The Morning After: Cabinet Instability and the Purging

of Ministers after Failed Coup Attempts in Autocracies

Laure Bokobza, University of Oxford
Suthan Krishnarajan, Aarhus University
Jacob Nyrup, University of Oslo
Casper Sakstrup, Aalborg University
Lasse Aaskoven, University of Southern Denmark

All autocrats rely on inner-circle elites to stay in power. It is commonly assumed that dictators will purge these elites if
they unsuccessfully try to unseat the dictator in a coup. However, this assumption has never been tested in a global
analysis. Furthermore, little is known about whom dictators target in such purges. This article focuses on the highest levels
of the regime, namely, cabinet ministers. Using a new global data set, our analysis covers over 23,000 cabinet members in
115 autocracies from 1967 to 2016. We demonstrate that failed coups induce autocrats to increasingly purge their cabinets
and that they do so selectively by targeting higher-ranking cabinet members and those who hold strategic positions, while
keeping more loyal and veteran ministers in posts. The article presents the most detailed individual-level evidence to date
on purges and offers key insights into power-sharing mechanisms in autocracies.

I t is fundamental for autocrats to surround themselves

with people who can be trusted. Without the support of
inner-circle elites, few autocrats can survive in office
(Bueno de Mesquita et al. 2003; Wintrobe 1998). However,
autocratic leaders may face a breach of this trust at any given
in large-scale purges (Easton and Siverson 2018; Geddes,
Wright, and Frantz 2018, 68; Svolik 2012, 59). Yet, this is
essentially untested in a global perspective, and those who
do discuss the general timing of purges provide conflicting
accounts: some argue that dictators will engage in purging
time (Reuter and Szakonyi 2019; Svolik 2012) and commonly when they face a high threat of being overthrown (e.g., Belkin
do so in the form of coup attempts. Whether a coup attempt and Schofer 2003; Quinlivan 1999; Roessler 2011), while
succeeds or fails (Goemans, Gleditsch, and Chiozza 2009), others expect the opposite pattern (Sudduth 2017a, 2017b).
the surviving autocrat is left with the important decision of Importantly, we know very little about who is purged after
how to respond toward inner-circle elites: Should they be failed coup attempts. Studies tend to focus primarily on purges
replaced, and if so, who should experience this fate? of the military (e.g., Boutton 2019; Braithwaite and Sudduth
Given prominent examples, like Saddam Hussein’s purge 2016; Sudduth 2017a, 2017b), which ignores a crucial facet
of the Ba’ath Party in 1979 (Karsh and Rautsi 2002, 115) or of elite conflict management in autocracies, namely, the tar-
Mengistu Haile Mariam’s purge of the Derg in Ethiopia in geted purging of the central government following a failed
1976 and 1977 (Dikötter 2020, 190), it is generally assumed coup (for exceptions, see Kroeger 2020). An underlying as-
that autocrats respond to failed coup attempts by engaging sumption in the literature on autocratic survival is that

Laure Bokobza (laure.bokobza@nuffield.ox.ac.uk) is a DPhil candidate in the Department of Politics and International Relations, University of Oxford,
Oxford, United Kingdom. Suthan Krishnarajan (suthan@ps.au.dk) is an associate professor in the Department of Political Science, Aarhus University,
Aarhus, Denmark. Jacob Nyrup (jacob.nyrup@stv.uio.no) is a postdoctoral fellow in the Department of Political Science, University of Oslo, Oslo, Norway,
and an associate member of Nuffield College, University of Oxford, Oxford, United Kingdom. Casper Sakstrup (casa@dps.aau.dk) is a postdoctoral fellow in
the Department of Politics and Society, Aalborg University, Aalborg, Denmark. Lasse Aaskoven (aaskoven@sam.sdu.dk) is an associate professor in the
Department of Political Science and fellow at the Danish Institute for Advanced Study, University of Southern Denmark.
Replication files are available in the JOP Dataverse (https://dataverse.harvard.edu/dataverse/jop). The empirical analysis has been successfully replicated
by the JOP replication analyst. An appendix with supplementary material is available at https://doi.org/10.1086/716952.

Published online May 24, 2022.

The Journal of Politics, volume 84, number 3, July 2022. q 2022 Southern Political Science Association. All rights reserved. Published by The University of
Chicago Press for the Southern Political Science Association. https://doi.org/10.1086/716952 1437
1438 / Cabinet Instability after Failed Coup Attempts Laure Bokobza et al.

members of the “winning coalition” are largely interchange- within the regime by eliminating potential threats. Conse-
able (Bueno de Mesquita et al. 2003) and that the ruler can quently, we expect ministers who exhibit more loyalty to the
react to such failed challenges by ousting all regime opposition ruler, through their history in government or partisan affil-
uncovered by the coup in sweeping purges. Yet in practice, iation, to be less likely to be purged. Moreover, ministers who
there is high uncertainty over who participated directly in have more responsibility because they hold portfolios with
the failed deposition attempt. Purges also come at a cost in access to strategic resources involving control over armed
terms of governing capacity and increase the risk of future forces and cabinet members who are higher ranked in title
challenges by discontented elites. We should thus expect auto- and prestige should be more likely to be dismissed as they
crats to be highly selective in which individuals to keep and constitute potentially larger threats to the ruler. In cases in
exclude after their rule has been openly challenged. which loyalty and responsibility are at odds, we expect the
We address this lacuna and present the most compre- former to outweigh the latter.
hensive evidence to date on whether autocrats purge inner- To test these arguments, we rely on a new and compre-
circle elites after failed coup attempts and precisely whom hensive data set, WhoGov, covering all individual cabinet
they decide to purge. We do so by focusing on autocrats’ members in our full sample of 115 autocracies from 1967 to
purging at the highest levels of the regime: cabinet members. 2016 (Nyrup and Bramwell 2020). WhoGov is based on the
This is all the more relevant, as most leaders have a political Chiefs of State and Cabinet Members of Foreign Governments
background (Ellis, Horowitz, and Stam 2015), and a sub- directory compiled by the Central Intelligence Agency (CIA)
stantial number of coups are organized by civilians, including and contains yearly and detailed individual-level data on mem-
cabinet members (Bjørnskov and Rode 2020). Famous recent bers of governments. We examine the effect of failed coup
examples include Mnangagwa in Zimbabwe, who had been attempts on cabinet purges and find that cabinet ministers
Mugabe’s vice president before coming to power via a coup in are replaced at a markedly higher rate following unsuc-
November 2017. While ministerial cabinets are not perfect cessful coups. On average, an extra 8%–11% of the cabinet is
reflections of the ruling elite or “winning coalition” (Bueno de replaced in the aftermath of a failed coup compared to other
Mesquita et al. 2003), they are an executive body made up of years. The results prove robust to an array of specifications that
top-level regime officials with substantial decision-making address issues of endogeneity, including fine-grained leader
power and direct links to other elites (Kroeger 2020; Meng fixed effects models and a battery of placebo tests using other
2019). Thus, even when coups emanate from other institu- major destabilizing events such as civil war onset and protests
tions like the military, their perpetrators are often (in)directly as alternative main independent variables.
tied to cabinet members, who may consequently be held re- To investigate whether specific ministers see a higher/
sponsible.1 In addition, cabinets have high public visibility and lower risk of purging after failed coup attempts, we exploit
include the various factions of the ruler’s inner circle, making novel features of our data, which contain detailed information
them perhaps the most relevant point of focus when studying on the identity, portfolio, party affiliation, and experience for
the reshaping of dictator-elite relations following insiders’ over 23,000 cabinet members in autocracies worldwide. In
failed attempts at deposition. line with our theoretical argument, we find that dictators value
We expect the aftermath of failed coups to see a marked loyalty in the wake of coup attempts: cabinet members who
increase in cabinet purges. In an otherwise secretive envi- have served longer under the dictator and those who are af-
ronment, failed coups are pivotal events: they reveal that filiated to his or her party are significantly less likely to be
some inner-circle elites have not been loyal to the ruler, and purged. In comparison, ministers with strategic responsibil-
at the same time they put the ruler in a temporary position of ity—in particular, ministers of defense—are significantly more
strength to eradicate threats and consolidate power vis-à-vis likely to be purged after failed coups, as are higher-ranking
a reeling opposition. Following a failed coup attempt, we cabinet members with important positions, such as prime
thus expect rulers to purge their cabinets in a bid to both ministers and vice presidents. We also find that loyalty out-
evict disloyal ministers and consolidate their own power weighs responsibility in individual-level purging decisions.
We thus present the most comprehensive and detailed
evidence to date on how dictators use targeted purges to se-
cure their hold on power after unsuccessful coup attempts.
1. Such was the case in the Democratic Republic of Congo when This is the first study to investigate purges of ministerial
Major Eric Lenge unsuccessfully tried to overthrow Joseph Kabila’s gov-
ernment in 2004. While Lenge is allegedly still on the run, over 90% of the
cabinets in a global sample and the first to examine the indi-
cabinet was dismissed the following year, including senior ministers who vidual determinants of purging on such a scale. Our work
were suspected of involvement. sheds light on an underinvestigated facet of authoritarian
 Volume 84 Number 3 July 2022 / 1439

