.089) (0.096) (0.105) (0.115)Observations 947 947 947 947Log Likelihood −2,623.012 −2,621.222 −2,097.330 −2,096.487Akaike Inf. Crit. 5,270.024 5,270.443 4,218.660 4,220.975Bayesian Inf. Crit. 5,328.264 5,338.389 4,276.900 4,288.921

Note: Negative binomial regressions with leader random effects. Models 3 and4 failed to converge. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

47

Table A4: Modeling First Order Autocorrelation

Dependent variable:Dismissals Horizontal Moves

(1) (2) (3) (4)Party Regime -0.529*** -0.558*** -0.336 -0.343

0.115 0.118 0.180 0.182Military Regime -0.221 -0.028

0.145 0.148Election 0.351*** 0.347*** 0.469*** 0.418***

0.076 0.082 0.092 0.101Party X Election 0.420** 0.430** 0.337* 0.388*

0.131 0.135 0.169 0.174Military X Election -0.076 0.423**

0.178 0.137Multiparty -0.195* -0.222** -0.288** -0.284**

0.082 0.081 0.098 0.102Coup Attempt 0.238 0.227 0.203 0.211

0.144 0.142 0.175 0.174Senior Partners -0.206 -0.219 0.768*** 0.770***

0.158 0.160 0.198 0.199Resource Rents 0.066 0.048 -0.069 -0.063

0.084 0.081 0.124 0.126Ln(GDP per Capita) -0.111 -0.110 0.135 0.135

0.060 0.060 0.071 0.071Growth 0.150 0.112 -0.362 -0.358

0.281 0.284 0.290 0.291Leader Tenure 0.039 0.036 0.045 0.043

0.027 0.027 0.032 0.032Leader Tenure2 -0.003 -0.003 -0.004 -0.004

0.002 0.002 0.002 0.002Leader Tenure3 0.000 0.000 0.000* 0.000*

0.000 0.000 0.000 0.000Constant -0.617 -0.538 -2.979*** -2.976***

0.372 0.374 0.485 0.501Leader tenure polynomial? Yes Yes Yes YesObservations 907 907 907 907Leaders 82 82 82 82Wald χ2 163.85 168.65 85.95 163.14

Note: Negative binomial generalized estimating equations regressions withAR(1) working correlation structure and semi-robust standard errors clusteredby leader. ∗p<0.05; ∗∗p<0.01; ∗∗∗p<0.001

48

Table A5: Distinguishing between pure party and hybrid party regimes

Dependent variable:Dismissals Horizontal Moves

(1) (2) (3) (4)Pure Party −0.551∗∗∗ −0.577∗∗∗ −0.292 −0.301

(0.128) (0.130) (0.200) (0.201)Hybrid Party −0.130 −0.169 −0.116 −0.122

(0.142) (0.146) (0.186) (0.191)Military −0.221 −0.038

(0.145) (0.145)Election 0.328∗∗∗ 0.332∗∗∗ 0.413∗∗∗ 0.360∗∗∗

(0.076) (0.082) (0.093) (0.103)Pure Party X Election 0.561∗∗∗ 0.563∗∗∗ 0.283 0.335∗

(0.131) (0.135) (0.186) (0.191)Hybrid Party X Election −0.064 −0.058 0.290 0.340

(0.176) (0.178) (0.231) (0.236)Military X Election −0.160 0.415∗∗∗

(0.190) (0.161)Multiparty −0.149∗ −0.178∗∗ −0.225∗∗ −0.220∗∗

(0.081) (0.080) (0.102) (0.107)Coup Attempt 0.309∗∗ 0.295∗ 0.187 0.197

(0.152) (0.152) (0.176) (0.175)Senior Partners −0.249 −0.268 0.702∗∗∗ 0.705∗∗∗

(0.187) (0.190) (0.226) (0.227)Resource Rents 0.049 0.033 −0.070 −0.065

(0.087) (0.085) (0.119) (0.121)Ln(GDP per Capita) −0.141∗∗ −0.139∗∗ 0.106∗ 0.106∗

