· contents 1 letting sleeping dogs lie . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....

Monika Nalepa

Ritual Sacrifices?

Transitional Justice and the Fate ofPost-Authoritarian Democracies

June 14, 2020

Contents

1 Letting Sleeping Dogs Lie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.1 Neither retribution, nor reconciliation . . . . . . . . . . . . . . . . . . . . . . 61.2 Who goes into politics, what do they do while in office, and

how do they delegate? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.3 The contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141.4 A novel approach to collecting TJ data . . . . . . . . . . . . . . . . . . . . . . 161.5 Roadmap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2 Blackmail and Transparency Regimes . . . . . . . . . . . . . . . . . . . . . . . . . . 252.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.2 The Formal Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.2.1 Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312.2.2 Sequence of play . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332.3.1 Quality of representation under the three equilibrium

types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362.3.2 Potential criticism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3 Purging the Authoritarian State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433.1 Illustrative examples of purges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3.2.1 The decision to purge as a simple delegation model . . . 543.2.2 Discretionary limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.2.3 Modeling the value of expertise . . . . . . . . . . . . . . . . . . . . . 59

3.3 Discussion and comparative statics . . . . . . . . . . . . . . . . . . . . . . . . . 653.3.1 Contrast with Representation under the Lustration

Blackmail Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 663.5 Conclusion to Part I: Normative implications . . . . . . . . . . . . . . . . 67

VI Contents

4 Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 794.1 Measuring the quality of representation . . . . . . . . . . . . . . . . . . . . 794.2 Measuring Transitional Justice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

4.2.1 Why do we need a new Transitional Justice dataset? . . . 874.2.2 Taking Time Seriously: Transitional Justice as a time

series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 904.3 Constructing measures of Transitional Justice . . . . . . . . . . . . . . . 924.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

5 Lustration and Programmatic Representation . . . . . . . . . . . . . . . . . . . 1035.1 How does lustration affect who wins office and how they

behave? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1035.1.1 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

5.2 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

6 Truth Commissions and the Quality of Democracy . . . . . . . . . . . . . 111

7 Purges and the Quality of Democracy . . . . . . . . . . . . . . . . . . . . . . . . . . 113

8 Conclusion: Beyond Ritual Sacrifices . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

9 Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1179.1 Proofs to Chapter 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

9.1.1 Players, Preferences, strategies and Beliefs . . . . . . . . . . . . 1189.1.2 Proposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1189.1.3 Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1199.1.4 Pure separating equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . 1199.1.5 Pure pooling equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

9.2 Proofs to Chapter 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1229.2.1 Baseline model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1229.2.2 Model with discretion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1239.2.3 Model with imperfect expertise . . . . . . . . . . . . . . . . . . . . . . 124

9.3 Quantitative Appendix to chapter 5 . . . . . . . . . . . . . . . . . . . . . . . . 1269.3.1 Quality of Representation (cosal 3) . . . . . . . . . . . . . . . . . . . 1269.3.2 Robustness checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

9.4 GIS appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

References Cited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

Contents 1

Summary. Transitional Justice–that is–the act of reckoning with a former authoritar-ian regime after it has ceased to exist, has direct implications for democratic processes.It influences who decides to go into politics, shapes politicians’ behavior while in of-fice, and finally, influences how politicians delegate policy decisions. Mechanisms oftransitional justice, far from being the epilogue of an outgoing authoritarian regime,are thus a constitutive part of the new democratic order. How successful democra-cies become at remaining democratic and the overall quality of these democracies isa direct consequence of transitional justice. Within transitional justice, this book dis-tinguishes between transparency mechanisms—the revelation of authoritarian lega-cies that were concealed—and purges—the firing of open collaborators of the ancienregime. Analysis of this distinction throws into stark relief the contrasting effects thesemechanisms have on sustaining and shaping the quality of democratic processes.While transparency regimes unequivocally reduce corruption and improve the pro-grammatic representation political parties can offer, the effect of purges is more am-biguous and contingent on features of the former authoritarian regime and currentcircumstances. At the same time, not all new states are equally well-positioned to carryout purges in the first place. Hence, the availability of transitional justice as remedy isnot uniform. A highly disaggregated global transitional justice dataset allows readersto evaluate these theoretical claims and confirms that the choice between democracyand justice that scholars have for decades presented to newly transitioned societies ismisconstrued.

1

Letting Sleeping Dogs Lie

I present to you a cabinet that bears no responsibility for the mortgage it is inheriting today.Even though it affects the circumstances in which we must operate, we are separating ourselvesfrom the past with a thick line

—-Tadeusz Mazowiecki in his 1989 expose (Gostkiewicz 2013)

After Generalissmo Francisco Franco died, the elites who succeeded him resolvedon behalf of the Spanish people to let bygones be bygones. Formally, in 1977, theypassed an Amnesty Law; informally, they agreed to a “pact of forgetting.” The ra-tionale offered for this deliberate decision was to chose democracy over justice. Al-though the Francoist regime had committed numerous atrocities during the Civil Warof 1936-1939 as well as after the Nationalists’ victory, Spanish elites decided to “sealthe archives” of the Guardia Civil and the Policia Armada (Franco’s secret police1) andnot attempt any reckoning with the past. Even private conversation concerning theCivil War and the authoritarian regime that succeeded it was rendered taboo. This“pact of forgetting” was shared widely by all sides of the political spectrum, includ-ing the communists against whom Franco’s Nationalists had fought in the Civil War.”For instance, Santiago Carillo (General Secretary of the Communist Party at the timeof transition) was quoted as saying: “In our country, there is but one way to reachdemocracy, which is to throw anyone who promotes the memory of the Civil War. Wedo not want anymore wars, we have enough of them already.”2

The informal pact became so entrenched that when in 2007, a Socialist governmenttried to revisit the past by proposing the mildest of transparency measures, includingvictim rehabilitation, it was met with staunch criticism.3 The Spanish example stoodin contrast with its geographic and temporal neighbors, Greece and Portugal, both ofwhich thoroughly purged their former authoritarian leadership and its agencies. Tosee this contrast, one need only look at Figures 1.1 and 1.2 below. Both are based on

1 Both were security agencies designed to preserve the power of the Spanish dictator.The Guardia Civil was mainly active in rural areas, whereas the Policia Armadapatrolled cities and metropolitan areas.

2 See Europe: Painful memories; Spain’s civil war (2006)3 See Europe: A Rude Awakening; Spain’s Past (2007).

4 1 Letting Sleeping Dogs Lie

data from the Global Transitional Justice Dataset.4 Figure 1.2 shows the removal ofleaders associated with the ancien regime in 84 countries that had such leadershippurge events. Most notably, since zeroes have been omitted from these figures, Spainis not even listed among the countries that underwent such purges.

Fig. 1.1: Severity of Thorough Purges

Fig. 1.2: Severity of Leaderhip Purges

Incidentally, neither Greece nor Portugal fared as well recovering from their au-thoritarian pasts as Spain. By the early 1990’s, Spain had risen to be one of the leadingnations in the European Community, with a GDP per capita of almost 68% of that of

4 The details on how this dataset was prepared are explained later in this chapter aswell as in Chapter 4.

1 Letting Sleeping Dogs Lie 5

the US in 1991 (based on purchasing power parity and the IMF). 5 and a Polity IV scoreof 10. Scholars of democratization overwhelmingly agree that Spain consolidated be-cause of its consensus surrounding letting bygones be bygones rather than despite it.(Przeworski 1991, Linz et al. 1978, Encarnacion 2008)

In light of this, perhaps unsurprisingly, when a wave of democratization spreadacross Eastern Europe twenty years later, Spain’s approach to reckoning with the au-thoritarian past was used as a model for (not) dealing with the legacies of commu-nism. In Poland, this approach was summarized with the term “gruba kreska” (“thickline”). Although originally intended to represent the idea of a clean slate for thenew non-communist cabinet, which would only be held accountable for policies im-plemented after assuming office, it quickly came to represent the idea of forgivingthe communists for all human rights violations committed during their 45-year-longtenure in Poland.

The international community of scholars largely applauded this decision. JackSnyder argued that “the prosecution of perpetrators of atrocities according to uni-versal standards risks causing more atrocities than it would prevent, because it paysinsufficient attention to political realities” (Snyder & Vinjamuri 2004, 5). Samuel Hunt-ington maintained that sometimes “amnesty...is necessary to establish a new democ-racy on a solid basis” (Huntington 1993, 214) and that “even if a moral and legal argu-ment could be made for prosecution, this would fall before the normative imperativeof creating a stable democracy.”

This policy was also widely endorsed by a group of social scientists formed aroundthe law journal “The East European Constitutional Review” published in the 1990sfirst at the University of Chicago and later at New York University. Jon Elster andStephen Holmes spearheaded this movement and used Spain’s most benign way ofdealing with former authoritarian collaborators to build their case that “doing noth-ing” is the best approach for new democracies to deal with past authoritarian regimes(Elster 2004a). “Gruba Kreska” and the “Pact of Forgetting” in contrast to “witchhunts” and “ritual sacrifices,” they argued, allow typical political cleavages of left andright to reemerge without the development of a “regime divide” where opponents ofthe previous regime overlook their ideological differences and maintain a united frontagainst successor autocrats, even when these members of the opposition actually shareideological identities with the successor autocrats.6 Delayed democratic consolidationdue to a party system that fails to develop programmatically is but one of the possiblecosts of making transitional justice central to political debates.

Poland and Hungary, which transitioned in short succession, both refrained atleast initially from harsh decommunization. In Poland, President Walesa and his sup-

5 As a point of comparison, Portugal’s GDP per capita relative to that of the US forthe same year was 56%, barely higher than the 52% recorded in 1982, and Greece’swas 60 %, down from 72% in 1980.

6 In the the context of Post-Communist Europe, for reasons explained later, “GrubaKreska” failed and a regime divide did emerge, formally defined by Grzymala-Busse (2001) as the persisting conflict between the successors to the pre-1989 Com-munist parties and the parties emerging from the Communist-era opposition. Thedeeper this divide, the lower the chances of coalition-formation between the Com-munist successors and their opposition counterparts based on shared policy goals.Contours of such a regime divide could also be traced back to Reconstruction-eraUS politics. For a discussion, see Carroll & Nalepa (2014)


porters from the Citizens’ Committee prevented any decommunization bills spon-taneously passed on the legislative floor from being implemented. 7 Latter attemptswere halted by the Constitutional Tribunal, a constitutional court established a fewyears before the transition. In Hungary, the Constitutional Court was also responsiblefor first halting and then significantly delaying attempts to deal with the past.

1.1 Neither retribution, nor reconciliation

The arguments in support of the “pact of forgetting” and ”gruba kreska” cited aboveimplicitly assume that transitional justice (TJ)—the combined set of mechanisms de-signed to deal with past authoritarian legacies—cannot be democracy-enhancing. Thewidely held belief is that TJ can at most promote reconciliation,8 though any broodingin the past comes at the cost of delaying normal democratic processes, normal politi-cal cleavage formation, and party system institutionalization. In this book, I challengethe belief that one can and must choose between democracy and justice. First, I arguethat democracies do not emerge in a vacuum. Legacies of former authoritarian statespermeate the new polity through unsettled scores of human rights abuses, staff of for-mer security agencies, and archives of the former secret police that list the names ofcollaborators of the authoritarian security apparatus.

Not all forms of authoritarian dominance are as transparent as repression (Tyson2016, Nugent 2019, Mattingly 2019, King et al. 2013). In many instances, the veryacts that sustained the authoritarian regime were secret collaboration (Blaydes 2010),cooptation (Magaloni 2006), and sabotage (Dragu & Przeworski 2019). A key as-pect of dealing with the past is the disclosure of such actions and the revelationof the identities of collaborators, spies, and saboteurs. This class of transitional jus-tice mechanisms—transparency regimes and the revealing of who among personsholding public office collaborated with the secret police to the detriment of the anti-authoritarian opposition—is often left out of classical critiques of transitional justice.Yet my first goal in this book is to show that truth revelation mechanisms, such as lus-tration and truth commissions have a critical impact on who selects to enter politics andwhat policies politicians pursue once in office. Such transparency regimes can improvethe quality of representation by bringing policy proposals closer to the preferences ofthe median voter than non-transparency regimes would.

Secret authoritarian legacies have a way of undermining successor regimes evenyears after the transition. For instance, collaborators of the former regime may haveengaged in acts that under a new democratic regime could tarnish their reputation. Iftheses acts remain secret while these former collaborators rise to positions of power,these politicians may be subject to blackmail from those who threaten to release their“skeletons in the closet.”

7 In a spectacular takedown of legislation aiming to reveal the identities of secretpolice agents by the Polish Minister of Interior, Walesa and his allies brought downthe entire cabinet and replaced it with one that vowed to stay clear of dealing withthe past.

8 However, see criticisms of the reconciliation goal that hold that it is too ambitiousand imposes a collectively shared belied about the truth of past events instead ofallowing competing, conflicting narratives of that past Muller (2002).