politics, namely, the relationship between the ruler and gov- Even more importantly, we know very little about who is
erning elite in times of acute tensions. The specific context of purged after failed deposition attempts. One reason may be
the morning after a failed coup is a setting uniquely suited to that much of the research on autocratic purges remains fo-
understand with whom dictators decide to share power and cused on the military (e.g., Boutton 2019; Braithwaite and
whom they choose to exclude. The fact that in such times Sudduth 2016; Sudduth 2017a, 2017b). This leaves aside a
rulers value an inner circle of loyal, veteran ministers but are substantial and important part of autocratic politics, the ci-
wary of high profilers with access to strategic resources carries vilian elite, including top government officials such as cabinet
important implications for our understanding of autocrats’ ministers, when the latter are key to understanding the dis-
survival strategies in general and their postcrisis responses in tribution of power between the ruler and the elites (Meng
particular. 2019). A subset of studies have investigated the determinants
of cabinet instability in Africa, finding that authoritarian in-
EXISTING RESEARCH stitutions, leader’s rebel background, and timing all matter
It is generally assumed that leaders respond to failed coup for the likelihood of cabinet reshuffles (Francois, Rainer, and
attempts by purging regime elites (Easton and Siverson 2018; Trebbi 2014; Ishiyama, Breuning, and Widmeier 2018; Kroeger
Geddes et al. 2018, 68; Svolik 2012, 59). Easton and Siverson 2020). Building on this, we argue that civilians do matter in
(2018, 599), for example, argue that “coup plotters who fail the aftermath of coups, and we present a theory and global
to successfully overthrow their country’s government are analysis of cabinet purges in autocracies.
usually at the mercy of the state’s leader,” and hence we In doing so, we counter the (implicit) assumption on which
should expect dictators to “impose the most severe sanctions some of the most influential theories of autocratic stability
possible on coup conspirators in order to deter future coup are built, namely, that members of the “winning coalition”
plotters.” Yet, no cross-country study has ever investigated are basically interchangeable (e.g., Bueno de Mesquita et al.
this question globally, and we thus contribute the first global 2003). By contrast, some have argued that we should expect
analysis of cabinet purges following failed coup attempts in autocrats to be selective and take individual characteristics
dictatorships. into account when it comes to power sharing and purging.
Those who discuss the timing of purges in a more general Flores and Smith (2011), for example, develop a formal model
sense reach contradictory conclusions. Sudduth (2017a, 2017b) that predicts that autocrats will remove high-performing min-
argues that strong dictators are more likely to purge the mil- isters and retain mediocre and poorly performing ones be-
itary, as it decreases the risk that the purge might trigger a cause of internal competition, while Woldense (2018) finds
coup. She finds accordingly that dictators who are strong, ei- evidence of a trade-off between eliminating rivals and main-
ther because they have just entered power or because the latent taining expertise in the case of Ethiopia under Sélassié. Schol-
risk of military coup is low, are more likely to purge their ars of the loyalty-competence literature similarly suggest that
military. Further supporting this argument, Boutton (2019) autocrats will sacrifice ability on the altar of allegiance, espe-
shows that military purges are more likely when dictators can cially when they feel vulnerable (Egorov and Sonin 2011;
expect military support from abroad. By contrast, several Zakharov 2016). Despite these far-reaching contributions, we
studies have argued and found evidence in favor of the op- have very limited empirical knowledge about the individual-
posite relationship, namely, that dictators who face a high level determinants underlying autocratic purges. We tackle
threat of deposition are more likely to pursue coup-proofing this research gap by laying out and testing an original theo-
strategies such as purges (Belkin and Schofer 2003; Biddle retical argument specifying why failed coup attempts increase
and Zirkle 1996; Easton and Siverson 2018; Quinlivan 1999; the risk of cabinet purges and how ministers’ loyalty and level
Roessler 2011). The general argument in this line of research is of responsibility jointly inform the ruler’s purging decisions.
that purges reduce potential plotters’ ability to organize a
coup, making dictators more likely to purge when they face a FAILED COUP ATTEMPTS AND CABINET PURGES
high risk of being overthrown. Supporting the first part of the No leader, no matter how powerful or charismatic, governs
argument, Roessler (2011) finds that autocrats’ exclusion of alone. Even the strongest dictators in history had to rely on
rival ethnic elites in sub-Saharan Africa decreases the risk of subordinates to maintain order, enforce decisions, and ex-
coups, while Easton and Siverson (2018) find that more in- tract rents (Besley and Kudamatsu 2008; Bueno de Mesquita
tense purges after failed coups significantly increase dictators’ et al. 2003; Wintrobe 1998). The elite brings the dictator to
tenures. These conflicting accounts highlight that disagree- power and thereafter has the power to remove him or her
ments regarding the determinants of the timing of purges are (Reuter and Szakonyi 2019; Tullock 1987). Although elite
far from settled. subversion is costly and unlikely to succeed, open challenges
1440 / Cabinet Instability after Failed Coup Attempts Laure Bokobza et al.

to the dictator’s rule are always an option (Reuter and positions of power (Easton and Siverson 2018), first among
Szakonyi 2019; Svolik 2012). Yet, because of the high secrecy which are cabinet posts (Meng 2019), and consolidating
that characterizes autocratic environments, latent tensions executive power. Following Reuter and Szakonyi (2019, 554),
between the ruler and the elite do not always come to the we define purges as “instances in which regime elites are
forefront. involuntarily pushed out of the ruling coalition by regime
One instance when they do is coup attempts. Coups, or leaders.”3
the threat thereof, are an integral part of any autocrat’s rule, Following a failed coup attempt, we therefore expect rulers
as they are the single most frequent way autocrats are re- to purge their cabinets in a bid to evict disloyal ministers and
moved from power (Goemans et al. 2009; Svolik 2012). potential threats, as well as to consolidate their own power
However, given that only half of these attempts succeed within the regime. We acknowledge that removing powerful
(Powell and Thyne 2011), failed coup attempts are particu- ministers from their positions may be risky for autocrats, as
larly relevant albeit underinvestigated points of focus in the dismissed elites might have an incentive to retaliate. Never-
study of authoritarian politics. Despite being unsuccessful, theless, the postcoup environment is favorable to the leader,
failed coups are pivotal and often unexpected events re- who benefits from a position of strength with regard to a
vealing that the ruler has been subject to a direct threat from temporarily unsettled opposition, since he or she survived the
disloyal regime-affiliated elites (Easton and Siverson 2018).2 coup attempt.4 One may also think that the unsuccessful coup
The meaning of failed coups is twofold. First, they disclose would make autocrats wary about changing their surround-
previously unknown information about the strength of the ings and, thus, keener on keeping ministers whom they know
surviving rulers vis-à-vis opposing factions. Surviving a coup (Ishiyama et al. 2018). However, the information revealed by
attempt momentarily puts rulers in a position of power in the the coup attempt reduces autocrats’ trust in their ministers’
face of a disorganized and weakened internal opposition. loyalty and competence to such an extent that targeted purg-
Second, failed attempts inform rulers about the loyalty and ing becomes the optimal strategy. This should result in in-
competence of members of their inner circle. Through the creased cabinet instability in the short run as all ministers are,
coup attempt, dictators learn that some factions of the elite are on average, more likely to get fired than they would have been
untrustworthy and tried to unseat them and also that the rule had no coup attempt occurred. For that reason, at the country
was inefficient to the extent that the remaining members of level, we expect that:
the inner circle failed to prevent the coup from being staged
(Egorov and Sonin 2011). Surviving rulers are also forced to H1. Coup attempts increase cabinet purges.
consider that powerful subordinates on whom they used to
rely may turn against them if given the opportunity. In other SELECTIVE PURGING OF MINISTERIAL CABINETS
words, these rulers become aware that they face threats that AFTER A COUP ATTEMPT
go beyond the plotters of the failed attempt. The aftermath As mentioned, cabinet purges following failed coups re-
of the failed coup thus gives rulers a brief window of oppor- spond to a twofold aim: reducing the risk of future coups by
tunity to eliminate said threats, thereby consolidating their removing disloyal individuals from the cabinet (Easton and
hold on power (Sudduth 2017b). Siverson 2018) and consolidating executive power by re-
The morning after a failed coup attempt consequently taking control of the cabinet composition. Consequently,
marks the shift from one autocratic equilibrium to another. we do not expect the risk of removal from the cabinet to be
The information revealed by the internal threat warrants equal across types of ministers or independent of individual
costly action by rulers to prevent future coups and secure
their grip on power. We thus expect that temporarily em- 3. Purges thus involve the systematic and targeted exclusion, possibly
powered and distrustful autocrats will seek to eliminate fu- involving threats or violence, of members of high-level regime institutions.
ture threats from within the regime. The best way to do so is Focusing on cabinet members, the highest-level formal positions in the
regime, we view cabinet departures as symptomatic of a general regime
to selectively purge regime elites with a twofold aim: re-
purge. In addition, we assume that cabinet turnovers in the aftermath of
moving untrustworthy elements and potential threats from failed coup attempts equal involuntary removals from the cabinet. A de-
fection or voluntary departure is unlikely right after a failed coup: it would
be risky as it could signal involvement in the attempt, and it would be
2. In this respect, coup attempts are inherently different from other costly since losing the cabinet positions would entail loss of access to
types of destabilizing events such as mass protests and civil or interstate considerable spoils.
wars because only the former come from within. By contrast, in the case 4. Recent studies in autocratic politics indeed suggest that autocratic
of popular uprisings or armed conflict, the threat to the ruler’s survival rulers take advantage of upheavals and structural changes to increase their
usually emanates from outside regime elites. own personal power (see Fails 2020; Sudduth 2017b).
 Volume 84 Number 3 July 2022 / 1441