(0.059) (0.059) (0.063) (0.064)Growth −0.004 −0.033 −0.396 −0.392

(0.248) (0.250) (0.283) (0.284)Constant −0.408 −0.336 −2.791∗∗∗ −2.781∗∗∗

(0.357) (0.357) (0.421) (0.433)Cubic Splines? Yes Yes Yes YesObservations 947 947 947 947Log Likelihood −2,624.289 −2,622.206 −2,113.196 −2,112.472θ 1.317∗∗∗ 1.326∗∗∗ 1.306∗∗∗ 1.310∗∗∗

Akaike Inf. Crit. 5,278.579 5,278.412 4,256.393 4,258.944

Note: Pooled negative binomial regressions with standard errors clustered byleader. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

49

Table A6: Controlling for Polity2

Dependent variable:Dismissals Horizontal Moves

(1) (2)Party Regime −0.486∗∗∗ −0.224

(0.111) (0.170)Military Regime −0.235 −0.030

(0.145) (0.143)Election 0.344∗∗∗ 0.343∗∗∗

(0.082) (0.105)Party X Election 0.426∗∗∗ 0.364∗∗

(0.134) (0.172)Military X Election −0.168 0.444∗∗∗

(0.191) (0.171)Multiparty −0.240∗∗ −0.047

(0.109) (0.117)Polity2 0.003 −0.031∗∗∗

(0.012) (0.011)Coup Attempt 0.287∗ 0.202

(0.153) (0.177)Senior Partners −0.348∗∗ 0.620∗∗∗

(0.159) (0.198)Resource Rents 0.054 −0.064

(0.080) (0.119)Ln(GDP per Capita) −0.116∗∗ 0.115∗

(0.058) (0.066)Growth −0.023 −0.407

(0.275) (0.277)Constant −0.469 −3.009∗∗∗

(0.357) (0.458)Cubic Splines? Yes YesObservations 947 947Log Likelihood −2,626.604 −2,109.513θ 1.310∗∗∗ (0.081) 1.326∗∗∗ (0.106)Akaike Inf. Crit. 5,285.209 4,251.025

Note: Pooled negative binomial regressions with standard errorsclustered by leader. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

50

Table A7: Controlling for Parliamentary Systems

Dependent variable:Dismissals Horizontal Moves

(1) (2)Party Regime −0.453∗∗∗ −0.240

(0.114) (0.177)Military Regime −0.234 −0.044

(0.145) (0.144)Election 0.338∗∗∗ 0.362∗∗∗

(0.083) (0.102)Party X Election 0.426∗∗∗ 0.340∗∗

(0.134) (0.170)Military X Election −0.166 0.411∗∗

(0.193) (0.162)Multiparty −0.191∗∗ −0.229∗∗

(0.079) (0.104)Parliamentary −0.293 −0.218

(0.196) (0.212)Coup Attempt 0.314∗∗ 0.193

(0.158) (0.175)Senior Partners −0.384∗∗ 0.613∗∗∗

(0.159) (0.201)Resource Rents 0.039 −0.056

(0.077) (0.120)Ln(GDP per Capita) −0.115∗∗ 0.129∗

(0.057) (0.068)GDP Growth 0.007 −0.376

(0.274) (0.280)Constant −0.479 −2.922∗∗∗

(0.351) (0.469)Cubic Splines? Yes YesObservations 947 947Log Likelihood −2,625.036 −2,112.414θ 1.316∗∗∗ (0.081) 1.311∗∗∗ (0.104)Akaike Inf. Crit. 5,282.071 4,256.829

Note: Pooled negative binomial regressions with standard errorsclustered by leader. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

51

Table A8: Removing assistant ministers

Dependent variable:Dismissals Horizontal Moves

(1) (2)Party Regime −0.478∗∗∗ −0.246

(0.114) (0.175)Military Regime −0.184 −0.027

(0.148) (0.144)Election 0.326∗∗∗ 0.376∗∗∗

(0.080) (0.100)Party X Election 0.424∗∗∗ 0.340∗∗

(0.133) (0.171)Military X Election −0.159 0.417∗∗∗

(0.192) (0.157)Multiparty −0.199∗∗ −0.247∗∗

(0.079) (0.102)Coup Attempt 0.308∗∗ 0.213

(0.153) (0.175)Senior Partners −0.323∗∗ 0.674∗∗∗

(0.156) (0.197)Resource Rents 0.042 −0.079

(0.081) (0.122)Ln(GDP per Capita) −0.133∗∗ 0.114∗

(0.053) (0.064)GDP Growth −0.003 −0.428

(0.271) (0.277)Constant −0.475 −2.869∗∗∗

(0.330) (0.451)Cubic Splines? Yes YesObservations 947 947Log Likelihood −2,592.494 −2,085.621θ 1.323∗∗∗ (0.083) 1.312∗∗∗ (0.105)Akaike Inf. Crit. 5,214.987 4,201.243