1.1 Neither retribution, nor reconciliation 7

To illustrate this point, consider the following case from Poland. In February of2016, the widow of Czesław Kiszczak, the former Chief of Communist Poland’s Se-cret Police, discovered a thick secret police file containing evidence that Lech Wałesa,Nobel Peace Prize Laureate and former Solidarity leader, had collaborated with theCommunist Secret Police between 1970 and 1976. The revelation raised serious con-cerns about the quality of Wałesa’s presidency, which had ended before Poland im-plemented its lustration law.9 Although Wałesa’s collaboration preceded his career asSolidarity trade union organizer, Kiszczak couuld have pressured him to avoid im-plementing certain policies by threatening to release the compromising file. This leftmany questioning the extent to which he represented interests of the electorate in-stead of those of the former secret police. More generally, it left scholars wonderingabout the prevalence of such acts of blackmail. How often did former authoritarianelites pressure their former spies into following their policy preferences over those ofthe voters? To the extent that one considers representing voters’ wishes as a marker ofhigh-quality representative democracy, the revelation left many wondering just howrepresentative Polish democracy was of its electorate.

Because transitional justice regulates when and how such skeletons in the closetare released, it has implications for the quality of democracy and the success of thedemocratization project. If democracy survives, damaging information collected bythe former authoritarian secret police for the benefit of authoritarian elites may, ifkept secret, turn elected politicians into clients of agents who threaten to reveal theirskeletons in the closet (Nalepa 2010b). Forgiving and forgetting may sabotage electedpoliticians’ capacity to represent voters, a phenomenon that is hard to pick up on bystudies focusing on the immediate aftermath of a transition. Thus, even if one agreeswith the arguments of Elster and Holmes questioning the validity of the normative(or so-called “backward-looking”) arguments for rehashing the past, exposing thefiles and secrets of the former authoritarian police can undercut the growth of au-thoritarian networks.

According to the argument I present in this book, revealing evidence of humanrights violations and collaboration with members of the ancien regime prevents for-mer authoritarian elites from influencing policy in new democratic polities. Wheretransparency is lacking, former authoritarian elites can pressure politicians into policyconcessions by threatening to reveal compromising information that could jeopardizethese politicians’ careers.

In sum, new democracies do not have the luxury of starting with a blank slate.There are cobwebs of former authoritarian regimes everywhere but while some ofthem are secret, others are perfectly transparent: the old regime relied on the militaryand staff of the state administration. These networks are not secret, but dealing withthem is no less consequential.

All new democracies face the dilemma of bureaucratic crisis: who should run theirstate? On one hand, there are both normative and practical cases to be made for re-moving agents of the ancien regime. At the same time, there is a clear trade-off topurging the state bureaucracy and replacing existing actors with new agents: inexpe-rienced bureaucrats lack the expertise of people who used to run the agencies of the

9 A lustration program would have vetted Wałesa for connections to the communistsecret police. Had it been in effect when Wałesa ran for or held office, his politicalcareer could have ended had he falsely maintained his innocence.


ancien regime. How can new democracies, emerging out of recent authoritarianism,construct a government that will be accountable and proficient?

Consider the case of Tunisia, the trailblazer of countries that transitioned duringthe Arab Spring. Ben Ali’s predecessor, Bourguiba, was famous for appointing all stateadministrators from among loyalists of the authoritarian party, Neo-Dustur. To ensurethat none of them became strong enough, he frequently reshuffled persons at the top.The result of this was a bureaucracy with a very poor skillset. Ben Ali, in contrast,began to professionalize his governing apparatus and enforcement agencies. In sharpjuxtaposition to Ba’athist states in the region, where positions were awarded accordingto partisan status, Ben Ali developed a cadre of professional bureaucrats.

This should in no way distract from Ali’s authoritarian methods. In sustaining hisdictatorship, he relied for support on his police forces. On the other hand, according toSafwan Masri, the army had “neither the power or political will” to quell the protestsagainst the regime (Masri 2017). According to numerous accounts, in the critical mo-ment of the protests, Rachid Ammar (Chief of Staff of the Armed Forces) outrightrefused orders to fire on protesters (Henry 2007). Later during the transition, it wasthe army that would provide cover for the protesters, while security forces under com-mand of the Interior Minister fired on protesters. These political alignments of agen-cies of authoritarian enforcement have consequences for how new democracies maywant to deal with the staff of the previous regime.

Following the transition, the security forces were thoroughly purged by the newdemocratic minister of interior, Rajhi.10 Meanwhile, the army was largely left intact;their loyalties were clearly with the new democratic government. However, the issueof the police is more complex. As later events showed, Tunisia became the leader in theMiddle East in terrorist attacks and supplying fighters to Syria (Macdonald & Wag-goner 2018). Arguably, this happened because of Tunisia’s weakened internal securityforces after the firing of Ben Ali’s enforcement apparatus.

Thus, banning known agents of the ancien regime may, under certain conditions,damage democratic quality. A purge of open collaborators removes agents whose ex-pertise may be of use in the new democratic state. Although such agents may use thisexpertise to advance their own goals, if the are not purged and these goals are trans-parent to the new democrats, there may exist ways of harnessing this expertise in theservice of the new regime.

Tunisia’s case stands in sharp contrasts with that of Bolivia where the military wasactively involved in propping up authoritarian governments from 1964 until 1982. Af-ter the transition, only extremely mild ways of dealing with the past were used. Fol-lowing a general amnesty for political prisoners, the only mechanism for acknowledg-ing the past was a short-lived truth and reconciliation commission. 11 After 2 years,the commission was disbanded one year before the expiration of its mandate. Over adecade later, criminal sentences against one of the military dictators and over forty ofhis collaborators were handed down. The absence of transitional justice could be at-tributed to the strength of the military at the time of transition, but it is not uncommon

10 According to a March 7, 2011 decree issued by the Interior Ministry, the secret policeand security apparatus were abolished “to bolster freedoms and civil rights” andto eradicate “outdated institutions that are vestiges of the regime.”

11 The National Commission for the Investigation of Forced Disappearances was es-tablished by Presidential decree within just days of the transition on October 28,1982.

1.2 Who goes into politics, what do they do while in office, and how do they delegate? 9

for reckoning with the past to take place even when the military is strong (Greece be-ing a leading example here). Moreover, the strong Bolivian military which would haveshielded them from transitional justice, does not explain why purges did not extendto other sectors of the Bolivian state.

A more plausible explanation for the absence of purges in Bolivia is that eventhough the collaborators of the military dictatorship were not ideologically alignedwith the new democratic forces that took over power in Bolivia, they were the onlyones who knew how to operate the state. In other words, they were kept in place be-cause dismissing them would result in loss of valuable expertise.

In sum, while there are normative reasons and clear pressures from public opin-ion to hold accountable the committers of past crimes, such punishment may not befeasible and at times, may even be counterproductive. New democracies may needthe expertise of compromised bureaucrats and agents of the regime, and may thusbe forced to keep in office an agent whose loyalties are with the ancien regime if thisagent can offer much needed expertise. As I argue in this book, the pressing project fornew democracies is to learn to use former bureaucrats when they have usable skills tooffer and when their political alignments do not make them a liability.

In this book, I depart from the normative and backward-looking approach to tran-sitional justice according to which the rationales for engaging in or forgoing these poli-cies are retribution or reconciliation. Instead, I pursue a forward-looking argument.First, I argue that making public the secret information that former autocrats coulduse to influence policy-making in the new democracy, should improve the quality ofrepresentation. By formally reconstructing the blackmail mechanism that transitionaljustice institutions undercut, I uncover circumstances under which politicians delivertheir mandate and avoid pressure from former authoritarian elites. This allows me toanswer the key question: are states that engage in uncovering the secret collaboratorsof the former authoritarian regime better off than states that do nothing? Second, Ilook at the effects of firing staff from agencies of the former authoritarian state andof disbanding state agencies. I point to the fact that such purges are not universallyavailable to new democracies. Some new states need the expertise of ancien regimeagents. Some may take advantage of the fact that preferences of the state agents arenot misaligned with those of the new democrats because they were never reall loyalto the outgoing dictator.

The remainder of this introduction is organized as follows. The next section orga-nizes the concepts that will be used in this book. Then, I describe my contributions tocomparative politics. This book will I explain how TJ is related to issues of selection(Who goes into politics?, how it affects behavior in office (democratic representation), andhow it affects delegation to bureaucrats. The last subsection will offer a roadmap of thebook.

1.2 Who goes into politics, what do they do while in office,and how do they delegate?The mechanisms of dealing with the past described above fall into the broad categoryof transitional justice (TJ), which refers to ways in which new democracies reckonwith their former authoritarian past. According to Kaminski et al. (2006), TJ refersto the “formal and informal procedures implemented by a group or institution of ac-cepted legitimacy around the time of transition out of an oppressive or violent social


order, for rendering justice to perpetrators, and their collaborators, as well as victims.”The first association many readers may have with transitional justice is the criminaltrials of those, who in the name of an authoritarian ideology, committed atrocities.Indeed, the event that founded the very discipline of transitional justice was the trialof Nazi perpetrators in Nuremberg (Teitel 2003). This association is strengthened bythe outstanding efforts of scholars to document trials and amnesties of perpetratorsof human rights violations around the world and to organize them in easily accessibledatasets.12

In this book, my focus is not on trials but on personnel TJ—that is, non-criminalforms of TJ. For democracy to even have a chance to consolidate, and for peace to havea possibility of taking hold, those responsible for human rights violations must at min-imum be removed from office. Trials go further. They actually hold those responsibleaccountable for what they did in the past. Personnel TJ—lustration, truth commis-sions, and purges—can be interpreted as a first step in the direction of accountability.Before trials can take place, the leadership of the former regime must first be removedfrom positions of power.

This book will look closely at two types of mechanisms: purges and transparencyregimes. The key difference between the two is whether new information is uncoveredin the process.

I will call purging the act of banning from office a known offender. Purges comein two forms. First, a purge can be thorough, whereby every member of an institutioncreated by the former authoritarian regime—the security apparatus, the military, thepolice, or the department of justice for example—is fired. Sometimes the entire agencyis disbanded. The East German secret police, popularly referred to as the “Stasi,” waspurged in this way following German Unification in 1990. No former Stasi officer re-tained his or her position. Instead, domestic surveillance operations were taken overby West Germany’s Bundesamt fur Verfassungsschutz, the Federal Office for the Protec-tion of the Constitution (Vilasi 2015).

Instead of disbanding an entire authoritarian agency, one can also limit a purge tothe leadership of that institution or organization. I will refer to this type of purge asa leadership purge. Typically, these policies are termed “decommunization” (Holmes1994),“denazification” (Capoccia 2015), or “de-Baathification” (David 2006). In 2003,De-baathification prevented 185 members of Saddam Hussein’s party from runningfor the legislature (David 2006). An example of a decommunization bill is the Al-banian legislature’s creation of a commission to reassess law licenses issued by thestate’s former communist government. The law had the potential to apply to all per-sons licensed by the state as lawyers. Initially, the special commission revoked thelicenses of 47 lawyers.13 Among the disqualifying conditions were membership in topcommittees of Albania’s Labor Party (the authoritarian communist party) and hav-ing graduated from the Faculty of Law at the state’s postgraduate school for trainingcommunist cadres. 14

12 See for instance Mallinder (2008), Olsen et al. (2010) and most notably Sikkink &Walling (2007), Dancy & Wiebelhaus-Brahm (2018), and the ongoing efforts of theTransitional Justice Research Collaborative (TJRC): Dancy & Montal (2017).

13 These licenses were subsequently returned after the Constitutional Court struckdown key provisions of the law.

14 See amendment to law Nr. 7541 from December 18, 1991, “On advocacy in the Re-public of Albania.”


Fig. 1.3: Purges and Transparency events for 84 Post-Authoritarian and Post-conflict states as a function of years lapsed since transition

The green line in Figure 1.3 below shows the mean occurrence of purges—boththorough and leadership—disaggregated into single events for 84 countries in theGlobal Transitional Justice dataset. The shaded grey area shows the confidence inter-vals around the mean. It indicates that purges continue to occur even years after thetransition has concluded. Their frequency declines over time, though not at the samerate. The blue line below it shows the occurrence of transparency regimes, which Iturn to next.

Among what I call transparency regimes, the first mechanism investigated in thisbook is lustration, which vets candidates for public office for ties to the former au-thoritarian secret police. An example of a lustration law is Poland’s April 1997 bill,which required all candidates running for office to declare in advance of the electionswhether they had spied on their fellow citizens for Bezpieka, the secret police under thecommunist regime. Former spies who owned up to their collaboration were allowedto run for office, but the information on their collaboration was revealed to voters atthe time they cast their ballots. Negative declarations were forwarded to the LustrationBureau of the Institute of National Remembrance (IPN), which attempted to recon-cile them with evidence from Bezpieka’s archives, also housed at the IPN. Failure toconfirm the declaration would result in a lustration trial. Such a trial could produce aformal declaration that the politician was a collaborator, and the politician in questioncould end up being banned from running for public office for up to 10 years (Nalepa2010b).

The United States’ Federal Congress passed a similar provision to the Polish statuteafter the end of the Civil War. The law required that persons who wished to carryout “certain occupations” (including the practice of law) subscribe to an oath thatthey had never “offered aid to the rebellion.” Subsequently, in 1866 in a case called ExParte Garland, the Supreme Court struck down the provisions dealing specifically with


the disbarring of former members of the Confederate government from practicing asattorneys.15

A second type of mechanism that I classify as transparency regimes are truthcommissions. These are bodies appointed by the government of a newly democraticstate or of a state that has recently recovered from civil strife to disclose the natureof atrocities committed during wartime or under a dictatorship. Truth commissionsshare many characteristics with lustration. First, both deal with truth revelation. Tothe extent that truth commissions reveal embarrassing information with the potentialof ruining a politician’s career, truth commissions and lustration have a similar effecton the quality of representation.