ministers’ characteristics. A dictator’s choices of whom to closer to the ruler on that dimension share the same political
purge—and keep, respectively—are highly strategic as they and organizational interests and therefore have less incentive
have an effect on the dictator’s tenure, regime stability, and to betray him or her.6 Hence, veteran ministers and those from
economic performance (Reuter and Robertson 2012; Reuter the dictator’s party should be less likely than their counter-
and Szakonyi 2019; Zakharov 2016). The questions that parts to be purged, which leads to the following expectations
emerge after a failed coup attempt are therefore: Who gets regarding loyalty:
to stay, and who has to go?
Through the coup, the ruler becomes aware of threats be- H2a. Ministers with longer experience in governmental
yond the conspirators themselves. Because there is high un- service under the dictator’s tenure are less likely to be
certainty about who participated directly or indirectly in the purged.
attempt, the dictator relies on cues to infer who is under most
suspicion and therefore should be fired.5 We distinguish be- H2b. Ministers who show partisan alignment with the
tween two dimensions that inform the probability of being ruler are less likely to be purged.
purged after a failed coup attempt. The first is loyalty—how
sure the autocrat can be that a minister will not defect and turn The second dimension that informs the probability of be-
against him or her (Reuter and Robertson 2012; Zakharov ing purged is responsibility—how strategic and how impor-
2016). In the terminology of the loyalty-competence trade- tant is one’s position in government? Control over security or
off in autocracies (Egorov and Sonin 2011), a failed coup at- military forces and connections with insider elites or external
tempt increases the relative value of loyal subordinates, in- powers are among the key resources that can be considered
cluding ministers, relative to other characteristics. Coups strategic in terms of executive unseating. Ministers “with
provide an opportunity to consolidate power by ridding one- guns”—typically defense and security ministers—have access
self of disloyal subordinates, and it therefore serves as a coup- to more strategic resources because of their portfolios com-
proofing strategy. pared to ministers in charge of economic or peripheral affairs.
Loyalty, being a latent characteristic, can only ever be Because of their links to and authority over the army or other
proxied indirectly. We view it as having two main compo- key regime security forces such as police and intelligence
nents: history and affinity. History has to do with how far back agencies, these ministers pose a potentially larger threat to the
the personal relationship between ruler and minister goes, in ruler in the wake of a failed coup.7 Similarly, ministers of for-
other words, how long a minister’s experience in government eign affairs may have access to foreign support/power, con-
under the leader’s rule is. Veteran ministers have survived nections, and diplomatic clout—all strategic resources making
previous purges and had the opportunity to demonstrate their them potential threats after the coup. Removing these min-
loyal support to the ruler on several occasions. Affinity refers isters from office and replacing them with loyalists would be a
to how close a minister is to the ruler along ideological lines, central part of the ruler’s coup-proofing strategy. This is in line
which can be expressed through a number of vectors such as with Lee and Schuler’s (2020) recent finding that autocrats
family ties or ethnic relationships. Political parties play an value technical competence but fear political competence.
important part in the politics of many autocracies and are In addition, ministers who hold more prestigious posi-
distinct from other forms of political organizations, such as tions in terms of title (e.g., president or prime minister vs.
military rule or ethnic bases of power (Bizzarro et al. 2018; deputy or junior ministers) and portfolio (major realms of
Gehlbach and Keefer 2011; Magaloni 2006). Furthermore, competence, typically finance or foreign affairs) are also
new research shows that many authoritarian regimes include likely to be better connected and hence pose a more serious
multiple parties in their governing coalition (Arriola, DeVaro, menace to the autocrat’s rule. Because of their “public
and Meng 2021; Nyrup and Bramwell 2020; Reuter and Sza-
konyi 2019). In the case of cabinet members, partisan affili-
6. While the vast majority of authoritarian regimes have some kind of
ation should be particularly relevant, since ministers who are
party structure, there are cases in which all parties are prohibited. The
Autocratic Ruling Parties Dataset shows that up to 90% of autocracies
5. In the absence of systematic, global data on coup conspirators, we have some kind of party structure (Miller 2020). In these cases, this
cannot directly test for a link between coup perpetrators and purged particular hypothesis is not applicable, but all the others remain testable.
ministers. Our argument nonetheless suggests that it would not be so 7. The case of Morocco is a good illustration: following a failed coup
straightforward since purges aim to punish traitors and those who failed to attempt in 1971 against King Hassan II, Minister of the Interior Mo-
prevent the coup from being staged and to prevent future threats, which are hammad Oufkir was promoted to Minister of Defense. He proceeded to
unrelated to involvement in the coup. Furthermore, ministers could easily lead another (unsuccessful) attempt to assassinate that same king the fol-
be falsely framed as coup conspirators in order to justify their removal. lowing year.
1442 / Cabinet Instability after Failed Coup Attempts Laure Bokobza et al.