Note: Pooled negative binomial regressions with standard errorsclustered by leader. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

52

Table A9: Promotions and Demotions

Dependent variable:Promotions Demotions

(1) (2)Party Regime −0.412∗∗∗ −0.285∗∗

(0.147) (0.138)Military Regime −0.204 −0.174

(0.134) (0.166)Election 0.260∗∗ 0.298∗∗

(0.123) (0.118)Party X Election 0.385∗∗ 0.484∗∗∗

(0.174) (0.156)Military X Election 0.421∗ 0.170

(0.252) (0.332)Multiparty −0.073 −0.143

(0.111) (0.105)Coup Attempt 0.299∗ 0.336∗

(0.169) (0.195)Senior Partners 0.427∗∗ 0.121

(0.183) (0.200)Resource Rents −0.053 0.158

(0.104) (0.116)Ln(GDP per Capita) 0.002 0.104∗

(0.079) (0.058)GDP Growth 0.257 −0.090

(0.300) (0.284)Constant −3.380∗∗∗ −4.185∗∗∗

(0.534) (0.407)Cubic Splines? Yes YesObservations 947 947Log Likelihood −1,273.247 −1,178.925θ 2.697∗∗∗ (0.504) 4.166∗∗∗ (1.123)Akaike Inf. Crit. 2,576.494 2,387.851

Note: Pooled negative binomial regressions with standard errors clustered byleader. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

53

Table A10: Effect of Party Regime on Founding and Non-Founding Leaders

Dependent variable:Dismissals Horizontal Moves

(1) (2)Party (NFL) −0.569∗∗∗ −0.359

(0.163) (0.271)Party (FL) −0.411∗∗∗ −0.171

(0.110) (0.173)Military −0.236 −0.045

(0.145) (0.144)Election 0.342∗∗∗ 0.366∗∗∗

(0.084) (0.102)Party (NFL) X Election 0.498∗∗ 0.487∗∗

(0.200) (0.227)Party (FL) X Election 0.360∗∗∗ 0.150

(0.129) (0.218)Military X Election −0.158 0.412∗∗

(0.194) (0.165)Multiparty −0.222∗∗∗ −0.247∗∗

(0.082) (0.103)Coup Attempt 0.286∗ 0.174

(0.154) (0.169)Senior Partners −0.376∗∗ 0.626∗∗∗

(0.153) (0.182)Resource Rents 0.067 −0.044

(0.078) (0.135)Ln(GDP per Capita) −0.095 0.142∗

(0.065) (0.076)GDP Growth −0.043 −0.441

(0.282) (0.280)Constant −0.609 −2.998∗∗∗

(0.401) (0.545)Cubic Splines? Yes YesObservations 947 947Log Likelihood −2,625.870 −2,111.766θ 1.312∗∗∗ (0.081) 1.313∗∗∗ (0.104)Akaike Inf. Crit. 5,285.741 4,257.533

Note: Pooled negative binomial regressions with standard errorsclustered by leader. Party (NFL) = 1 for party regime leaderswho did not found the party. Party (FL) = 1 for party regimeleaders who founded the party. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

54

Table A11: Controlling for Cubic Polynomials of Regime Tenure

Dependent variable:Dismissals Horizontal Moves

(1) (2) (3) (4)Party Regime −0.365∗∗∗ −0.455∗∗∗ −0.162 −0.233

(0.103) (0.115) (0.153) (0.173)Military Regime 0.548∗∗∗ 0.345∗∗∗ 0.573∗∗∗ 0.383∗∗∗