Consider, as an example, the El Salvadoran truth commission, which between July13, 1992 and March 15th, 1993 investigated serious acts of violence occurring since1980. Victims filed 22,000 complaints with the commission. 60 percent of these com-plaints concerned extrajudicial killings, 25 percent involved disappearances, and 20percent pertained to torture (many complaints alleged more than one form of vio-lence). State agents were found responsible for as much as 85 percent of the violence.The commission attributed “only” 5 percent of the responsibility to the rebel group“Farabundo Martí National Liberation Front” (FMLN), the leftist paramilitary orga-nization. The report then went on to name specific individuals responsible for theabuses and recommended the disqualification of any culpable military men and civilservants from public employment.

Although some truth commissions stop short of explicitly banning from officeperpetrators that have been found guilty of human rights violations, shaming alonecan have a powerful effect. If the public condemnation of such acts is strong enough,the mere revelation of the wrongdoers’ names can effectively destroy their politicalcareers. Relatedly, revealing the truth about the authorship of human rights abusesprevents the blackmail of perpetrators holding public office at the hands of thosein possession of incriminating evidence. A well-functioning truth commission de-prives anyone in possession of information about “skeletons in a politician’s closet”of the ability to extract concessions in exchange for keeping the embarrassing—orworse, incriminating—information secret. Hence, truth commissions and lustrationlaws have similar effects.

Truth commissions are also similar in format to lustration because they are government-appointed institutions and so carry with them more legitimacy and credibility thando grassroots or NGO-inspired processes of gathering data about past human rightsabuses.

In light of all of the aforementioned similarities, I refer to truth commissions andlustration jointly as transparency regimes. The capacity of lustrations and truth com-missions to add transparency stands in sharp contrast with purges—thorough orleadership-restricted—as the latter deal with overt collaboration and membership inauthoritarian and criminal political organizations. Together, these four transitionaljustice mechanisms, portrayed in Table 1.1, constitute the institutions for reckoningwith the authoritarian past and civil war legacies that I focus on in this book.

The columns of table 1.1 underscore why it is important to distinguish betweenthe vetting of open and secret members and collaborators of the ancien regime. Themechanisms through which transparency and purges affect the quality of democracyin the long term are not the same. In a nutshell, politicians who have in the past been

15 See 71 U.S. (4 Wall.) 277 (1867).


Table 1.1: Transitional Justice Mechanisms Discussed in this book

Nature of CollaborationReach into the Population Secret OpenLeaders Lustration Leadership PurgesLeaders & Rank and File alike Truth Commissions Thorough Purges

clandestine collaborators of the authoritarian regime or secret committers of atrocitiescan be blackmailed by those who have credible access to information on these “skele-tons in the closet.” Needless to say, if the public still cares about what happened inthe past, the revelation of such “skeletons” could end a politician’s career. In returnfor their silence, individuals in possession of this evidence can demand rents or pol-icy concessions. Regardless of the currency in which the ransom is paid out by theblackmailed politician, the quality of democracy suffers.

In contrast, purges deal with known collaborators. Getting rid of members whoran the agencies of the former authoritarian regime is like getting rid of bureaucrats;here one can think of administrative purges as the reverse of a delegation problem.If a new politician comes into office and carries out a thorough purge, he is forcedto implement policy in inherently uncertain conditions: without the expertise of peo-ple who used to run the agencies of the ancien regime, he cannot know how policyimplementation will be affected by states of the world unknown to him. Conversely,a decision to forgo a purge can be thought of as the equivalent of delegation to anagent who is equipped with expertise and thus able to adjust policies to the state ofthe world.

The “reach into the population” dimension (the rows of Table 1.1 is no less impor-tant, particularly when it comes to the temporal impact of transitional justice mech-anisms. Thorough purges, in contrast to leadership purges, apply a blanket rule toworkers of the former authoritarian state. Because of this collective responsibility,the reach into the population of thorough purges is greater as some workers of stateagencies are fired even though they personally bear no responsibility for the regime’stransgressions. Leadership purges only extend to those rank and file members of thestate who participated in the regime’s transgressions, however these are defined. Thor-ough purges typically appear early in the transition. Leadership purges, however, canstretch out long into the transition.

A comparison of lustrations and truth commissions reveals variation along thesesame dimensions. There is no place a perpetrator of humans rights violations can hidefrom a truth commission with a sufficiently far-reaching mandate. Not being in thespotlight of public office does not keep one’s name from not being mentioned during ahearing or as part of a report. In contrast, the operation of lustrations is limited to elites.Even though who is considere an elite and who is not can vary from one lustration lawto the next, those who want to avoid having their skeletons in the closet exposed maysimply forgo running or holding the positions to which lustration extends. In practice,this could mean that lustration works slower than truth commissions. However, this


depends also on what the recommendations of truth commissions are and how wellthey are implemented (Zvobgo 2019b,a).

1.3 The contributions

The above discussion highlights that transitional justice mechanisms play a key rolein determining who is selected into political office. Lustration accomplishes this mydisincentivizing former collaborators from running. Purges do this more explicitly byfiring those with direct (leadership) or indirect (thorough) links to the ancien regime.Despite this, neither the literature on candidate selection nor the literature on politi-cal economy of bureaucracies has paid attention to authoritarian legacies.16. This bookwill bring legacies of the ancien regime to the forefront of understanding democraticstability. The literature on democratization, save for a few exceptions that deal mostlywith party systems (Grzymala-Busse 2002, Pop-Eleches & Tucker 2011, Pop-Eleches2007, Riedl 2014), has also failed to appreciate just how important authoritarian lega-cies are. Authoritarian regimes and their democratic successor-states are, for the mostpart, studied by separate groups of scholars.

On the other hand, Transitional justice scholars are consumed by the debate be-tween retribution and reconciliation (Encarnacion 2014). The first contribution of thisbook is to stress the role of authoritarian legacies in new democracies, and particularlythe importance of these legacies in regulating who works for the new state. This alsobreaks with existing research on transitional justice, which has focused on the norma-tive imperatives of dealing with crimes committed in the past. Because of this norma-tive framing, transitional justice literature has not been able to appreciate that deal-ing with transgressions that have not even been made public may be more importantthan reckoning with known perpetrators of the ancien regime. The forward-lookingorientation of this book fills that lacuna and brings to light just how threatening se-cret legacies of authoritarian rule are to the new democratic state. Thus, by explaininghow TJ regulates selection into office, I also contribute to the literature on democraticbacksliding in important ways.

Second, this book explains how TJ affects how politicians behave in office con-ditional on being elected, appointed, retained. This question belongs squarely in thecomparative democratization literature. In the last twenty-five years, scholars of com-parative democratization have shifted their focus from studying democratic transi-tions to studying the quality of democracy. This phenomenon is associated with theproliferation of hybrid regimes—that is—regimes that hover between autocracy anddemocracy (Levitsky & Way 2010). Hybrid regimes are states that employ democraticprocedures, such as elections (Gandhi & Lust-Okar 2009), constitutions (Ginsburg &Simpser 2013), and legislatures (Wright 2008, Jensen et al. 2014), but are in fact ruledby closet autocrats (Nalepa et al. 2018). One of the aspects of democratic quality isthe question of programmatic representation. The contribution my book offers to thisliterature zeroes in on the linkages between citizens and political parties. When politi-

16 Indeed, when political economist talk about purges, they restrict their attention toauthoritarian purges (Montagnes & Wolton 2019, Jiang & Yang 2016)

1.3 The contributions 15

cians are blackmailed by those who threaten to release skeletons in their closet, theydepart from the programmatic commitments they made to their voters.17

The nature of these linkages has a critical impact on democratic stability (Tavits2005, Pop-Eleches 2010). Scholars have noted that parties in newer democracies rely ona number of strategies to build linkages with citizens; these strategies range from offer-ing programmatic party platforms to relying on charismatic candidates or clientelism(Kitschelt & Wilkinson 2007). Mainwaring (1999) argues that under-institutionalizedand fragmented party systems with volatile electorates are low-hanging fruit for pop-ulist political elites who wish to engage in clientelistic practices. The establishmentof programmatic parties is hindered not only by a lack of adequate institutions, butalso by the poor quality of political elites. The inexperienced politicians that charac-terize young democracies are simply more credible when promising private goodsthan when they promise policy. Establishing clientelistic ties is easier than compet-ing for votes on the basis of programs (Keefer 2007). Political elites might also hold amonopoly on goods or services that the electorate wants to access, which also hindersrepresentativeness by, for example, reducing the competitiveness of elections (Med-ina & Stokes 2007). Yet, the literatures on the nature of party-voter linkages and onthe quality of democratic representation have largely left out considerations of tran-sitional justice: it is not clear why we have not only no tests, but also no theories onhow dealing with former authoritarian elites and their secret legacies will affect theability of politicians to represent voters’ policy preferences. My book fills this gap byshowing that blackmail with secret files hurts representation unequivocally.

My third contribution is to the literature on delegation. Notice that the dilemmafacing the new politician is familiar to students of delegation or principal-agent mod-els. The agent—in this case, the ex-authoritarian bureaucrat—may have preferencesthat are so divergent from those of the principal (the new democratic politician) thathe uses his expertise to implement the policy he himself prefers. On balance, this pol-icy outcome may be worse for the principal than his own implementation, even whenit is lacking in expertise. There is a familiar tradeoff between the expertise the ancienregime bureaucrat can offer and the extent to which loyalty to the previous regimerender his services a liability. The specific transitional justice context that makes thisdilemma particularly interesting is that both of these features—expertise and pref-erence divergence—have roots in the preceding regime. I will investigate how themode in which the authoritarian state appoints bureaucrats influences the formerstate’s agent’s expertise and alignment with the new democratic agents. The final fac-tor I will model is uncertainty, which is one of the key features of regime transitions.Yet, some transitions are more uncertain than others. For instance, in the context ofpost-communist transitions, not only political regimes but also economic systems un-derwent an overhaul. In light of this, the delegation problem in the context of post-authoritarian purges offers a unique opportunity to apply an old, if not somewhatstale literature to a completely different area of social science.

17 I define programmatic representation as running on platforms that are communi-cated to voters, being elected to office because of those platforms, and finally im-plementing the platforms upon being elected.


1.4 A novel approach to collecting TJ data

Though not a theoretical contribution, a non negligible by-product of my research isthe production of a new transitional justice dataset. In order to test the implications ofmy theory, the Transitional Justice and Democratic Stability Lab that I direct has col-lected the Global Transitional Justice Dataset comprising of a time-series cross-sectionof transitional justice events in all post-authoritarian and post-conflict states since theend of World War II. Despite a growing number of empirical studies examining thebroad impact of transitional justice on democratic stability and peace (Olsen et al.2010, Van der Merwe et al. 2009, Thoms et al. 2010), there is a knowledge gap per-taining to the impact of purges and transparency regimes on the long-term quality ofdemocratic representation.

The first generation of empirically-oriented transitional justice research concen-trated on explaining patterns and trajectories of the implementation of specific tran-sitional justice measures (Stan et al. 2009, Lundy & McGovern 2008, Mallinder 2008,Pettai & Pettai 2014). More recently, to understand the impact of transitional justiceon the quality of democracy, scholars have begun to investigate transitional justice asan independent variable (Sikkink 2011, Sikkink & Walling 2007). Noteworthy becauseof their coverage of lustration policies are contributions by Horne (2017b) and Olsenet al. (2010). Horne examines the relationship between the nature of transitional jus-tice mechanisms and trust, suggesting that revealing the notoriety with which citizenswere spying on one another may decrease interpersonal trust in countries with a largenetwork of covert collaborators. At the same time, her research indicates that a wideand compulsory lustration procedure may result in substantial bureaucratic turnover,thereby increasing political trust in governmental institutions. Horne’s research wasamong the first to disaggregate trust in this manner; much of the previous literatureassumed that transitional justice was essential for trust-building. Yet, although sheprovides a rich description of lustration trajectories for as many as 12 countries, theyare all in post-communist Europe.18 Moreover, Horne extends the use of the term “lus-tration” to post-authoritarian purges that is, the removal from office of collaboratorswhose nature of collaboration is well-known. This contradicts the etymology of theword “lustrate,” which comes from Latin and refers to shedding light and makingsomething transparent.19 Hence, her finding of negative effects for the application ofwhat she calls “lustration laws” to institutions of the state is consistent with my find-ings on the ambiguous effects of purges.

Data collection efforts that expand beyond post-communist Europe have also beenundertaken. One example is Olsen et al. (2010). These authors, however, reduce an

18 Most contributions to the lustration literature suggest that this transitional policyis limited to post-communist Europe (De Greiff & Mayer-Rieckh 2007, Ellis 1996,Closa Montero 2010, Letki 2002, Stan 2013, Stan & Nedelsky 2015). This is, however,an artifact of restricting the search for personnel transitional justice to the term “lus-tration” and its derivatives, when in fact, vetting the state apparatus for secret andopen members and collaborators of the former authoritarian regime is called bydifferent names in different parts of the world—these names range from “vetting”and “purging” and “house-cleaning” to specific terms such as “denazification,”“decommunization,” and “de-Baathification.”