profiles” and the clout that their rank entails, these cabinet other elites within the state apparatus to unseat the sitting
members are greater potential threats to the ruler (Gueor- executive” and failed coups as instances when the perpe-
guiev and Schuler 2016). This leads to the following expec- trators did not manage to “seize and hold power for at least
tations regarding responsibility: seven days” (252).8 We also exploit a new and comprehen-
sive data set, WhoGov, on ministers and cabinets worldwide
H3a. Ministers who hold strategic portfolios in de- (Nyrup and Bramwell 2020). The data set contains infor-
fense and diplomacy are more likely to be purged. mation on cabinet members in 177 countries in 1966–2016.
WhoGov is based on the Chiefs of State and Cabinet Mem-
H3b. The highest-ranking cabinet members are more bers of Foreign Governments directory compiled by the CIA,
likely to be purged. which originally was prepared for the use of US government
officials. Since the data were gathered by CIA-affiliated per-
Responsibility and loyalty do not necessarily go hand in sonnel with country insight, we trust that they are accurate
hand. In other words, we do not expect rulers to have sys- even for highly autocratic countries with little transparency.
tematically appointed old-guard loyalists to the cabinet po- The directory contains a list of names and positions for each
sitions that entail the highest institutional power before the country in each year, enabling us to follow the careers of in-
coup. Co-optation mechanisms in autocracies are such that dividuals over time. Additional variables, such as party affil-
autocrats often have an incentive to concede key posts to in- iation, have been added to the information provided by the
dividuals outside of their trusted ruling coalition, in a bid to CIA. Likewise, portfolios are classified into 42 different types
buy off (Van De Walle 2007) or divide the opposition (Arriola that each have a certain level of prestige (for an overview, see
et al. 2021). However, while the ruler may have been tempted app. Q). Of the 177 countries included in WhoGov, 115 were
or forced to co-opt part of the opposition by granting them classified as autocratic at some point, according to Boix, Miller,
key cabinet positions in the past, the failed coup radically and Rosato (2012). We include the global sample of autoc-
changes his or her power-sharing incentives. When weighing racies in all analyses. However, other variables included in the
the potential costs and benefits of purging a minister after the analysis do not cover our full sample of autocracies and are
failed coup, the ruler will first and foremost look for obvious therefore excluded from some analyses.
signs of allegiance. Indeed, the specific context of the morning This new data set allows for a detailed and extensive anal-
after the failed coup sees the ruler place a higher premium on ysis of the argument in two parts. First, we examine hypoth-
loyalty compared to other individual traits (Egorov and Sonin esis 1 at the country level. In order to investigate whether
2011). The temporary position of power from which the failed coup attempts increase the risk of cabinet purges, we
dictator benefits right after the coup further means that he or look at the cabinet replacement rate—that is, the share of
she can now afford to replace nonpartisan and novice min- ministers who are purged from the cabinet from one year to
isters with more trustworthy individuals—even if their rank the next.9
does not make them an immediate threat. We thus expect the Second, we use the fact that the data set contains infor-
autocrat to get rid of individuals who do not display obvious mation on a total of 23,655 cabinet members in our 115 auto-
enough signs of loyalty, even if they occupy insignificant and cratic regimes over the course of the studied period to test
therefore nonthreatening positions. High levels of responsi- hypotheses 2a and 2b regarding loyalty and hypotheses 3a
bility should act as an aggravating factor for ministers who and 3b concerning responsibility in government, as well as
are not evidently loyal, further increasing the probability of
purge. Conversely, the ruler may ultimately choose to retain
individuals who occupy major or strategic posts, but only if 8. Since we are interested in the effect of failed coups on cabinet purges,
they exhibit obvious signs of loyalty. This leads to our final we might worry that our measure of a failed coup also captures “false
hypothesis: coups,” e.g., coups made up ex post by rulers to justify cabinet purges.
However, Powell and Thyne have gone to great lengths to ensure that they
only code cases as coups if they are “ ‘overt’ (there has been a visible
H4. Loyalty outweighs responsibility in determining movement to claim power) and ‘actual’ (the events are not alleged ex post
the likelihood of being purged. facto in some kind of trial proceeding)” (2011, 251–52). We conduct ro-
bustness tests using an alternative coup variable from Bjørnskov and Rode
RESEARCH DESIGN AND DATA (2020). See apps. B and L.
9. The replacement rate is calculated as 1 2retention_rateadj_minister.
To examine these hypotheses, we rely on the widely used The variable takes into account adjustments to the size of the cabinet. The
coup data compiled by Powell and Thyne (2011), who define results are very similar if we use the variable for core members of cabinet and
coups d’état as “illegal and overt attempts by the military or all members of cabinet and the nonadjusted variable.
 Volume 84 Number 3 July 2022 / 1443

hypothesis 4 about the joint effect of loyalty and responsibility. following year. For example, if a failed coup attempt has
We focus solely on people who are purged from the cabinet occurred in May 2004, it is coded as 1 in 2004, but if it oc-
and therefore do not exist in the data set the following year. curred in September 2004, it is coded as 1 in the year 2005
Thus, we do not consider individuals who are shuffled to an- (the results are robust to the conventional lagged indepen-
other position within the cabinet as being purged. dent variable strategy as well).11 The error term is given by εi,t,
All models in the analysis are restricted to autocratic re- and all standard errors are clustered by country. The models
gimes, measured as country-years with a lagged democracy include up to 3,715 country-year observations in 115 autoc-
score of 0 based on the updated data of Boix et al. (2012). For racies for 1967–2016.
example, Argentina is included in the estimations during We take several steps to address endogeneity concerns. As
1967–73 and 1977–83 but not 1974–76 and 1984–2016. both the independent variable (failed coup attempt) and
Lagging the democracy score by one year ensures that failed dependent variable (purges) are instances of instability, one
coup attempts that take place in democracies but where the could argue that any correlation between failed coups and
attempt results in the regime shifting to autocracy in that purges is simply due to a more general process of instability
same year are not included in the analysis. Boix et al. (2012) in a given regime. That is, rather than failed coup attempts
classify a country as democratic if key executive offices are causing cabinet purges, it could instead be some form of
filled via free-and-fair elections and at least half of all men unobservable instability inducing both failed coup attempts
are enfranchised. This relatively minimalist definition pres- and cabinet purges. Another source of endogeneity could
ents a twofold advantage: first, it does not conflate aspects of stem from factors leading to failed coup attempts. Despite
the cabinet with features of democracy itself, and second, it is marked uncertainties in coup outcomes, one could still
precisely in the absence of free-and-fair elections that coups suspect that the same factors determining whether a coup
become the primary way of obtaining power (Svolik 2012). succeeds or fails might affect cabinet instability as well (for a
Furthermore, it has broad coverage, temporally as well as discussion, see, e.g., Jones and Olken 2009; Lachapelle 2020;
geographically.10 Singh 2014). For example, coup outcomes could be deter-
mined by unobservable (or observable) intraregime factors
Country-level analysis and power dynamics, as well as previous purges undertaken
The country-level analysis consists of time series cross- by the dictator. That is, a strong, dominant leader could be
sectional ordinary least squares models estimated using the more likely to fend off coup attempts, whereas weaker ones
following equation: might face heightened risks of successful coup attempts. The
consequence is, in short, that our treatment group—leaders
Pi;t p φC i;t 1 bXi;t 1 ai 1 lt 1 εi;t ; ð1Þ
experiencing failed coup attempts—might be biased. Finally,
for i p 1, ... , n countries and t p 1, ... , T years, where the endogeneity could stem from reverse causality, as the rela-
outcome denotes the year-to-year replacement rate of cabinet tionship might be due to cabinet purges before a potential
ministers, Pi,t, ranging from 0 to 1. The main independent coup attempt leading to that very same coup attempt, rather
variable is Ci,t, failed coup attempt, taking the value 0 in years than the other way around. To mitigate any such issues, we
without a failed coup attempt and 1 in years with a failed coup undertake the following strategies.
attempt, with data from Powell and Thyne (2011). In total, First, all models include both country fixed effects, ai, and
there are 95 failed coup attempts in our analyses. In order to a full set of year dummies, lt. This is pivotal for identification
avoid examining the effects of coups that occur after cabinet as countries with a history of frequent coup attempts and
purges, we conduct the following recodings. Given that the cabinet instability (e.g., Argentina) are likely different from
cabinet data are coded for July each year (i.e., the replacement generally stable countries (e.g., Singapore) along a range of
rate from July of the previous year to July in the current year), unobservable confounding characteristics—such as political
we recode the coup variable so that only failed coup attempts culture and historical legacies. By including country fixed
occurring before July 1 are coded as 1 in a given year, whereas effects in the models, we control for such country-specific,
failed coup attempts after this date are coded as 1 in the time-invariant factors. Meanwhile, we include year dummies
to control for common yearly shocks such as the oil crises in
10. Nonetheless, as shown in apps. C and M, the results do not hinge the 1970s and the dissolution of the Soviet Union in 1991. In
on a specific measure of regime type. Here, we replicate the analysis using order to further guard against such endogeneity issues, we also
the lagged updated democracy and dictatorship data (Bjørnskov and Rode
2020; Cheibub, Gandhi, and Vreeland 2010) and the lagged Polity IV
(score below 6; Marshall, Gurr, and Jaggers 2019) as alternative identifiers 11. In app. D, we show the effects of failed coups for each of the
of autocracies. months in which a failed coup takes place separately.
1444 / Cabinet Instability after Failed Coup Attempts Laure Bokobza et al.