(0.075) (0.085) (0.084) (0.102)Election −0.231∗ −0.235 0.028 −0.041

(0.139) (0.145) (0.133) (0.143)Party X Election 0.439∗∗∗ 0.333∗∗

(0.136) (0.168)Military X Election −0.143 0.417∗∗

(0.198) (0.164)Multiparty −0.219∗∗∗ −0.219∗∗∗ −0.265∗∗ −0.265∗∗

(0.083) (0.085) (0.108) (0.109)Coup Attempt 0.296∗ 0.283∗ 0.134 0.151

(0.158) (0.154) (0.154) (0.155)Senior Partners −0.333∗∗ −0.356∗∗ 0.652∗∗∗ 0.649∗∗∗

(0.151) (0.158) (0.197) (0.197)Resource Rents 0.057 0.058 −0.043 −0.043

(0.079) (0.080) (0.121) (0.121)Ln(GDP per Capita) −0.110∗ −0.115∗ 0.135∗∗ 0.130∗∗

(0.058) (0.059) (0.065) (0.064)GDP Growth −0.002 −0.014 −0.350 −0.347

(0.266) (0.272) (0.274) (0.275)Constant −0.471 −0.429 −2.865∗∗∗ −2.815∗∗∗

(0.356) (0.367) (0.456) (0.453)Cubic Splines, Leader? Yes Yes Yes YesCubic Splines, Regime Yes Yes Yes YesObservations 947 947 947 947Log Likelihood −2,629.615 −2,625.487 −2,112.616 −2,110.535θ 1.297∗∗∗ (0.080) 1.313∗∗∗ (0.081) 1.308∗∗∗ (0.104) 1.319∗∗∗ (0.105)Akaike Inf. Crit. 5,291.230 5,286.974 4,257.232 4,257.070

Note: Pooled negative binomial regressions controlling for both leader tenure and regimetenure cubic polynomials. ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

55

0

20

40

−0.525 −0.500 −0.475 −0.450 −0.425

Party Regime Estimate

Den

sity

0e+00

5e+04

1e+05

4.5e−05 5.0e−05 5.5e−05 6.0e−05

Party Regime X Election Estimate

Den

sity

0

20000

40000

60000

0e+00 2e−05 4e−05 6e−05

Party Regime Estimate P−Value

Den

sity

0

500

1000

0.001 0.002 0.003

Party X Election Estimate P−Value

Den

sity

Figure A2: Party Regime Estimates from Model 2 (dismissals) in Table 1 After DroppingIndividual Leaders

(P-values calculated using standard errors clustered by leader)

Table A12: Extreme bounds analysis

Variable Lower Extreme Bound Upper Extreme Bound % Signficant (0.05)Party Regime -0.941 -0.221 100Party X Election 0.118 0.764 100Military Regime -0.743 0.176 42Military X Election -0.595 0.301 0Election 0.047 0.519 100Note: Leamer’s extreme bounds analysis.

56

0

10

20

30

−0.50 −0.45 −0.40

Party Regime Estimate

Den

sity

0

20000

40000

60000

4.5e−05 5.0e−05 5.5e−05 6.0e−05

Party Regime X Election Estimate

Den

sity

0

10000

20000

0.00000 0.00005 0.00010 0.00015

Party Regime Estimate P−Value

Den

sity

0

100

200

300

400

0.001 0.002 0.003

Party X Election Estimate P−Value

Den

sity

Figure A3: Estimates from Model 4 (horizontal reshuffles) in Table 1 After DroppingIndividual Countries

(P-values calculated using standard errors clustered by leader)

57

Party Regime

x.values

−0.7 −0.5 −0.3 −0.1

02

46

810

Military Regime

x.values

−0.4 −0.3 −0.2 −0.1 0.0

01

23

45

6

Election

x.values

0.00 0.10 0.20 0.30

05

1015

20

Party X Election

0.0 0.1 0.2 0.3 0.4

05

1015

2025

Military x Election

−0.15 −0.10 −0.05 0.00

05

1015

20

(a) DismissalsParty Regime

x.values

−0.4 −0.3 −0.2 −0.1 0.0

01

23

45

6

Military Regime

x.values

−0.20 −0.10 0.00

02

46

8

Election

x.values

0.0 0.1 0.2 0.3

05

1015

2025

Party X Election

0.0 0.1 0.2 0.3 0.4

05

1015

2025

Military x Election

0.0 0.1 0.2 0.3 0.4

05

1015

20

(b) Horizontal Reshuffles

Figure A4: Extreme bounds analysis

58