19 See Webster’s II New Riverside Dictionary: lustrate [Lat. lustrare, lustrat] to purify,make bright.

1.4 A novel approach to collecting TJ data 17

entire process of implementing lustration to the one year when it was implemented.The danger of simplifying the sometimes complex progression of a transitional justiceproposal through the legislative process to a single date is underscored by the figurepresented below, which presents data on lustration as a time series of progressive andregressive events for 5 countries that I will use as archetypal cases throughout the book:Spain, Poland, Tunisia, Bolivia and South Africa.

To allow for better understanding of the figure, I clarify what positive and nega-tive events are. I define a positive event as the submission of a lustration proposal tothe floor of the legislature, the passage of such legislation, the upholding of such leg-islation as constitutional by a supreme court, or the overturning of a presidential vetoagainst such legislation. I define a negative lustration event, in contrast, as the votingdown, vetoing, or striking down by the constitutional court of lustration provisions.Similarly, expanding the set of persons targeted by transitional justice or broadeningthe set of “offenses” (where “offense” is defined in light of the transitional justiceprocedure in question) to include more past or present positions constitutes a posi-tive lustration event, whereas attempts to narrow the set of targets or “offenses” arenegative lustration events. Positive and negative events in truth commissions, thor-ough purges and leadership purges, though not shown here, are defined similarly.I will argue that ambiguities in the extant empirical research stem from data collec-tion strategies that are not sensitive to the temporal dimension of transitional justice.None of these questions can be addressed before knowing when, how, and for howlong personnel TJ has been implemented. Our data brings scholars a key step closerto understanding the impacts of personnel TJ.

The raw data used to create the above figures (and the dataset) start with eventsearches in Keesings Record of World Events, Lexis Nexis Academic Universe, andnumerous secondary sources for each post-conflict and post-authortarian state, be-ginning with its most recent democratic spell. The number of positive and negativetransitional justice events is then aggregated to create an annual panel, with countriesas the cross section and years since the transition as the temporal dimension. A panelassembled in this way allows for the creation of many different measures of transi-tional justice and the raw chronologies themselves allow researchers to experimentwith different systems of disaggregation. One of the measures is severity.20 The figureabove shows data for lustration which is the first transparency regime analyzed here.

It is worth elaborating on the rationale for focusing on personnel transitional jus-tice in this book at the expense of focusing on trials of former human rights violatorsand on victim compensation. Personnel TJ mechanisms should stand out as criticallyimportant for the objective of limiting the influence of former authoritarian elites be-cause these mechanisms deal with these elites most directly. Yet, scholars analyzingdata on transitional justice have found little evidence that these mechanisms make adifference in preventing former autocrats from reasserting their political dominance(Olsen et al. 2010, Binningsbø et al. 2012) or that they matter for the long-term qual-ity of democratic representation (Olsen et al. 2010, Van der Merwe et al. 2009, Thomset al. 2010).

Another reason for focusing on personnel transitional justice is that it does notrun into the stark problems with the principle of non-retroactivity that criminal trialsdo. The principle of nullum crimen sine lege (“no crime without law”) poses a problem

20 For a more elaborate discussion of this and the two other measures I develop in thisbook,volatility and urgency, see Chapter 4.


0

1

2

3

4

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29Years Lapsed since Transition

Num

ber

of L

ustr

atio

n E

vent

sType of Lustration Event

negative

positive

Lustration Events in Poland

0

1

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23Years Lapsed since Transition

Num

ber

of T

ruth

Com

mis

sion

Eve

nts

Type of Truth Commission Event

negative

positive

Truth Commission Events in South Africa

0

1

2

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30Years Lapsed since Transition

Num

ber

of T

ruth

Com

mis

sion

Eve

nts

Type of Truth Commission Event

negative

positive

Truth Commission Events in South Korea

0

1

2

1 2 3 4 5Years Lapsed since Transition

Num

ber

of L

ustr

atio

n E

vent

s

Type of Leadership Purge Event

negative

positive

Leadership Purge Events in Tunisia

0

1

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41Years Lapsed since Transition

Num

ber

of L

ustr

atio

n E

vent

s

Type of Lustration Event

negative

positive

Lustration Events in Spain

0

1

2

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30Years Lapsed since Transition

Num

ber

of T

rans

ition

al J

ustic

e E

vent

s

Type of Transitional Justice Event

negative

positive

Transitional Justice Events in Bolivia

Fig. 1.4: Severity of transitional justice mechanisms in archetypal countrycases of the book

for trials of former perpetrators of human rights violations because such trials crim-inalize actions that were not only legal, but indeed encouraged, under the previousauthoritarian regime. In the case of trials, the choices of policymakers from differentcountries are restricted by statute of limitation provisions (Posner & Vermeule 2004,Siegel 1998) and/or by the extent to which they are able to adhere to Natural Lawas opposed to Formalism (Schwartz 2000). Because it is difficult to assess the extentto which different countries’ legal traditions restrict their implementation of criminalforms of TJ, we are better able to isolate the effects of TJ if we can assume that eachcountry had available to it the same range of mechanisms. In the case of personnelTJ, it is safe to assume this. Turnover in personnel, implied by lustration, purges, andtruth commissions, is not as sensitive to the principle of non-retroactivity regardless


of how permissive to the autocrats the law on the books was at the time human rightsviolations were committed.21

Figures 1.6 and 1.8 above disaggregate TJ events over time. As this chapter also em-phasizes a crucial distinction highlighted lies in distinguishing between purges andtransparency regimes: the mechanisms behind them affect selection into politics, be-havior in office, and delegation to bureaucrats in starkly different ways. It is thereforeno surprise that states embarking on transitional justice programs have used these TJmechanisms in various degrees of severity. As mentioned above, the Global Transi-tional Justice Dataset covers 84 countries; However, for easier reading, I summarizehere the severity of the four transitional justice mechanisms in the five countries thatI will use at various stages of this book to illustrate my argument.

Fig. 1.5: Truth Commissions Fig. 1.6: Lustrations

Fig. 1.7: Thorough Purges Fig. 1.8: Leadership Purges

21 For other excellent arguments on why limiting the focus of TJ to trials is misleading,see Murphy (2017).


The top two panels illustrate the severity of lustration and truth commissions; thebottom two illustrate purges.22 Countries with no events of a certain type are not evenincluded in the figure. We see that Tunisia had the highest severity of truth commis-sions, although it did engage in all four transitional justice mechanisms, includinglustration. In fact, the severity of lustration event in Tunisia was almost as high as inPoland. Poland, similarly to Tunisia, had all four types of transitional justice events;however, among transparency regimes, it invested in lustrations instead of truth com-missions. Further, it invested slightly less in leadership purges than Tunisia did.23.

Interestingly, South Africa, most famous for its Truth and Reconciliation Commis-sion established after the negotiated transition to democracy in 1994, has only expe-rienced truth commission events, and the severity of those events is only at 0.5 (on ascale from 0 to 1). One can compare this to Tunisia’s 0.8. This is consistent with thereconciliation orientation of the South African TRC whereby, through a confession,perpetrators were granted amnesty from criminal prosecutions.

Notable here is also Spain, where despite the “Pact of Forgetting,” the country em-braced transparency regimes in the form of both lustrations and truth commissions.Yet, as figure 1.6 above indicates, this happened relatively recently. The case with thelowest scores on any of the transitional justice mechanisms studied in this book is Bo-livia. Indeed, this dearth of TJ is the rationale behind including this case as a moredetailed illustration of the phenomena in question. The only transitional justice mech-anisms used were truth commissions, and these events only reached a level of severityequal to slightly over 0.25.

The Global Transitional Justice dataset can also be used to justify the disaggrega-tion of transitional justice into purges—that is, dealing with known forms of collabo-ration with the ancien regime or engagement in human rights violations—and lustra-tion and truth commissions—that is, dealing with secret forms of collaboration andhuman rights violations that have not been revealed. The left panel of Figure 1.9 belowplots the total number of positive transitional justice events net of negative events as afunction of time lapsed since the transition (left/upper panel) and as a function of theyear in which the transition took place (left/lower panel). Here, all transitional justiceevents have been pooled together, and there appears to be no relationship between TJand two variables that ought to be good predictors of TJ: time lapsed since transitionand year of transition (Elster 2004b, Huntington 1993, Barahona de Brito et al. 2001).

However, once we disaggregate the transitional justice mechanisms into purges(thorough and leadership), lustration, and truth commissions, a clear pattern emerges.Consider first the lower right panel of Figure 1.9, illustrating positive TJ events netof negative events for the four mechanisms as a function of transition year. Lustra-tion prevails in countries that transitioned around 1990, which tend to be the EasternEuropean ones (Albania, Bulgaria, East Germany, Estonia, Hungary, Latvia, Poland,

22 “Severity” is defined as the ratio of positive transitional justice events of a certaintype to the total number of events. The full justification for this measure is providedin chapter 4.

23 As a side note, Tunisia’s leadership purge score is somewhat deflated by the factthat many of its leadership purges took place before its first elections. The GlobalTransitional Justice Dataset, for instance, does not include the removal of “30 toppolice officers removed by the Ghannouchi administration” on February 2, 2011(Preysing 2016).


Fig. 1.9: Disaggregating Transitional Justice Data

Slovakia, and Slovenia), as previous scholarship has speculated.24 There is also anuptick in truth commissions around the beginning of the third wave of democrati-zation, but in contrast to lustration events, truth commission events trend upwardagain in countries with mid-1990s transitions as well as in countries transitioningaround 2010.25 This is consistent with the scholarship on truth commissions: truthcommissions abound in South America (in Paraguay, Ecuador, and Peru) and Africa(in Kenya, South Africa and Liberia); they can also be found in Indonesia and EastGermany (United States Institute of Peace 2011g,b,a,h,e, Gibson 2006, United StatesInstitute of Peace 2011f,c,d).

24 See De Greiff & Mayer-Rieckh (2007), Ellis (1996), Closa Montero (2010), Letki(2002), Stan (2013), Stan & Nedelsky (2015). Note, however, that there are alsoinstances of lustration in other countries. A deeper look into my data reveals thatthey these other countries include Argentina, Spain, and Guatemala.

25 These include the Arab Spring countries,Tunisia and Egypt,as well as several coun-tries in South and Southeast Asia.


The story with purges is quite different. First, the occurrence of thorough purgesis flat across the range of transition years in my dataset. If they do occur, they occurin the immediate aftermath of a transition (the slight uptick on the left end of theupper right panel of Figure 1.9. Leadership purges, on the other hand, seemed tobe more popular in the beginning of the Third Wave transitions (in Latin Americancountries that transitioned in the seventies and eighties) and their popularity seems tobe increasing again after 2005. As in the case of thorough purges, they are concentratedin the early post-transition years.

The insight from the upper panel of Figure 1.9 is that the timing of lustrationclearly differs from the timing of purges. While purges—both thorough and leadership-only—take place in the immediate aftermath of transitions to democracy, lustrationspeak about ten years following the transition. Truth commissions are implementedsoon after the transition or conflict termination but continue to be implemented longerthan purges or lustration.

In sum, patterns of purge activity are dissimilar to patterns of lustration and truthcommission activity. Similar inferences can be drawn from the GIS-coded version ofmy data in the form of world maps illustrating all three of my measures. Data onseverity, volatility, and urgency values for lustration, truth commissions, and thoroughand leadership purges are provided in the GIS Appendix.26

1.5 Roadmap

This book is organized as follows. Chapter 2 presents a formal model that focuseson how lustration can prevent blackmail, arguing that transparency regimes, suchas lustrations and truth commissions, have a positive effect on the long-term qual-ity of democratic representation. I capture the blackmail potential of “skeletons in thecloset,” understood here as evidence of collaboration. If left in the hands of formeragents of the ancien regime, these “skeletons” have the potential to threaten demo-cratic representation. The model uncovers the extent to which implementing trans-parency regimes can mitigate this danger. Transparency regimes accomplish this bymaking bluffing more difficult by lowering the a priori belief that embarrassing se-crets may still be in the hands of the blackmailers. A more normative implication ofthis model is that transparency regimes do not always work the way advocates offorward-looking arguments for transitional justice would want them to. In some equi-libria, blackmail with secret police files occurs even without evidence. Yet, an identityresult presented in this chapter shows that no matter what the equilibrium, departurefrom perfect representation is always the same function of two parameters: (1) howmuch voters pay attention to being represented by politicians who in the past had en-gaged in secret spying or human rights abuses and (2) the severity of the transparencyregime.

Chapter 3 presents a formal model of purges (both of the thorough type and atthe leadership level) and clarifies why the effects of purges on the long-term quality ofdemocratic representation are quite different from the effects of lustration. To capturethe dynamic of administrative purges, I make use of an old workhorse of models of bu-reaucracy from American Politics: the delegation model. In order to compare the effectof lustration with that of purges, the model’s results are stated in terms of the same26 They are also available at an interactive website: https://tinyurl.com/ybmcj7hf.

1.5 Roadmap 23

outcome variable as in the model of lustration blackmail: the quality of democraticrepresentation. The model builds on the literature on delegation in American Politics,but crucially, shows that purging more is not always better for representation. Newdemocracies face trade-offs between using the ancien regime’s agents’ expertise to re-duce noise around policy implementation and facing the possibility that these agentshave misaligned preferences. An important difference between purges and truth rev-elation procedures made clear as a result of this modeling exercise is that while pref-erence divergence between the new democratic politician and the agent of the ancienregime has no effect on the quality of representation in the case of truth revelationprocedures, it has a critical moderating effect in the case of purges.