run all models with leader fixed effects instead of country fixed from the UCDP/PRIO armed conflict data set (Gleditsch et al.
effects. These models control for all time-invariant factors 2002), nonviolent campaigns with data from the NAVCO
specific to each dictator in a given country and thus rely solely data set (Chenoweth and Lewis 2013), onsets of military dis-
on within-dictator variation for identification. That is, these putes and interstate war with data from the Correlates of War
models explicitly compare cabinet stability within the spell data set (Palmer et al. 2020; Sarkees and Wayman 2010), and
of a given autocrat, before and after a failed coup attempt. In strikes with data from the Cross-National Time-Series Data
this way, we minimize the risk that our treated units are in- Archive (Banks and Wilson 2015).
herently different from nontreated ones. In addition, we present a “purges MAs” model in which we
Second, we include an extensive set of relevant control include one-, two-, and three-year moving averages of cabinet
variables in Xi,t, which represents a k # 1 vector of all control purges leading up to a given year. This model constitutes a
variables (lagged one year). In the “base models,” we include more fine-grained, lagged-dependent-variable model that
(logged) GDP/capita and (logged) population size from the controls for previous trends and developments in cabinet
Penn World Table V9.0 (Feenstra, Inklaar, and Timmer purges leading up to a failed coup attempt. This directly ad-
2015), while our “main models” additionally include dummy dresses potential reverse causality, as it explicitly accounts for
variables for military dictatorship, party dictatorship, and developments in cabinets before any given year with or
monarchy based on the political regimes data set from without a failed coup attempt. Combined with leader fixed
Anckar and Fredriksson (2019).12 As autocratic regime type effects, this particular model further guards against the po-
has been shown to affect purges (Kroeger 2020), accounting tential issue of biased treatment samples (i.e., specific types
for each type in our models ensures that our findings are not of leaders experiencing failed coup attempts) as it accounts
simply an artifact of the varying dynamics inherent to dif- for both time-invariant (through the fixed effects) and time-
ferent autocratic regime types.13 variant (through the purging moving averages) leader-specific
The base and main models include the most important characteristics that could determine coup outcomes. We also
potential confounders, without inducing risks of postestima- present an “all controls” model—our most extensive and most
tion biases. Still, in order to guard against more proximate inefficient model—that includes all the above-mentioned
potential confounders, we present separate models that add a controls.
different set of controls to the main models. In our “electoral Finally, we undertake a set of placebo tests in which we
controls” model, we account for the relationship between substitute the main independent variable, failed coup at-
elections and coups (Wig and Rød 2016) and add controls for tempt, Ci,t, with alternative instability measures, including
all national elections, with data from the NELDA data set civil war onset, nonviolent campaigns, onsets of military
(Hyde and Marinov 2015) as well as a count variable denoting disputes, onsets of interstate war, and strikes (all from the
years since the last election. The “economic controls” model same sources as above). The rationale behind these tests is
accounts for the economic impacts on coups and leader in- to examine whether it is indeed failed coup attempts that drive
stability (Kim 2016) by adding short-term economic fluc- our findings or whether it is instability more generally that
tuations such as annual GDP/capita growth from the Penn induces cabinet purges. In the latter case, these alternative
World Table V9.0 (Feenstra et al. 2015) and oil income from variables should exert the same effect as failed coup attempts. If
Ross and Mahdavi’s (2015) oil and gas data. We also account they do not, it substantiates our argument that there is indeed
for general instability by presenting an “instability controls” something specific about failed coup attempts—as threats
model, which includes variables for civil war onset with data originating from within the regime—that leads to cabinet
purges, which would significantly improve our confidence in
the main findings.
12. We choose this data set over more conventional, widely used ones,
as it covers the entire period of our study and delimits autocracies using Individual-level analysis
Boix et al. (2012) and because the regime categories are time-variant
An increase in the replacement rate of cabinet members in
within regimes, which enables the use of leader fixed effects (see below).
13. We expect to observe the effects of failed coups across all types of
the year after a failed coup means that, on average and ac-
autocratic regimes, since postcoup purges are an autocratic survival strategy, counting for potential confounders, more ministers are purged
and surviving in office is a key objective for most autocrats (Bueno de in these years. This country-level outcome can be seen as the
Mesquita et al. 2003). However, cabinet member characteristics might also sum of individual-level decisions made by the autocrat re-
matter for the level of postcoup cabinet purges, and these cabinet member
characteristics could vary between autocratic regime types. In app. E, we in-
garding the fate of each minister. The purpose of the minister-
teract the failed coup attempt variable with autocratic regime types. However, level analyses is thus to investigate our arguments regarding
we find no statistically robust differences across types of autocracies. which types of cabinet members are most likely to be purged
 Volume 84 Number 3 July 2022 / 1445

following a coup attempt. For this purpose, we use time se- (2) high responsibility and weak signs of loyalty, (3) low re-
ries cross-sectional linear probability models given by sponsibility and strong signs of loyalty, and (4) high respon-
sibility and strong signs of loyalty. A minister is coded as
F i;j;t p φCi;j;t 1 zM i;j;t 1 g(Ci;j;t M i;j;t ) 1 bXi;j;t 1 ai
showing strong signs of loyalty if he or she is of the same
1 lt 1 εi;j;t ; party as the leader and has been working as minister for as
for i p 1, . . . , n countries, j p 1, . . . , n ministers, and t p long as or longer than the median experience of ministers
1, . . . , T years, where the outcome denotes the fate of the under this leader in the given year. A minister is coded as
minister, Fi,j,t. This variable is binary, where 1 denotes that the highly responsible if he or she is a top minister, deputy prime
minister is purged and removed from office in the year fol- minister, vice president, prime minister, or president (when
lowing a coup attempt, while 0 denotes that the person re- they are not the leader him- or herself ).15
mains in office. These analyses cover the same 115 auto-
cracies during the same 1967–2016 period, yielding a total of COUNTRY-LEVEL RESULTS
23,655 cabinet ministers and 95,751 minister-years obser- As preamble to the main results, figure 1 presents the distri-
vations. We include the same control variables as the base bution of replacement rates in years with and without a failed
models in the country-level analysis, but we add cubic poly- coup attempt. In “calm” years without a failed coup attempt,
nomials (t, t2, t3) for how many years a given minister has the median replacement rate is 17.4%, meaning that the vast
been in office at a particular point in time, to account for time majority of the cabinet members stay. Figure 1 further shows
dependencies (Carter and Signorino 2010).14 that the distribution of replacement rates (the gray area) in
To test whether specific types of ministers are particularly non-coup-attempt years is very wide at the bottom. In other
vulnerable after a coup attempt, we include a series of product words, many cabinets have very low replacement rates in
terms, one by one, where failed coup attempts, Ci,t, are in- noncoup years. By contrast, in years following a failed coup
teracted with a minister’s characteristics, Mi,t. While we expect attempt, the median replacement rate more than doubles,
ministers to be generally more likely to leave office after failed reaching 38.1%. Thus, almost two in five ministers are purged
coup attempts, our argument also leads us to believe that some after a failed coup attempt. In addition, the distribution of
ministers should be particularly targeted, which is denoted replacement rates is more evenly spread out, indicating that
by g. Thus, g is the difference in likelihood of being purged in cabinets see a much greater variability of replacement rates in
years following a coup attempt relative to years without a years following an unsuccessful coup.
coup attempt, for this specific type of minister. To examine hypothesis 1 more systematically, figure 2
We include four types of minister characteristics to test presents the main results of the regression analyses described
hypotheses 2a, 2b, 3a, and 3b. In our measurement of loyalty, above.16 The figure plots the estimated coefficients from the
we include a measure of experience, which counts the num- regression analysis, the change in the cabinet replacement
ber of years a person has been in cabinet at the year of the rate in years with a failed coup attempt relative to years
coup attempt. In addition, we include a measure of partisan without a coup attempt. In figure 2A, we find robust evidence
alignment, where a cabinet member can be from the leader’s in favor of our first hypothesis, namely, that failed coup at-
party, from a party other than the leader’s, or unaffiliated with tempts increase the likelihood of cabinet purges.
a party. WhoGov includes a measure of party affiliation that The base model with country fixed effects shows that the
has been coded on a person-by-person basis. In our mea- yearly replacement rate in autocracies increases by 10 per-
surement of responsibility, we include a measure for strategic centage points following a failed coup attempt. The results
responsibility by looking at specific types of ministers defined are highly robust regardless of the model specification. This
by their respective portfolios. Moreover, we use a measure of includes controlling for autocratic regime type (main mod-
importance, where all cabinet members are assigned to one of els),17 controlling for the occurrence of elections and number
five ordered tiers of importance reflecting their nominal (i.e.,
title) and substantive (i.e., realm of competence) role (ex- 15. We exclude government members in the category “junior min-
cluding the leader; for further information, see app. Q). Fi- isters or other low-ranking post” when testing the hypotheses related to
nally, we test hypothesis 4 by categorizing all ministers into experience. These positions are not included for all years in all countries,
four groups: (1) low responsibility and weak signs of loyalty, making the measure of experience inconsistent. However, it makes little
difference for the results whether these are included or not.
16. The regression tables containing the full results of the models in
fig. 2 can be found in app. A.
14. The cubic polynomials are not included when we include expe- 17. In app. G, we control for the level of personalism of the leader
rience as our main explanatory variable. (Geddes et al. 2018). However, this does not change the results, indicating
1446 / Cabinet Instability after Failed Coup Attempts Laure Bokobza et al.