The models presented in chapters 2 and 3 share a dependent variable—the qualityof representation. At the same time, important differences exist between them: themodel of lustration deals with how to select a pool of representatives that will not beheld hostage by what they did in the past; the model of purges deals with the challengeof staffing post-authoritarian bureaucracies.

The introduction to Part II of the book summarizes the empirical implications ofmy theory and discusses the empirical strategy for testing who selects to go into poli-tics, what politicians do while in office, and how they delegate tasks to their bureau-cracies.

Chapter 4 is devoted to a discussion of the data that operationalizes the dependentand independent variables of this book. First, it discusses alternative operationaliza-tions of the dependent variable: the quality of democratic and specifically, program-matic representation and political corruption. Additionally, I also develop a measureof authoritarian elite reproduction to be used in subsequent empirical chapters. Asfar as the independent variables are concerned, since no existing datasets differenti-ate between open and closed forms of collaboration, my data collection on transitionaljustice not only separates transparency regimes (truth commissions and lustration)from purges (thorough and leadership-only). I also code transitional justice data asevents. This chapter develops three novel measures of personnel transitional justicecharacteristics: urgency, severity, and polarization. Using these measures, I show otherdifferences between the Global Transitional Justice Database and other existing datacollection efforts.

In Chapter 5 the blackmail model is tested with a hierarchical model where theunits of analysis are political parties nested within Post-Authoritarian states.

This strategy is appealing because lustration is an elite-centered transparencyregime with most of the affected elites concentrated in political parties. For trans-parency regimes, with a wider reach in society, two things are true. First, the extent towhich they affect selection into politics is negligible (because anyone can be exposedin the process of a truth commission operating, there is no disincentive to specificallyavoid politics). Second, they are more fast acting than lustration. This second charac-teristic means that a design that looks at change over time can be employed. Hence toexamine the effects of truth commissions, in Chapter 6 I use a difference-in-differencedesign. Identifying the causal relationship between personnel transitional justice andquality of democracy is crucial because it is highly plausible that the same charac-teristics that lead a country to embark on a transitional justice program may be theones that down the line allow it to enjoy a high quality of democratic representation.This design consistently identifies a causal relationship between transitional justiceand democratic quality.


Moving on to testing my theory on purges, recall that one of its key findings is thatnot all states can embark on purges because at times, the loss in expertise is too greatto conduct a thorough purge even when the ideological distance between the bureau-cracy of the ancien regime and the new succeeding democrats is large. In light of this,some countries simply cannot implement purges at all. This calls for an instrumentalvariable approach with authoritarian regime institutionalization as the instrument forpurge activity. This is the subject of Chapter 7

Chapter 8 draws conclusions as it evaluates the role and availability of transitionalmechanisms in shaping democratization. Far from being “ritual sacrifices,” trans-parency regimes regulate how politicians behave in office. Truth commissions makepolities less corrupt while lustrations are better at dissociating political power fromeconomic power and allowing politicians to represent better. The effect of purges ismore complex as even the feasibility of purges depends heavily on the institution-alization of the previous authoritarian regime. Moreover, this chapter also indicatesthat there are important differences between transparency regimes. Namely, the effectof truth commissions on democratization becomes apparent sooner, but fizzles outover time as truth commissions continue in operation. Lustration, on the other hand,seems to require surpassing a certain threshold of intensity before positive effects onthe quality of democracy come to fruition.

The overall conclusion is to categorically discredit the virtues of the so-called“Spanish Model” of transitional justice. Letting sleeping dogs lie, particularly whenit comes to yet-to-be-revealed crimes perpetrated by former autocrats, is exactly thewrong way to go about dealing with the past.

2

Blackmail and Transparency Regimes

2.1 Introduction

In the previous chapter, I argued that democracies do not emerge in a vacuum andthat legacies of former authoritarian states permeate the new polity in the form of un-settled scores of human rights abuses, staff of former security agencies, and archivesof the former secret police that list who worked as a collaborator of the authoritariansecurity apparatus. This chapter is devoted to studying how transparency regimes—the disclosure of the identities of collaborators, spies, and saboteurs—influences theselection of who gets elected to office and how these representatives behave once in of-fice. In sum, this chapter will explain why lustrations and truth commissions improvethe quality of representation in nascent and more mature democracies that have ex-perienced authoritarian rule.

Transparency regimes—the revelation of who, among persons holding public of-fice, collaborated with the secret police to the detriment of the anti-authoritarianopposition—are often left out of the classical critiques of transitional justice. Hence,when Samuel Huntington (1993) writes that “even if a moral and legal argumentcould be made for prosecution, this would fall before the normative imperative ofcreating a stable democracy,” he implicitly assumes that transitional justice cannot bedemocracy-enhancing. Other classical contributors to the comparative study of de-mocratization have also made the implicit assumption that all of transitional justiceis punitive. For example, Snyder & Vinjamuri (2004) argue that “the prosecution ofperpetrators of atrocities according to universal standards risks causing more atroci-ties than it would prevent, because it pays insufficient attention to political realities.”Holmes (1994) calls some forms of transitional justice “witch hunts.” Voytech Cepl(1992) refers to such acts as “ritual sacrifices.” Even Huntington (1993) maintains thatsometimes “amnesty ... is necessary to establish a new democracy on a solid basis.”Born out of this understanding is the use of the term “retroactive justice” as synony-mous with transitional justice (Rev 2005).

The key contribution of this chapter is to show that transparency regimes—lustrationand truth commissions—by making public secrets that former autocrats or theiragents could use in blackmail, improve the quality of representation. I will showthat by revealing acts of secret collaboration with the authoritarian regime, these pro-cedures prevent former authoritarian elites from extorting policy concessions from

26 2 Blackmail and Transparency Regimes

former collaborators who have become elected politicians, but have “skeletons intheir closet.” Implementing transparency regimes such as lustration laws should helppoliticians become responsive to their constituencies. Even though the game I presentin this chapter has three types of equilibria, I show that transparency regimes alwaysenhance representation. Additionally, the quality of democratic representation de-pends on the transparency regime’s severity, the extent to which dissidents-turned-politicians suffer if their skeletons come out, and on the extent to which independentmedia are able to check bluffs from agents who blackmail politicians with “fake” skele-tons.

Although the formal model and the commentary that follows use the terminol-ogy of lustration laws, the same model can be used to explain the operation of truthrevelation procedures: fundamentally, both procedures perform the same function,revealing past human rights violations that have thus far been kept secret. In the em-pirical chapters that follow, I will demonstrate that the theory provided in this chapterdoes indeed underpin the operation of both lustrations and truth commissions.

Among TJ mechanisms, lustration, truth commissions, and purges are all linkedto the objective of barring former authoritarian elites from policy influence. Lustrationdoes this by vetting candidates for public office for secret ties to the former authori-tarian secret police. It accomplishes this by opening archives of the former regime’ssecret authoritarian apparatus to uncover who, among persons running for public andpolitical office, had worked as an informer or spy prior to the democratic transition.The proven collaborators are then explicitly banned from holding office, or informa-tion about their collaboration is released to the voters, who can then decide whetherto still cast their votes on thus-compromised politicians. Truth commissions can onlyindirectly ban former members and collaborators from office, but as long as the vot-ing public or nominating agencies react negatively to truth revelations about the past,they can prevent former authoritarian elites and their agents of repression from serv-ing the new state. Purges prevent agents of the the ancien regime from exploiting theirexpertise in a way that would sabotage the policy goals of the state’s new leaders.As explained in the previous chapter, the removal of known collaborators is accom-plished by a different mechanism than the uncovering of secret collaborators of theancien regime. That is why purges are analyzed in a separate chapter (chapter 3).

Lustration is neither a new nor even a modern procedure used by democracies.The Athenian constitution made room for an institution similar to lustration. Doki-masia, translated best into English as “scrutiny” applied to anybody who had beenappointed to public office, whether by election or by lot. It involved a judiciary pro-ceeding where orators offered arguments on why a candidate may not be fit for office.Dokimasia became very popular following the coup d’etat of 404-403 BC, when un-covering secret ties to the Thirty Tyrants (Todd et al. 2007) became a tool of evaluatingwhether a politician in question had the moral character required to hold public office.Collaboration with the Thirty Tyrants, if demonstrated, was considered disqualifyingbecause it indicated a lack of loyalty to the constitution.1

In the 21st century, skeletons in politicians’ closets became known under the term“kompromat.” In 2016, the term kompromat—used to describe Russia’s attempts to in-tervene in US policy—made front page news. Generally, kompromat (compromising

1 According to Todd et al. (2007), while dokimasia was routinely conducted outsideof the the 403-380 B.C. period, it was only contested in the two decade s that fol-lowed the coup.

2.1 Introduction 27

materials) refers to either embarrassing information or evidence of a person’s illegalactivity that, if revealed, could damage that person’s career or open her up to prose-cution. With a few exceptions, the use of secrets and blackmail to effect policy changehas not received much scrutiny from political scientists. Among the few exceptions tothis trend is the work of Keith Darden (2001), who notes that this kind of data wasroutinely collected by authoritarian secret services in the Soviet Era and used to con-trol people through blackmail: the “compromised” were threatened with the releaseof damaging information to the public or—worse—to prosecutors. In a more recentarticle, Hubert & Little (2020) conceptualize the collection of embarrassing informa-tion by principals on their agents as a way of disciplining subordinates under an au-thoritarian regime. The authors use a cheap talk game to account for the possibilityof kompromat being leaked. Embarrassing or damaging information, even if collectedby authoritarian security forces, may also be put to use when the authoritarian regimeis long gone. In an even more recent paper, on which I draw in this chapter, Nalepa& Sonin (2019) focus on a blackmailer who cannot generate kompromat, but can useit to force compromised politicians into making concessions. In their model, votershave the ability to update beliefs about the probability that they are represented by acompromised politician.

The reason that mechanisms for dealing with former secret collaborators of theancien regime have largely eluded analysis is that we cannot directly observe or knowexactly how often former authoritarian elites pressure their past collaborators throughsuch acts of blackmail. However, if succumbing to such pressures prevents democrat-ically elected politicians from representing voters, one may wonder whether formerauthoritarian regimes suffer from a deficit of representation and accountability whenthey fail to make their authoritarian legacies public.

Transparency regimes, such as lustration and truth commissions, can reduce theunfair informational advantage that members of the authoritarian regime hold longafter their rule is over. The Polish lustration law, a classic example of lustration, re-quires all persons holding or running for public office to declare in advance whetherthey had collaborated with the secret authoritarian police prior to the transition. In-formation from declarations admitting collaboration is put on the ballot, and votersdecide whether to cast their vote for a former collaborator. Negative declarations aresent to a special division of the Institute for National Remembrance, where they areverified against information assembled in the archives of the former secret politicalpolice. Proven collaborators who lied on their declarations are banned from runningfor office for 10 years. Although this is the most cited example of lustration (Kaminski& Nalepa 2006, Nalepa 2010a, 2012, Letki 2002, Williams et al. 2005, Szczerbiak 2002),it is hardly typical because it allows two types of collaborators to escape direct sanc-tions: (1) the collaborator who admits he worked as a secret collaborator2 and (2) thecollaborator who failed to own up to his past but was not uncovered. A more typicallustration law carries with it an explicit sanction for anyone who is proven to haveworked for the secret police as an informer (as in Hungary) or who fails to provideevidence of his innocence (as in the Czech Republic).

A classic example of a truth commission is the South African Truth and Reconcilia-tion Commission (Hayner 2001, Gibson 2006). The Commission was formed in 1995 toinvestigate crimes committed against the South African people during the apartheid

2 See Nalepa (2008) for a discussion of whether a positive declaration is, in fact, nota sanction.


regime (1960-1994), covering human rights violations committed by both the stateand various liberation movements.3 The Commission’s mandate provided it with theability to offer amnesty to those who fully participated in the process and truthfullyconfessed the full extent of their crimes. It released a five-volume final report to then-President Nelson Mandela in 1998. The report detailed the abuses committed by theapartheid-era National Party government, the state-opposition-turned-ruling-partyAfrican National Congress (ANC), and other “leading political figures on both sidesof the anti-apartheid struggle” (Keesing’s Record of World Events 1998). Yet, just likethe Polish lustration law, the South African Truth Commission is hardly typical. Mosttruth commissions have a narrower mandate, certainly not allowing those who confesstheir crimes to enjoy amnesty from criminal responsibility. Many truth commissionsare not allowed to complete their full mandate, because they run out of financial re-sources, political backing, or both.

Regardless of small differences within truth commissions, within lustrations, andbetween lustrations and truth commissions, all such transitional justice mechanismsincrease transparency.

By revealing information about past abuses committed by anyone, and not justby politicians, truth commissions also prevent the blackmail of decision-makers—bethey in politics or in other positions—with embarrassing information about formerhuman rights violations. In short, both lustrations and truth commissions influencewho is selected to run for office. The decision whether to implement a transparencyregime has a direct impact on how politicians behave once in office: if implemented,lustrations and truth commissions prevent former authoritarian elites from influenc-ing policy in new democratic polities. Where transparency is lacking, former authori-tarian elites can pressure politicians into making policy concessions by threatening toreveal compromising information that could jeopardize their careers. If blackmail iseffective, politicians lose their ability to faithfully represent voters.

The most important finding from the model I solve below is that the more strin-gent the lustration law, the better the quality of democratic representation. I showthat this holds regardless of the class of equilibrium I identify. Since what I proposein this chapter is a signaling model, the three types of equilibria are separating, semi-separating and pooling. The conventional wisdom on how lustration operates matchesthe intuition of the separating equilibrium: lustration separates compromised politi-cians from uncompromised ones. The former are immune to blackmail, while the lat-ter succumb to it. Because in a separating equilibrium, lustration separates the twotypes of politicians, it would seem that only this type of equilibrium serves the nor-mative goal of transparency. Yet, I show that under the two other equilibria—semi-pooling, which partially separates the two types, and pooling, which does not separateat all—increasing the severity of lustration universally increases the quality of repre-sentation. This is somewhat surprising, as under both of these additional equilibria,blackmail—and the accompanying extraction of policy concessions by the blackmail-ing agent— occurs even with lustration policies in place. For this reason, separatingequilibria are particularly attractive, and it is worthwhile knowing the circumstancesunder which they arise. The second finding from the analysis of the model is thusthat the chances of a separating equilibrium decrease as the preferences of the newly

3 Specifically, it was established via the Promotion of National Union and Reconcili-ation Act, passed by the South African parliament in July 1995 (Waldmeir 1997).

2.1 Introduction 29

elected politician and the former secret police agent diverge, but the quality of demo-cratic representation is not affected by this distance.4

Since separating equilibria most closely reflect how conventional wisdom under-stands lustrations to work, it may seem worrisome that these other equilibria existat all. After all, in the pooling equilibrium, lustration does nothing to modify vot-ers’ beliefs regarding the past collaboration of the politician they are about to reelect.Semi-separating equilibria only provide partial information. Because of this, it is vi-tal to stress that regardless of the type of equilibrium that is, regardless of whetherblackmail with information about skeletons in the closet actually occurs, the severityof lustration always improves the quality of democratic representation. Moreover, thisresult is theoretically robust regardless of assumptions about the functional form ofplayers’ preferences.5

Before I introduce the formal model and the main intuition behind it, it is worthexplaining that my arguments in support of transparency regimes differ quite sig-nificantly from those made in the rest of the literature on transitional justice. Thisliterature, on the one hand, makes few distinctions between purges and lustration.On the other hand, it treats lustrations and truth commissions as completely differ-ent mechanisms. Arguments in support of truth commissions range from recognizingvictims’ need to “reconcile” with the past to accentuating victims’ rights to know theidentity of those who harmed them. Arguments for lustration, in the meantime, rangefrom pointing out that victims of spying have the right to know who informed the au-thorities of their activities (Stan 2012) to emphasizing the importance of preventingformer spies and their leading officers from playing key roles in public service (Nedel-sky 2013, Stan et al. 2009, David 2011). The common argument made in favor of bothlustrations and purges is the normative and backward-looking one: that former spies,their leading officers, and members of the former regime should be prevented fromplaying key roles in public service (Nedelsky 2013). Such arguments assume that thequestion of whether the truth about transgressions by the former authoritarian elitesand their collaborators should be unearthed has a definite answer. In fact, as I arguein this book, this question is complex and far from settled.

The next section reconstructs the blackmail mechanism that lustration and truthcommissions are ideally supposed to undercut with a game theoretic model of incom-plete information. The game directly models how lustration laws limit the potential ofblackmailing members of the former security apparatus to interfere with politicians’ability to faithfully represent policy preferences of the electorate.

A key strength of the model is that the potential for blackmail, which is present be-cause of specific authoritarian legacies that affect certain new elites, creates a problemfor the quality of political representation. Critically, this is the only friction to politicalaccountability that produces the theoretical result. Hence, the game theoretic modelcaptures only pathologies in representation that are due to a lack of transparencyregimes.

4 I do not put much stock in this claim, however, as altering players’ utility functionsto quadratic makes distance significant in the intuitive way: the greater the distancebetween the blackmailing agent and the new democratic politician, the higher thepolicy concessions ?.

5 In a related article, Ang & Nalepa (2019) show that changing the players’ utilityfunctions to quadratic does not alter this finding.


2.2 The Formal Model

Can transparency regimes such as lustration improve the quality of representation byeliminating the effectiveness of blackmail with dark authoritarian secrets? My modelof incomplete information reconstructs the mechanism through which blackmail in-terferes with politicians’ ability to faithfully represent the policy preferences of theelectorate. Solving the model allows us to examine the extent to which transparencyof this kind reduces the vulnerability of politicians to their former leading officers.

Existing formal models of transitional justice focus on how autocrats disciplinetheir agents of repression Tyson (2016), Hubert & Little (2020), Montagnes & Wolton(2019). The dilemma faced by autocrats is highlighted by Powell (2014): “a weak mil-itary can leave them vulnerable to ... civil war, while a strong military...a coup d’etat.”Paine (2019) has recently argued that the same reasoning can be applied to internalsecurity forces. Authoritarian leaders have two choices:

1. They can maintain weak or fragmented security forces, thus guarding against acouple d’etat but risking resistance from below.

2. They can invest in strong state security at the risk of empowering a strong com-petitor.

Dragu & Lupu (2018) use a coordination game with incomplete information toshow that paradoxically, repression is most likely to hinge on expectations and be-come a coordination game when authoritarian leaders need it most (when dissentagainst them is at its highest). While in Dragu & Lupu (2018), the autocrat’s dilemmais limited to the authoritarian regime, Tyson (2016) extends the consequences of theautocrat’s actions and the consequences of actions of his agents of repression beyondthe authoritarian regime, allowing for transitional justice. He models the interactionbetween a leader and his repressive apparatus in circumstances where the stability ofthe authoritarian regime is uncertain. The autocrat in these circumstances must com-pensate his agents of repression to offset the risk they bear of being punished shouldthe regime collapse. Tyson’s model uses the prospect of transitional justice to modelrepressive agents’ incentives, and his theory assumes that the identity of these agentsis known. This need not be the case, however. Moreover, the availability of informationon who, in the past, was or was not a secret collaborator of the ancien regime dependson the transparency regime, which is itself a subject of strategic choice in my model.

Hubert & Little (2020) draw on both of these strands of transitional justice lit-erature and conceptualize the collection of embarrassing information by principalson their agents as a way of disciplining subordinates under an authoritarian regime.The authors use a cheap talk game to account for the possibility of kompromat beingleaked. My model departs from Hubert & Little (2020) in that it focuses on conse-quences rather than origins of kompromat. My blackmailer cannot generate kompro-mat, but he can use it to force compromised politicians into making concessions.

Another model with similarities to mine is that by Nalepa & Sonin (2020). There,voters also form beliefs about the probability that they are represented by a compro-mised politician, but in addition, voters are uncertain about the specific location of thepotentially compromised politician. Indeed, Nalepa & Sonin (2020) show that anygeneral model of kompromat must have at least four cross-cutting types that allowthe voter’s uncertainty to translate into a blackmailer’s leverage over politicians.6 An6 More specifically, they argue that first, there must be compromised and uncom-

promised politicians. Second, there must be uncertainty about the uncompromised

2.2 The Formal Model 31

important distinction between my model and that ofNalepa & Sonin (2020) is that Iallow the blackmailer to bluff with the threat of revealing skeletons even if no suchskeletons exist. This is only possible if the politician himself does not know whetherhe collaborated with the secret police. How might this be possible?

In Communist Europe, numerous incentives existed to recruit members of the op-position as collaborators even when they were not fully aware of the fact that theywere being recruited. Officers with more informers were paid more, were promotedfaster, and certainly had no incentives to tell their informers that they were recruitedby the secret police. For this reason, I assume that any opposition politician may havehad kompromat produced against them. This possibility allows blackmailers to bluff,making the signaling model in this chapter non-trivial. Henceforth, when I use theterm “collaborator’,’ I will be referring to politicians with kompromat or skeletons inthe closet and not necessarily to conscious collaborators.

2.2.1 Setup

My model features a former collaborator elected to public office following a transitionto democracy and his former officer, who threatens to reveal embarrassing informationabout the former collaborator unless he implements policies favored by the officer. Thequality of representation suffers when the politician is vulnerable to extortion by theagent and implements the policy preferred by his former secret police officer insteadof the policy preferred by voters. I assume that unless they are blackmailed, politi-cians implement the policy preferences of their voters. In making this assumption, Ido not claim that politicians always follow voter preferences; however, this assump-tion is necessary to isolate the effect of lustration on curbing blackmail with secretpolice files.

The key tension in my model comes from the fact that while the officer knowswhether he possesses the evidence he threatens to release, the former collaborator-turned-politician does not. Note that the more stringent the lustration law is, the lesslikely it is that the former agent possesses such incriminating evidence. Therefore, Imodel the severity of lustration as the probability that evidence of the politician’s col-laboration remains in the possession of the officer. Restating this in terms of the moti-vating example from the previous chapter, if harsh lustration had been implementedin Poland before Wałesa served as President, Czesław Kiszczak, the secret police chief,would not have been able to use the files against Wałesa.

2.2.2 Sequence of play

The game starts with a move of Nature, which determines with probability 1 − π ∈(0, 1) that the officer has evidence against the politician and with probability π that hedoes not. 1−π in this model represents the exogenously given severity of lustration: themore severe lustration is, the less likely it is that unearthed evidence of collaborationindeed remains in the hands of the officer. The Politician, P , knows the value of π butnot its specific realization. In the second stage of the game, the Officer, O , decideswhether to make a policy demand of the Politician. I represent this decision by one of

politician’s true policy preferences. Otherwise, any deviation from the uncompro-mised politician’s ideal policy would expose the elected leader as compromised.


two exogenously given actions in {demand, no demand}. If no demand is made, thegame ends and the Politician implements the policy corresponding to his ideal point.

For simplicity, I assume that the Politician’s ideal point is given by p = 0, and theOfficer’s ideal point is represented by a > 0. Thus, a captures the ideological distancebetween the Politician and the Officer. P observes whether a demand has been made,but not whether evidence against him exists (the realization of π). In the third stage,P decides whether to make a policy concession in response to the Officer’s demand,and how big this concession should be. I model this concession as proposing a policyx ∈ [0, a]. If evidence against the Politician does not exist, the game ends and theOfficer pays the cost of bluffing, 0 < c < a.7 The bluffing cost, c, captures the cost ofobtaining and disclosing some plausible information about the Politician’s past. Thisis costlier for the Officer in a world of independent media, where such disclosures areverified by journalists and, if false, exposed.

If evidence does exist, the Officer decides whether to reveal the evidence (R or∼ R). The revelation is interpreted as the Officer exercising his threat to reveal kom-promat against the Politician. In the event that evidence is revealed, the Politician paysthe cost of having skeletons in his closet revealed, or, in short, of being fired, F .

Due to the asymmetric way in which information is distributed in models suchas the one above, they are called models of incomplete information. Additionally, Iassume that the informed party—the Officer—moves first and places a policy demandon the Politician, who then has a chance to respond to this policy demand. Because thePolitician may learn something about the agent’s type from his action, this is a modelwith “updating,” also referred to as a dynamic model of incomplete information. Theinformed party, moving first, is referred to as the “sender of the signal,” while theuninformed party, moving second, is referred to as the “receiver of the signal.” Forthis reason, models such as these are frequently referred to as “signaling models.”These models have specific categories of equilibria, called “separating,” “pooling,”and “semi-separating.” In the first category, each type of informed party chooses adifferent action (in this sense, the types “separate”). In the second type of equilibrium,both types “pool” on the same action. Finally, in the third category, at least one of thetypes uses a mixed strategy, meaning that he sometimes pools with the other type andsometimes separates from the other type.8

The strategies, beliefs, and preferences are formally defined in the appendix,where the game is also solved for equilibrium, but the timeline of the game can besummarized as follows:

1. Nature decides whether the politician running for office is a collaborator (withprobability 1− π) or a non-collaborator (with probabilityπ).

2. The Officer observes the move of Nature and decides whether or not to make apolicy demand (D,∼ D).

3. The Politician, who does not observe the move of Nature but observes the actionof the officer and if a demand is made, makes a counteroffer in the form of a policyconcession: x.

4. The Officer decides whether to release the kompromat (R,∼ R).5. The game ends, and payoffs are distributed to both players.7 The second restriction is purely technical and merely simplifies equilibrium calcu-

lation.8 This explanation of signaling models assumes discrete types, although signaling

models also allow continuous types.

2.3 Results 33

Fig. 2.1: Lustration Blackmail Game

NatureOfficer

−a; 0

Politician

(−|a−x|;−x−F

)R

(−|a−x|;−x

)¬R

x

1− πOfficer

−a; 0

(−|a−x|−c;−x

)x

π

no demand

demand

The payoffs are a function of the distance between the players’ respective idealpoints and implemented policy as well as the two types of costs characterized above:(1) the cost to the Politician of being fired as a result of skeletons revealed in his closet,F ; and (2) the cost of bluffing incurred by the Officer if he makes an empty threat, c.Critically, there is no additional cost (or benefit) to the Officer from releasing kom-promat.

2.3 Results

The model is solved completely in the formal appendix, but I offer some intuitionshere about how the three types of equilibria come about. This is summarized in thefigure below as a function of F , the cost to the Politician of skeletons being exposed.