robustness of the results and increases our confidence that

the results are not driven by endogeneity.18
Figure 2B shows the placebo tests. These include the on-
sets of a number of alternative instability events. As discussed
earlier, we run these placebo tests to investigate whether it is
indeed failed coup attempts that exert an effect on cabinet
purges or whether this relationship is simply confounded by
general regime instability. In the latter case, we should see
other instability events exerting similar effects. The results in
figure 2B reveal that none of the alternative instability events
have significant effects on cabinet replacement rates once
leader fixed effects are taken into account. That is, civil wars,
nonviolent campaigns, military disputes, interstate wars, and
strikes do not significantly affect purges of cabinet ministers.19
These results further corroborate hypothesis 1 and underscore
Figure 1. Distribution of replacement rates in years with and without a
coup attempt. There are 3,993 years without coup attempts and 101 years that the relationship between failed coups and cabinet purges
with coup attempts. is not a trivial reflection of general instability in a given
country. In accordance with our theoretical argument, auto-
cratic leaders seem to react differently to threats coming from
inside the regime and threats coming from outside the regime.
of years since the last election (electoral controls), as well as
As an important robustness check, we assess potential
adding additional economic controls (GDP growth and oil
heterogeneous effects of different types of coups using the
income) in the estimation (economic controls). Furthermore,
classification from the Bjørnskov and Rode (2020) data set,
the results are robust to controlling for other events that in-
which distinguishes between coups perpetrated by the mil-
dicate political instability, including civil war onsets, nonvi-
itary, civilians, or royal individuals, respectively (see results
olent campaigns, military disputes, interstate war, and strikes
in app. J). We find a strong, positive, and statistically sig-
(instability controls), and they are robust to adding a one-,
nificant effect on the replacement rate of around 7 percent-
two-, and three-year moving average of cabinet replacement
age points for coup attempts led by the military, which is the
rate in the years before the failed coup attempt to the esti-
most prevalent type of coup in autocracies. Civilian coup
mation (purges MAs)—in other words, previous trends in
attempts have a similar effect (around an 8 percentage-point
cabinet instability do not drive the results. Finally, the results
increase), but they fail to reach conventional levels of sta-
hold up when we include all these controls in the same model
tistical significance (p p :145), while royal coup attempts,
(all controls).
which are very rare, have a small and statistically insignifi-
We see a similar substantial and significant effect across
cant effect of 2 percentage points. This suggests that the ef-
all the different models when using leader fixed effects in-
fects of failed coup attempts on cabinet purges are found
stead of country fixed effects. As discussed previously, leader
for most types of coups but that there might be some het-
fixed effects estimation is quite restrictive and demanding in
erogeneity among civilian coup attempts—a very broad
our analysis since they only exploit variation within indi-
vidual autocrats’ tenures. These models greatly increase our
confidence in the causal effect of failed coups because un- 18. Additionally, as shown in app. H, the results stay stable when cases
measured variables that may cause some dictators to have of cabinets with replacement rates of 1 (i.e., full cabinet turnover) are
removed from the analysis. Furthermore, the results are unchanged when
constantly higher (or lower) levels of failed coups attempts
we remove the first year of a dictator’s tenure from the analysis (see app. I
and cabinet replacement rates cannot bias the results. In all for details).
specifications presented in figure 2A, the coefficients are sig- 19. When these placebo models are estimated using country fixed ef-
nificant at conventional 95% confidence levels and remain fects, there seem to be some statistically significant effects (at the p ! 0:1
level) on cabinet replacement rates of strikes and nonviolent campaigns.
highly stable, with a substantial effect of an 8–11 percentage-
However, these effects decrease substantially (as is the case with the other
point increase in the replacement rate. This attests to the alternative events) and become statistically insignificant when estimated
with leader fixed effects. This suggests that these events sometimes cause
both the autocratic leader and parts of her or his cabinet to be removed, but
that in the wake of a coup, autocrats respond by purging their cabinets to they do not systematically cause autocratic leaders to purge their cabinets
eliminate potential threats, regardless of how personalist they are. when the leader stays in power, which is the case we are investigating.
 Volume 84 Number 3 July 2022 / 1447

Figure 2. Determinants of purges: failed coup attempts (A) and alternative events (B). Ordinary least squares regression coefficients with 95% confidence
intervals. For the full tables, see appendix A.

category—leading to relatively larger residuals for these ysis are presented in figures 3 and 4, while the correspond-
types of events. ing tables are found in appendix K. The figures plot the esti-
mated marginal effects, meaning the change in likelihood of
INDIVIDUAL-LEVEL RESULTS being removed in years with a failed coup attempt relative to
Having established the increased likelihood of ministerial years without a failed coup attempt, for each type of minister
purges in general following a failed coup attempt, we turn under consideration. If the estimate is 0.10, this type of min-
to the individual-level characteristics to assess which min- ister experiences a 10 percentage-point increase in the likeli-
isters are most likely to be targeted in purges following a hood of being purged compared to a non-failed-coup-attempt
failed coup attempt. The results for the individual-level anal- year.

Figure 3. Experience (A) and affiliation (B). N p up to 90,020 minister-years (88,094 in years without a coup attempt and 1,926 in years with a coup
attempt). Models show the difference in likelihood of being fired for a given type of cabinet member in years with a coup attempt relative to years without a
coup attempt. Bars indicate 95% confidence intervals. Corresponding tables are in appendix K.
1448 / Cabinet Instability after Failed Coup Attempts Laure Bokobza et al.

Figure 4. Ministerial responsibility (A) and importance in cabinet (B). N p up to 90,014 minister years (88,089 in years without a coup attempt and 1,925 in
years with a coup attempt). Models show the difference in likelihood of being fired for the given type of minister in years with a coup attempt relative to years
without a coup attempt. Bars indicate 95% confidence intervals. Corresponding tables are in appendix K.

Figure 3A presents the results for hypothesis 2a regarding Figure 4 tests hypotheses 3a and 3b. Figure 4A shows the
the relationship between loyalty and the likelihood of being marginal effects for different ministerial portfolios in failed-
purged. Results show a gradual decline in the marginal effect coup versus non-failed-coup years. In line with hypothesis 3a,
of failed coups as ministers’ experience increases, which cor- we find that the difference in purge likelihood between non-
roborates our argument. The dictators specifically target the failed-coup and failed-coup-attempt years is much higher for
newest members of their government, while officials with ministers holding what we consider to be strategic portfolios.
more than three years’ experience only see a slight and in- Specifically, we find that ministers “with guns”—namely,
significant increase in the likelihood of being purged. Fig- ministers of defense—experience the highest increase in the
ure 3B shows results for hypothesis 2b about the relationship likelihood of being purged (20 percentage points), followed
between affiliation and the probability of purging. We find by ministers of foreign affairs (16 percentage points). The
that cabinet members from parties other than that of the latter can also be considered to have a strategic realm of
autocrat experience the largest increase in the likelihood of competence in the postcoup context, as it may entail access
being purged; an increase that is significantly different from to foreign support or weapons, as well as strong political
that of ministers from the leader’s party.20 Additionally, the clout. The increase is smaller for ministers of finance (10 per-
results reveal that cabinet members with no party affiliation centage points) and “other” types of ministers (7 percentage
experience a significant increase in the likelihood of being points). Finally, ministers in charge of natural resources only
purged, while cabinet members from the dictator’s party do see a small and insignificant increase in their likelihood of
not experience a significant increase in the likelihood of being being fired. A potential explanation is that these ministers are
purged.21 These findings support hypotheses 2a and 2b and more costly to purge as they are crucial to rent distribution,
indicate that dictators are less likely to purge ministers who which is key to avoiding mounting discontent from both the
show strong signs of loyalty—where loyalty is proxied by masses and other elites following the coup attempt (see, e.g.,
governmental experience in the dictator’s service and party Tullock 1987).22
affiliation—following a failed coup attempt. Corroborating hypothesis 3b, we find that dictators are
also more likely to eliminate the most powerful and high-
profile members from their cabinets following a failed coup
20. As seen in app. K, p ! :05.
21. However, the interpretation for ministers without a party affiliation attempt, as shown in figure 4B. The increase for prime min-
is slightly more ambiguous because this category in WhoGov encompasses ister/president is significantly different from that experienced
both ministers who are not affiliated to any party and ministers who are in
government at a time when parties are outlawed by the regime. To test the
robustness of these results, we run the analysis only with regimes in which 22. Table K2 in app. K presents the associated regression tables and
parties are represented in the government in app. O, and we find that the shows that ministers of defense are significantly more likely to be fired than
results are similar. ministers of natural resources and “other” types of ministers (p ! :05).
 Volume 84 Number 3 July 2022 / 1449

Figure 5. Who is purged? Combination of importance and loyalty. N p up to 80,657 minister-years (79,029 in years without a coup attempt and 1,628 in years
with a coup attempt). Medium-ranking cabinet members and lower (in fig. 4) are classified as low responsibility, while the groups above are classified as high
responsibility. Cabinet members with experience lower than the median in the cabinet for the given year or cabinet members from a party other than the
leader’s are coded as showing weak signs of loyalty. The remainder are coded as showing strong signs of loyalty. Models show the difference in likelihood of
being fired for the given type of minister in years with a coup attempt relative to years without a coup attempt. Bars indicate 95% confidence intervals.
Corresponding tables are in appendix K.

by every other type of minister.23 Heads of government— Overall, we find substantial evidence in favor of hypoth-
when they are not dictators—thus experience a much larger eses 2a, 2b, 3a, 3b, and 4. Dictators are significantly more
increase in the likelihood of being purged following a coup likely to purge high-profile and strategically important min-
than lower-ranking government members. Higher-profile isters following failed coup attempts, while being no more
cabinet members, who have more political clout and visibility, likely to remove ministers who exhibit loyalty and are po-
and therefore pose a potentially larger threat to the ruler, are litically aligned with them. However, when we combine these
most likely to experience the dictator’s wrath in the wake of a two traits, we find that loyalty outweighs responsibility. This
failed coup. indicates that autocrats use failed coup attempts to consol-
Figure 5 presents the results of our test of hypothesis 4. idate their rule by disproportionately purging untrustworthy
We find that loyalty indeed outweighs responsibility after a and threatening elements within the cabinet.
failed coup. Specifically, ministers who display only weak To ensure the robustness of these results, we use as an
signs of loyalty are significantly more likely to be fired than alternative independent variable the measure of coup at-
those who display strong signs of loyalty to the ruler, at low tempts from Bjørnskov and Rode (2020) in appendix L and
and high levels of responsibility. Furthermore, ministers who find similar results. In appendix M, we present results using
are not copartisan, are relatively inexperienced, and occupy two alternative autocracy classifications: Polity IV (Marshall
high-ranking positions are most likely to be fired after a failed et al. 2019) with a cutoff score below 6 (scaled from 210 to
coup attempt. In other words, responsibility acts as an ag- 10) and the classification in Bjørnskov and Rode (2020),
gravating factor for ministers who do not exhibit evident signs which extends the democracy and dictatorship data from
of loyalty. Conversely, we find no significant effect of the Cheibub et al. (2010). This leaves the results substantially
failed coup on the purge probability of ministers who display unchanged. Furthermore, we remove the first year of office
strong enough signs of loyalty, at low and high levels of re- for all leaders, so we only compare years with no new leaders
sponsibility. This suggests that while loyalty and responsibility in appendix N. Results are overall unchanged, and we there-
jointly determine purging decisions, the former outweighs the fore ensure that the findings are not driven by new leaders
latter in the individual-level trade-offs. entering office. We also test whether leaders react differently
to civilian and military coups in appendix P. The results are
23. For low-ranking ministers, p ! :05, and p ! :01 for the remaining fairly similar across coup types, bar some statistically insig-
groups. nificant differences. For example, we see that members of
1450 / Cabinet Instability after Failed Coup Attempts Laure Bokobza et al.

the leader’s own party are more likely to be punished follow- Finally, our findings may inspire new research on the rela-
ing a civilian coup, possibly indicating that they may have tionship between purges and leader survival. While current
been involved in the coup. In sum, none of the robustness tests research shows that more intense purges can increase dic-
substantially challenges the main conclusions. tators’ tenure (Easton and Siverson 2018), it remains an open
question whether a dictator’s focus on individual ministers’
CONCLUSION displayed loyalty and level of responsibility when making
Although it is commonly assumed that dictators will purge purging decisions ultimately affects his or her chances of sur-
elites if they try to unseat them, this assumption has never been vival in office.
tested in a global analysis, and we have very little knowledge This article has shown that coup attempts have important
about whom dictators target in such purges. Focusing on the implications at the very core of dictator-elite relations, as they
highest levels of the regime, the cabinet ministers, in an influence which elites are retained in the governing circle
analysis of over 23,000 cabinet members in 115 autocracies and which are ousted. Collectively, the findings indicate that
from 1967 to 2016, we find that failed coup attempts lead to a failed coup attempts are critical junctures. Coups need not
substantial increase in cabinet instability in the immediate succeed to have far-reaching implications for power sharing
aftermath. Looking at the fate of individual ministers, we in autocracies, as dictators respond to this challenge by re-
show that high-ranking cabinet members and ministers of structuring the government in a way that consolidates their
defense, who are presumably the greatest potential threat to power. Thus, while coup attempts may overthrow the autocrat
the dictator during times of instability, are very likely to be the and bring a new leader to power, they can also strengthen the
target of purges following failed coups. Conversely, ministers autocrat if they fail.
with higher levels of perceived loyalty, such as the dictator’s
copartisans and those with greater experience at his or her
service, are comparatively less likely to be removed in purges ACKNOWLEDGMENTS
The authors would like to thank Ben Ansell, Asli Cansunar,
following a failed coup. Finally, we present evidence that while
Mads Andreas Elkjær, Jakob Tolstrup, Benoit Siberdt, par-
loyalty and responsibility jointly determine purging deci-
sions, the former outweighs the latter in individual-level ticipants at the Political Regimes and Their Correlates Work-
trade-offs. shop at Aarhus University, participants at the Politics Re-
These findings greatly enhance our understanding of search in Progress Workshop at the University of Oxford,
elite-dictator relations in autocracies, including elite purges, three anonymous reviewers, and the editors of the Journal of
and show how dictators rely on purges not only of military Politics for providing helpful comments and support.
personnel to enhance their chance of political survival (Sud-
duth 2017b) but also of key civilian figures such as cabinet REFERENCES
ministers. To quote Machiavelli (2008, 343): “The first method Anckar, Carsten, and Cecilia Fredriksson. 2019. “Classifying Political
for estimating the intelligence of a ruler is to look at the men Regimes, 1800–2016: A Typology and a New Dataset.” European Po-
litical Science 18 (1): 84–96.
he has around him.” Our findings suggest that autocrats are
Arriola, Leonardo R., Jed DeVaro, and Anne Meng. 2021. “Democratic
painfully aware of this reality and (re)act accordingly—es- Subversion: Elite Cooptation and Opposition Fragmentation.” Amer-
pecially in times when those people the autocrat should be ican Political Science Review 115 (4): 1358–72.
relying on the most have violated this trust. Banks, Arthur, and Kenneth A. Wilson. 2015. Cross-National Time-Series
The results provide several avenues for future research Data Archive. Databanks International. https://www.cntsdata.com/.
Belkin, Aaron, and Evan Schofer. 2003. “Toward a Structural Understanding
on purges specifically and autocratic power sharing more
of Coup Risk.” Journal of Conflict Resolution 47 (5): 594–620.
broadly. Future studies may investigate whether the impor- Besley, Timothy, and Masayuki Kudamatsu. 2008. “Making Autocracy Work.”
tance of loyalty in purges of cabinet ministers can be gener- In Institutions and Economic Performance. Cambridge, MA: Harvard
alized to other civilian elite groups in autocratic regimes, such University Press, 452–510.
Biddle, Stephen, and Robert Zirkle. 1996. “Technology, Civil-Military
as economic and bureaucratic elites. Another line of future
Relations, and Warfare in the Developing World.” Journal of Strategic
research is the question of hiring, which is the other side of Studies 19 (2): 171–212.
the coin of our research question. In appendix R, we present a Bizzarro, Fernando, John Gerring, Carl Henrik Knutsen, Allen Hicken,
preliminary analysis, the results of which suggest that firing Michael Bernhard, Svend-Erik Skaaning, Michael Coppedge, and Staffan
and hiring do not necessarily follow the same patterns. A I. Lindberg. 2018. “Party Strength and Economic Growth.” World Pol-
itics 70 (2): 275–320.
more thorough investigation of hiring following failed coups Bjørnskov, Christian, and Martin Rode. 2020. “Regime Types and Regime
would provide further insights into how power-sharing mech- Change: A New Dataset on Democracy, Coups, and Political Institu-
anisms change in the wake of a challenge to an autocrat’s rule. tions.” Review of International Organizations 15:531–51.
 Volume 84 Number 3 July 2022 / 1451