The intuition behind the pure separating equilibrium is fairly simple: all Officersequipped with evidence of kompromat against the Politician choose one strategy andall officers not equipped with such evidence choose the other strategy. Upon observingthe action of the Officer, the Politician can with certainty predict whether he has skele-


Fig. 2.2: Policy outcomes in the Lustration Blackmail Game

Fc c(1−π)

Policy outcome

F

Separating

c

Semi-pooling

(1− π)F

Pooling

tons in his closet or not. Defining this equilibrium amounts to finding a counteroffersufficient to keep the officer from switching strategies away from this equilibrium, butthat also tolerable for the Politician.

Key for this equilibrium to be sensible is that we avoid the pathological separationwhere the Officer without evidence makes a demand, while the Officer with evidenceof skeletons in the closet refrains from making one. Fortunately, this equilibrium iseasily eliminated because it is never rational for the Officer not to make a demandwhen he has evidence against the Politician.

Note that P cannot make a counterproposal x unlessO has made a demand. Thus,the worst O can do when making a demand is −a, which would be his payoff if P ’scounterproposal were x = 0; that is, if P made no concession at all. Without mak-ing any demand, O is guaranteed to receive −a and no more. This leaves only onepossible form of separating equilibrium: the Officer with no evidence does not de-mand concessions, the Officer with kompromat makes a demand; the Politician makesa counteroffer, and that counteroffer is large enough to stave off the Officer’s reveal-ing of kompromat. The requirement for this to be a Bayesian equilibrium is that thePolitician, upon seeing a demand, is certain that he is dealing with an Officer equippedwith kompromat; if he sees no demand, he knows with certainty that no such evidenceexists.

2.3 Results 35

A second intuition behind this equilibrium, as seen from Figure 2.2, is that becausethe Officer can expect a concession only 1− π of the time, the concession itself has tobe quite large—equal to F .

Finally, for this pure separating equilibrium to exist, the cost of having skeletonsrevealed (F ) relative to the bluffing cost (c) must be high enough. In this pure sepa-rating equilibrium, blackmail is effective with probability 1 − π. The flip side of thisresult is that lustration severity, represented by π is directly proportional to the qual-ity of democratic representation. Moreover, the quality of democratic representationis inversely proportional to the cost of having skeletons in the closet exposed. Para-doxically, the more voters care that the politician they are about to elect to office is nota former collaborator, the greater the concession by a politician who actually used tobe a collaborator must be.

I now turn to a discussion of the pooling equilibrium. In many respects, it is the po-lar opposite of the separating one. Here, Officers with and without evidence choosethe same action, implying that the Politician cannot update his prior beliefs to poste-rior beliefs by conditioning on the Officer’s action. In presenting the intuition for sep-arating equilibria above, I explained that the Officer will always place a demand whenevidence is present. Thus, the only possibility for a pooling equilibrium in this gameis one where both types of Officer—with and without evidence—make demands.

In this equilibrium, the Officer always places a demand and the Politician alwaysoffers the same counterproposal. As illustrated in Figure 2.2, this counterproposal isx′ = (1−π)F . This value is clearly lower than in the case of the separating equilibrium.Below I present the intuition behind why this is the case.

For such an equilibrium to hold, the dissident has to prefer having his proposalaccepted to having skeletons in his closet revealed. However, because only a fractionof the threats to reveal skeletons in the closet can actually materialize, the greatestpolicy concession the highest proposal the Politician will accept, given the Politician’sbeliefs, the Officer’s optimal rejection region is necessarily lower.

At the same time, this equilibrium exists only when the cost F of having skeletonsrevealed is quite high relative to the cost of bluffing c, and the probability that skele-tons exist is lower or, conversely, when the lustration law is more severe. Note that1 − π in the denominator corresponds to the availability of skeletons, and a smallervalue makes the entire expression c

1−π greater, leaving less “room” for the poolingequilibrium. Since lustration severity corresponds to π, the reverse is true for lustra-tion: higher lustration leaves more “room” for pooling. If this seems counterintuitive,the following further explains the reasoning: when lustration is severe, even if all of-ficers make demands, the chances that a demand is backed by a credible threat ofskeletons being revealed is actually low. As a result, the corresponding counteroffercan be correspondingly low as well. The only factor making the counteroffer higher,of course is, the cost of having skeletons revealed. Key here is that severity of lustra-tion can mitigate against a high cost of having skeletons revealed. To translate thisback to empirical implications in transitioning democracies: when new democraciesare deeply concerned about former collaborators remaining in office, more severe lus-tration laws make the policy concession the Politician must make to the blackmailersmaller, increasing the quality of representation.

Finally, I turn my attention to the effectiveness of blackmail with secret police filesunder the semi-pooling equilibrium.

In this equilibrium, the Officer plays a mixed strategy. This means that for at leastone of the signals, he uses a lottery over his actions: with probability λ he chooses


to make a demand, and with probability 1 − λ, he refrains. Technically, the Officercould also play this kind of mixed strategy for both signals. However, I have alreadyestablished above that making a demand is always rational when the Officer has accessto skeletons. I will only be looking for mixing in the eventuality that the Officer hasno skeletons.

Intuitively, this equilibrium will occur in the parameter space between the poolingand separating equilibria—that is, for moderate costs of having skeletons revealed andfor moderate lustration severity.

In this equilibrium, the officer always makes a demand when evidence is presentand, with some probability λ = (1−π)(F−c)

πc, also when evidence does not exist. The

beliefs supporting such a semi-pooling equilibrium are calculated using Bayes’ ruleand require the Politician to take into account that if he is seeing a demand, it mightbe coming from an Officer without information on skeletons in the closet; the proba-bility of this, however, is lower than in the pooling equilibrium. In light of this, moredemanding Officers than in the pooling equilibrium will have kompromat to punishPoliticians with if Politicians don’t satisfy demands. Consequently, the demand hasto be correspondingly lower. In the formal appendix, I calculate this equilibrium de-mand to be x′′ = c. The parameter λ, meantime, the probability that an officer withoutkompromat makes a demand is inversely related to the severity of lustration. This isintuitive: severe lustrations ought to make bluffing less common.

In the next subsection, I discuss the implications of these different counterpropos-als in the equilibrium for the quality of democratic representation.

2.3.1 Quality of representation under the three equilibrium types

How does the quality of democratic representation relate to the effectiveness of black-mail under the separating, pure pooling, and semi-separating equilibria? I notedabove that the size of the counteroffer is largest in the separating equilibrium andsmallest in the pooling equilibrium. At the same time, demands are much less fre-quent in the separating equilibrium than in the pooling or semi-pooling equilibria.

Before discussing an extension of my analysis to truth commissions, I also derivean identity result characterizing the PBE outcome. Consider Figure 2.3 below:

Recall that my model assumes that absent pressure from the Officer, the Politicianwould carry out the policies desired the voters, which correspond to his ideal point.Ultimately, what I am most interested in across all equilibria is the quality of repre-sentation: the extent to which the Politician can withstand pressures from the Officerto abandon his ideal point, 0. In addition to the values of x∗, x”, and x′, which corre-spond to the counteroffers proposed by the Politician (in the the separating, pooling,and semi-separating equilibria, respectively), I present the average levels of misrepre-sentation. Recall, that I interpret the quality of representation as resistance to the of-ficer’s blackmail. Conversely, misrepresentation is interpreted as the departures from0, the Politician’s ideal point.

In order to derive the expected level of misrepresentation, I weigh the PBE out-come in each equilibrium by the frequency of its occurrence. In the case of the pureseparating equilibrium, this is simply (1− π(F )) + (π)0, as the Officer only proposesx∗ = F when evidence exists, which is 1−π of the time. The remaining π of the time,he reverts to 0. In the case of the pure pooling equilibrium, the average policy is im-plemented at 1 ∗ ((1− π)F ), as the Officer always places a demand and the Politician

2.3 Results 37

Fig. 2.3: Equilibria

Fc cπ

x∗ = F

1− π of the time

Pure Separating

weighted avg:

(1− π)F

x′= c

(1− π) + πλ

of the time

Semi-separating

weighted avg:

(1− π)F

x′′= (1− π)F

all of the time

Pure Pooling

weighted avg:

(1− π)F

always responds with x” = (1− π)F . In the case of the semi-separating equilibrium,the calculation of the policy implemented is somewhat more complex because the of-ficer always places a demand when evidence exists, which occurs 1 − π of the time,and λ∗ of the time when evidence does not exist. Hence, the total frequency of placinga demand is given by (1−π)+π π(F−c)

πc. The Politician responds to this demand with

x′, bringing the expected policy outcome to (c)((1−π)+π π(F−c)πc

+π(1− (1−π)(F−c)πc

)0,which, as in the previous two cases, reduces to πF .

It appears that average misrepresentation is lower in the separating equilibriumthan in the pooling equilibrium,9 while misrepresentation under the semi-separatingequilibrium also depends on c and a. Yet, under all three equilibria, average misrepre-sentation decreases with lustration severity, which is represented by π. Additionally,note that under the semi-separating equilibrium, and especially under the poolingequilibrium, transitional justice does not entirely prevent blackmail. Under the pureseparating equilibrium, bluffing never occurs, and the amount of misrepresentationtracks the amount of evidence left in the Officer’s possession (so it is directly respon-sive to the severity of lustration). This equilibrium becomes easier to achieve as thecosts of bluffing increase because for higher values of c, it is easier for F to be lowenough to satisfy the c < F constraint.

In the semi-separating equilibrium, the Officer always makes a demand when ev-idence exists and with probability λ∗ when evidence does not exist. The key featureof the semi-separating equilibrium is that when it is played, the Officer extracts policy

9 Since π <√π


concessions, with probability λ∗, undermining representation, even when evidencedoes not exist. Misrepresentation does not exactly track transitional justice, but in1 − λ∗ of the cases, when evidence has been swept by the transitional justice pro-cess, bluffing does not take place. This equilibrium occurs when the cost of revealingskeletons, F , is moderate.

In the pooling equilibrium, bluffing occurs every time evidence does not exist re-gardless of how much evidence transitional justice has left in the hands of formerofficers. From the normative point of view, the separating and the semi-pooling equi-libria are better than the pooling equilibrium: under this equilibrium, the severity oflustration has no bearing on the frequency of blackmail. Nevertheless, the magnitudeof concessions is not as large as in the other two equilibria. This pooling equilibriumoccurs when having skeletons in the closet revealed by the former secret police is verydamaging to the Politician’s career (when F is high).

Consequently, a higher c makes the constraints on F easier to satisfy. Thus, itmakes the separating equilibrium more likely vis a vis the semi-pooling equilibriumor the semi-pooling equilibrium more likely vis. a vis. the pure pooling equilibrium.As a result, raising the costs of bluffing, c allows transparency regime to eliminateincentives to misrepresent voters’ preferences.

In Ang & Nalepa (2019), we examine whether these results are not being driven bymodeling assumptions, and specifically by the functional form of the utility functions.There, we solve the model for quadratic as opposed to “tent” preferences. Althoughmy result on the effect of costs of revealing skeletons on the separating equilibrium isrobust to this change in assumptions, my result on the effect of preference divergenceis not. After changing the functional form to quadratic, the distance is shown to havea negative effect on total misrepresentation. Critically, however, the result on the effectof lustration severity on total misrepresentation is preserved by the change.

These expected policy outcomes are described in Proposition 2.1, which followsdirectly from my derivation above. I summarize this analysis in a proposition that alsoforms the basis for the main hypothesis tested in Chapter 5, which is that lustrationseverity increases the quality of representation:

Proposition 2.1. The PBE outcome, interpreted as the expected level of misrepresentation, isgiven by the same formula across all three equilibria: πF .

A key implication that follows from Proposition 2.1 is that departures from per-fect representation are constant for all values of a. This means that the effectivenessof blackmail does not depend on how far apart the ideal points of the Officer and thePolitician are. This is somewhat surprising as intuitively, one might expect the ideo-logical proximity of ideal points to matter. Indeed, this specific result—the irrelevanceof distance between ideal points—is not robust to changes in the functional form ofthe players’ preferences: With quadratic preferences, the effect of distance is negativeon quality of representation; this is consistent with the empirical results presented inChapter 5. Yet, out of concerns for simplicity and because distance is not my mainvariable of interest, I keep the expression (1 − π)F as a succinct summary of my ex-pectations.

2.3.2 Potential criticism

My model is not impervious to criticism, however. First, the assumption that a politi-cian does not know that he is a collaborator may strike some readers as unrealistic.