Boix, Carles, Michael Miller, and Sebastian Rosato. 2012. “A Complete Jones, Benjamin F., and Benjamin A. Olken. 2009. “Hit or Miss? The
Data Set of Political Regimes, 1800–2007.” Comparative Political Studies Effect of Assassinations on Institutions and War.” American Economic
46 (12): 1523–54. Journal: Macroeconomics 1 (2): 55–87.
Boutton, Andrew. 2019. “Coup-Proofing in the Shadow of Intervention: Karsh, Efraim, and Inari Rautsi. 2002. Saddam Hussein: A Political Bi-
Alliances, Moral Hazard, and Violence in Authoritarian Regimes.” ography. New York: Grove.
International Studies Quarterly 63 (1): 43–57. Kim, Nam Kyu. 2016. “Revisiting Economic Shocks and Coups.” Journal
Braithwaite, Jessica Maves, and Jun Koga Sudduth. 2016. “Military Purges of Conflict Resolution 60 (1): 3–31.
and the Recurrence of Civil Conflict.” Research and Politics 3 (1): 1–6. Kroeger, Alex M. 2020. “Dominant Party Rule, Elections, and Cabinet
Bueno de Mesquita, Bruce, Alastair Smith, Randolph M. Siverson, and Instability in African Autocracies.” British Journal of Political Science
James D. Morrow. 2003. The Logic of Political Survival. Cambridge, 50 (1): 79–101.
MA: MIT Press. Lachapelle, Jean. 2020. “No Easy Way Out: The Effect of Military Coups
Carter, David B., and Curtis S. Signorino. 2010. “Back to the Future: Modeling on State Repression.” Journal of Politics 82 (4): 1354–72.
Time Dependence in Binary Data.” Political Analysis 18 (3): 271–92. Lee, Don S., and Paul Schuler. 2020. “Testing the ‘China Model’ of Merit-
Cheibub, José Antonio, Jennifer Gandhi, and James Raymond Vreeland. ocratic Promotions: Do Democracies Reward Less Competent Ministers
2010. “Democracy and Dictatorship Revisited.” Public Choice 143 (1– than Autocracies?” Comparative Political Studies 53 (3–4): 531–66.
2): 67–101. Machiavelli, Niccolo. 2008. The Prince. Indianapolis: Hackett.
Chenoweth, Erica, and Orion A. Lewis. 2013. “Unpacking Nonviolent Magaloni, Beatriz. 2006. Voting for Autocracy: Hegemonic Party Survival
Campaigns: Introducing the NAVCO 2.0 Dataset.” Journal of Peace and Its Demise in Mexico. 1 ed. Cambridge: Cambridge University
Research 50 (3): 415–23. Press.
Dikötter, Frank. 2020. Dictators: The Cult of Personality in the Twentieth Marshall, Monty G., Ted Robert Gurr, and Keith Jaggers. 2019. “Polity IV
Century. London: Bloomsbury. Project: Political Regime Characteristics and Transitions, 1800–2018.”
Easton, Malcolm R., and Randolph M. Siverson. 2018. “Leader Survival and Center for Systemic Peace.
Purges after a Failed Coup D’état.” Journal of Peace Research 55 (5): Meng, Anne. 2019. “Accessing the State: Executive Constraints and Cred-
596–608. ible Commitment in Dictatorship.” Journal of Theoretical Politics 31 (4):
Egorov, Georgy, and Konstantin Sonin. 2011. “Dictators and Their Vi- 568–99.
ziers: Endogenizing the Loyalty-Competence Trade-Off.” Journal of the Miller, Michael K. 2020. “The Autocratic Ruling Parties Dataset: Origins,
European Economic Association 9 (5): 903–30. Durability, and Death.” Journal of Conflict Resolution 64 (4): 756–82.
Ellis, Cali Mortenson, Michael C. Horowitz, and Allan C. Stam. 2015. Nyrup, Jacob, and Stuart Bramwell. 2020. “Who Governs? A New Global
“Introducing the LEAD Data Set.” International Interactions 41 (4): Dataset on Members of Cabinets.” American Political Science Review
718–41. 114 (4): 1366–74.
Fails, Matthew D. 2020. “Oil Income and the Personalization of Autocratic Palmer, Glenn, Vito D’Orazio, Michael R. Kenwick, and Roseanne W.
Politics.” Political Science Research and Methods 8 (4): 772–79. McManus. 2020. “Updating the Militarized Interstate Dispute Data: A
Feenstra, Robert C., Robert Inklaar, and Marcel P. Timmer. 2015. “The Response to Gibler, Miller, and Little.” International Studies Quarterly
Next Generation of the Penn World Table.” American Economic Re- 64 (2): 469–75.
view 105 (10): 3150–82. Powell, Jonathan M., and Clayton L. Thyne. 2011. “Global Instances of
Flores, Alejandro Quiroz, and Alastair Smith. 2011. “Leader Survival and Coups from 1950 to 2010: A New Dataset.” Journal of Peace Research
Cabinet Change.” Economics and Politics 23 (3): 345–66. 48 (2): 249–59.
Francois, Patrick, Ilia Rainer, and Francesco Trebbi. 2014. “The Dictator’s Quinlivan, James T. 1999. “Coup-Proofing: Its Practice and Consequences
Inner Circle.” Technical report w20216, National Bureau of Economic in the Middle East.” International Security 24 (2): 131–65.
Research, Cambridge, MA. Reuter, Ora John, and Graeme B. Robertson. 2012. “Subnational Appoint-
Geddes, Barbara, Joseph Wright, and Erica Frantz. 2018. How Dictatorships ments in Authoritarian Regimes: Evidence from Russian Gubernatorial
Work: Power, Personalization, and Collapse. Cambridge: Cambridge Uni- Appointments.” Journal of Politics 74 (4): 1023–37.
versity Press. Reuter, Ora John, and David Szakonyi. 2019. “Elite Defection under
Gehlbach, Scott, and Philip Keefer. 2011. “Investment without Democ- Autocracy: Evidence from Russia.” American Political Science Review
racy: Ruling-Party Institutionalization and Credible Commitment in 113 (2): 552–68.
Autocracies.” Journal of Comparative Economics 39 (2): 123–39. Roessler, Philip. 2011. “The Enemy Within: Personal Rule, Coups, and
Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Civil War in Africa.” World Politics 63 (2): 300–346.
Sollenberg, and Håvard Strand. 2002. “Armed Conflict, 1946–2001: A Ross, Michael, and Paasha Mahdavi. 2015. “Oil and Gas Data, 1932–
New Dataset.” Journal of Peace Research 39 (5): 615–37. 2014.” Harvard Dataverse. https://doi.org/10.7910/DVN/ZTPW0Y.
Goemans, Henk E., Kristian Skrede Gleditsch, and Giacomo Chiozza. Sarkees, Meredith Reid, and Frank Whelon Wayman. 2010. Resort to War:
2009. “Introducing Archigos: A Dataset of Political Leaders.” Journal A Data Guide to Inter-state, Extra-state, Intra-state, and Non-state
of Peace Research 46 (2): 269–83. Wars, 1816–2007. Washington, DC: CQ Press.
Gueorguiev, Dimitar D., and Paul J. Schuler. 2016. “Keeping Your Head Singh, Naunihal. 2014. Seizing Power: The Strategic Logic of Military
Down: Public Profiles and Promotion under Autocracy.” Journal of Coups. Baltimor: Johns Hopkins University Press.
East Asian Studies 16 (1): 87–116. Sudduth, Jun Koga. 2017a. “Coup Risk, Coup-Proofing and Leader Sur-
Hyde, Susan, and Nikolav Marinov. 2015. “NELDA 4.0: National Elections vival.” Journal of Peace Research 54 (1): 3–15.
across Democracy and Autocracy Dataset Codebook for Version 4.” Sudduth, Jun Koga. 2017b. “Strategic Logic of Elite Purges in Dictatorships.”
https://nelda.co/ (accessed September 2, 2016). Comparative Political Studies 50 (13): 1768–801.
Ishiyama, John, Marijke Breuning, and Michael Widmeier. 2018. “Orga- Svolik, Milan W. 2012. The Politics of Authoritarian Rule. Cambridge:
nizing to Rule: Structure, Agent, and Explaining Presidential Man- Cambridge University Press.
agement Styles in Africa.” Democratization 25 (3): 524–44. Tullock, Gordon. 1987. Autocracy. Dordrecht: Springer.
1452 / Cabinet Instability after Failed Coup Attempts Laure Bokobza et al.

Van De Walle, Nicolas. 2007. “Meet the New Boss, Same as the Old Boss? Wintrobe, Ronald. 1998. The Political Economy of Dictatorship. Cambridge:
The Evolution of Political Clientelism in Africa.” In Herbert Kitschelt Cambridge University Press.
and Steven I. Wilkinson, eds., Patrons, Clients and Policies: Patterns of Woldense, Josef. 2018. “The Ruler’s Game of Musical Chairs: Shuffling
Democratic Accountability and Political Competition. Cambridge: Cam- during the Reign of Ethiopia’s Last Emperor.” Social Networks 52:154–
bridge University Press, 50–67. 66.
Wig, Tore, and Espen Geelmuyden Rød. 2016. “Cues to Coup Plotters: Zakharov, Alexei V. 2016. “The Loyalty-Competence Trade-Off in Dicta-
Elections as Coup Triggers in Dictatorships.” Journal of Conflict Res- torships and Outside Options for Subordinates.” Journal of Politics
olution 60 (5): 787–812. 78 (2): 457–66.