2.3 Results 39

How could the collaborator himself be unaware of his involvement? Sadly, the real-ities of collaborator recruitment in robust authoritarian regimes, such as the ones inEastern Europe, show that such occurrences were common. Secret police officers wereencouraged and rewarded to recruit collaborators broadly and worked under cover.It was common for students and laborers involved in dissident movements to shareinformation with strangers who later turned out to be secret police officers. Officers,in turn, registered even casual conversations as recruitment events, as they soughtto maximize recruit numbers and not intelligence. As for intelligence itself, even themost benign information could be used to recruit additional collaborators. Minutedetails of the contents of a lecture detailed by a student marginally involved in thedissident movement, for example, could be put to use to recruit key student activists.All the secret police had to do was overwhelm them with information of their livesthat they already had in their possession. The target of recruitment would feel help-less and succumb to the recruitment effort. This technique worked in part because thefirst “collaborator” rarely thought he or she was doing anything wrong. After all, whodoes not like to share the details of a riveting lecture they just attended? Years later,following the transition, this has led to tragic discoveries when, after the opening ofsecret police archives, many Eastern European learned that they had provided thepolice with information. In Poland, the scale of such undercover recruitment effortsled the Constitutional Court to include a special exemption to the lustration law for“unconscious collaborators.” This allowed targets of lustration proceedings to arguethat they had been unaware that they were the subject of recruitment efforts. Poland’sfirst President, Lech Walesa, described in the introduction to this book may have beensuch an “unconscious collaborator” and in recent years convincing evidence support-ing such collaboration was discovered by historians (Chodakiewicz 2009).10 Anotherillustration of the ambiguity with the category of collaboration comes from the filmby FLorian von Donnersmarck , “The Live of Others.” In this fictional story about aStasi officer spying on a high-profile theater director, during the set-up of the eves-dropping devices at the director’s home, the Stasi officer notices that his team’s activ-ity was noticed by the director’s neighbor. He knocks on her door and threatens thatif she even says a word about the spy operation to the theater director, her daugh-ter will lose her spot at the University. Next, as he turns to leave he says “Make sureto send Frau Meineke a gift in appreciation of her collaboration.” Alongside the giftwould be a receipt that Frau Meineke would sign, providing the Stasi, as well as futureblackmailers, with evidence of her collaboration.

The second problem that concerns the model is that the action of the Officer in thelast stage is not unique in equilibrium. The blackmailing secret police officer is justas well off releasing the kompromat as he is keeping it to himself. Nalepa & Sonin(2019), in fact, show that any general model of kompromat with a unique equilib-rium in which kompromat works must have at least four cross-cutting types that allowthe voter’s uncertainty to translate into a blackmailer’s leverage over politicians. First,there must exist compromised and uncompromised politicians. Second, there mustbe uncertainty about the uncompromised politician’s true policy preferences. Other-wise, any deviation from the uncompromised politician’s ideal policy would expose

10 More specifically, according to Chodakiewicz (2009) Walesa later admitted that hehad spoken with Security agents prior to becoming Solidarity leader but categori-cally refused any kind of collaboration and evidence corroborating the latter claimwas produced.


the elected leader as compromised, and the transparency regime would simply ban ex-collaborators from running for office. In a transparency regime, moderate politicianswould be elected to office and the politician with the ideal point closest to the me-dian would be elected. Without transparency, Nalepa & Sonin (2019)’s model showsthat a potential blackmailer can exploit the fear of moderate compromised politicians(whose policy positions are preferred by the median voter to the policy position ofsome uncompromised politician). The blackmailer induces these moderate politiciansto move policy away from the median voter. The voter, in turn, anticipates this to be thecase and votes instead for a challenger with more distant policy positions but whosecollaborator status is not under question.11

The upshot of this analysis is that a lack of transparency always translates into pol-icy that that is farther from the median voter than it would brunder a transparencyregime. Hence, the consequences for the quality of democratic representation are ex-actly the same as in my model.

Extension to truth commissions

The formal model of lustration can be adapted to accommodate truth commissions.The game tree would look the same, with slight differences in the identities of theactors. Instead of a Politician taking an action, some other elite member would playthis role. Similarly, the person exercising blackmail does not need to be a secret policeofficer; anyone privy to information about human rights abuses committed by formerelites and interested in leveraging that information to extract concessions from elitemembers can play this role. These concessions may take the form of rents or privi-leged access to resources the elite member controls. Truth commissions, by revealinginformation about past abuses, work in the same way as lustration does to prevent thisinformation from being used.

2.4 Conclusion

This chapter has reconstructed the role of transparency regimes in determining who,in new democracies, is selected into office and how these politicians behave in of-fice. Regardless of their retroactive value, lustration and truth commissions enhancethe quality of democracy by making blackmail with secret police files impossible. Themodel clearly shows that blackmail is still possible even after lustration and truth com-missions have been implemented, at least in some circumstances. Such blackmail withsecret authoritarian legacies takes place when voters place a very high premium onhaving elites that are clean of suspicions of collaboration and when the costs of bluff-ing are low. Chapter 6 will develop this intuition further, but media independenceis the easiest empirical interpretation of the costs of bluffing. Bluffs by secret officerswho do not actually have evidence of collaboration can be uncovered more easily incontexts where independent journalism is well developed.

Even though lustration does not eliminate blackmail completely, the effect of lus-tration severity on the quality of democratic representation is positive. The quality of11 In the Nalepa & Sonin (2019) model, this kind of politician corresponds to the suc-

cessor autocratic party—that is, the party whose members could have been mem-bers of the former regime, but not secret collaborators.

2.4 Conclusion 41

representation is monotonically increasing in the the severity of truth revelation be-cause the concessions the Politician must make in the pooling and semi-separatingequilibria are smaller than those under the separating equilibrium.

The three types of equilibria are presented in Figure 2.3 as a function of F , thecost to the Politician of having his skeletons revealed. The set of conditions for thepooling equilibria is distinct from the set of conditions defining the separating andsemi-pooling equilibria, with the critical element being the magnitude of F relative toπ, the probability that evidence against the Politician has been made transparent by alustration or truth truth commission.

I have shown that for very low costs of revealing skeletons in the closet, there existsa separating equilibrium in which the Officer only makes a demand if evidence exists,and the Politician makes a counteroffer x∗ = F . Second, if the costs of revealing skele-tons are very high relative to the costs of bluffing, there exists a pooling equilibriumin which the Officer always makes a demand irrespective of whether evidence exists,and the Politician makes a counteroffer x

′′= (1 − π)F . Third, for moderate costs of

revealing skeletons in the closet, there exists a semi-pooling equilibrium where theOfficer sometimes bluffs (i.e., makes a demand even when he has no evidence), andthe Politician responds with a counteroffer x

′= c. In all equilibria, the counteroffer is

made and accepted.The main contribution of this model is that it develops a robust result about the

effect of lustration severity on the quality of democratic representation: increasing lus-tration severity always increases the quality of representation. This chapter also iden-tifies conditions under which former secret police officers extract policy concessionsfrom politicians even after lustration has been implemented. I identify (pooling andsemi-separating) equilibria under which lustration is compatible with bluffing. Underthese equilibria, lustration does not prevent departures from perfect representation.These empirical implications are developed further and illustrated in Chapter 5.

In a related article, Nalepa & Sonin (2019) suggest a compelling reason why in-cumbents refrain from implementing transparency regimes even when they themselveshave no skeletons in the closet. Incumbents suppress transparency, reducing the qualityof representation in new democracies, because they are more likely to win electionswhen their challengers can be blackmailed with kompromat concerning the latter’spast collaboration.

Our model, too, both shows that transitional justice—in the form of transparencyregimes—is good for democracy and explains why it is so rare.

Like the theory presented in this chapter, Nalepa and Sonin assume that the surfac-ing of kompromat instantly destroys the career prospects of the compromised politi-cian and that the incumbent in power has no skeletons in her own closet. In otherwords, we consider the case where transparency, on the surface, should benefit the in-cumbent most. Yet, we point to a wide range of circumstances where lustration is notimplemented. We focus on modeling the strategic interaction between a median voter,an incumbent, a (potentially compromised) challenger, and an agent of the ancienregime’s security apparatus. The key feature of the model is that the member of theancien regime’s security apparatus has private information about whether a memberof the opposition and challenger to the incumbent has done something embarrassing,such as collaborated with the security police prior to the transition. Hence, this is amodel of kompromat (Darden 2001).

3

Purging the Authoritarian State

Do you swear to faithfully serve the new Polish Republic?I do, to the very end, be it mine or hers. (Wladyslaw Pasikowski, Psy[Dogs])

Introduction

In the previous chapter, I presented the first of two theoretical arguments againstthe “Spanish Model” of transitional justice. Although the Spanish model or ‘doingnothing’ when applied to lustration policies, policies that deal with secret collabora-tors, may not produce immediate negative consequences for democratic stability, itstrengthens the power of authoritarian networks. In this chapter, I ask whether thesame can be said for purges— transitional justice mechanisms that deal with knowncollaborators.

Anecdotal evidence suggests that purges may actually hurt a nascent democ-racy. As an example, consider Iraq’s policy aimed at purging new democratic insti-tutions of former Ba’athists. According to sources cited by Roman David (2006), de-Ba’athification prevented 185 members of Saddam Hussein’s party, mostly Sunnis,from running in legislative elections in 2003. Despite its promise to promote societalreconciliation, this policy only ignited ethnic tensions. Jon Elster (2004a) describesFrance’s policy of epuration, banning former Vichy collaborators from holding officefollowing World War II, as having very similar effects.

In other contexts, purges may be a welcome transitional justice mechanism. Con-sider as an illustration the purge of the Tunisian Gendarmerie, Ben Ali’s political po-lice, which by some counts numbered 150,000 workers (or one security officer per 80Tunisians). In contrast to the military, which was quick to start disobeying Ben Aliin the midst of the protests, the Gendarmerie remained extremely loyal to the dicta-tor. As early as February of 2011 (less than one month following the self-immolation)30 top police officers were removed from their posts by the interim Prime MinisterMohammed Ghannouchi.1 A top military officer was appointed to head the national

1 Similarly, the party of Ben Ali, the Constitutional Democratic Rally (Rassemblementconstitutionnel democratique, RCD) did little more than to prop up Ben’ Ali’s ruleand had no independent role in state administration. Consequently, its disbanding

44 3 Purging the Authoritarian State

security service, and career military officers were named as new police chiefs for sevenkey regions in the country. Following this leadership purge, in March of 2011, the en-tire secret police agency was disbanded. Yet, as Bouguerra (2014) points out, the “in-telligence service in Tunisia was divided into two agencies... the Directorate Generalof Special Services (SS), which worked on general intelligence information, and theDirectorate General of Technical Services, which provided the same information butthrough technical instruments such as phone tapping or Internet control. The Direc-torate of State Security (DSE) coordinated the SS and ST.” The suspension of the DSEeffectively dissolved the political police and was welcomed by ordinary Tunisians.However, many experts cautioned that it would disrupt the smooth functioning of theentire intelligence system that filtered information and provided analysis. Beyond thesuspension of intelligence gathering, suspending the political police also entailed clos-ing down smaller units within the General Directive for Public Safety; these includedtraffic police, public safety police, crowd control police, and others. All servicemenemployed by these agencies were replaced by new, dramatically undertrained staff.

Taken together, these introductory remarks bring into stark relief that purges—that is, the removal of known members of the ancien regime and their organizations—are complex and call for analysis separate from transparency regimes.

This chapter employs the tools of formal theory to explore the relationship be-tween purges and the quality of democracy. First, it will show that not all nascentdemocracies can afford to embark on wide-scale purges. This is because any newdemocratic politician who is considering purges is presented with a fundamentaltradeoff between loyalty and expertise. This tradeoff is not unique to post-authoritariandemocracies. Jack Paine’s work on repression underscores the ubiquity of the loyalty-effectiveness dilemma. Paine (2019) considers the tradeoff between appointing a pro-fessional (hence equipped with expertise) army and a personal (hence loyal, but rel-atively less skilled) militia. He then explains that autocrats will appoint professionalrather than personalist militaries when they are more concerned about large revolu-tionary threats from below than about external threats. The reason is that neither kindof security force can survive a revolution from below: a post-revolutionary state muststart with a blank slate, and such turnovers result in everyone being fired from the mil-itary. In light of this, professional militaries, their higher expertise notwithstanding,are not more effective at fighting revolutionary threats than personal militias. Theyhave better expertise, but no incentive to use it. In contrast, in the case of externalthreats, professional militaries can count on leniency. Indeed, in some instances, theyeven collaborate with external challengers to the autocrat. In light of this, they put lesseffort into averting external threats. Consequently, the autocrat who fears such threatsis better off investing in a personal militia.

Implicit in this account, however, is that loyalty to the autocrat on the one handand professionalization on the other, cannot go hand in hand: a loyal military cannotbe professional; whenever there is a conflict between fulfilling the autocrat;s wishesand making the correct decision, the loyal military will ignore what their professionaltraining tells them to do. Consider, however, as a counter-example the military in ParkChung Hee’s post-coup South Korea, where according to Joo-Hong Kim, Park takingadvantage of the Korean War of 1950-1953 “had transformed a rag-tag army, consist-ing of former colonial officers and independence fighters of all ideological stripes split

had few bad consequences for democratization, and indeed enabled the creation ofnew parties disconnected from the previous authoritarian regime.

3 Purging the Authoritarian State 45

into innumerable factions, into a a professionalized military with the potential to leadthe country into modernity” (Baik et al. 2011, p.169). By the time Park launched hiscoup, Kim adds, “the armed forces had become the most cohesive and modernizedinstitution in South Korea.” And yet, despite this professionalization, Park succeededin transforming it into a loyal partner. The military enforced five of his installmentsof martial law and three garrison degrees during Park’s his 18-year rule. The profes-sionalization of the army was in no way hampered by Park’s politization of the armedforces and its expansion into non-military arenas. The South Korean case motivatesmy modeling decision to analytically separate loyalty (represented as preference di-vergence) from professionalization (represented as expertise).

Even if South Korea is an exception in the possibility of combining loyalty withexpertise in an authoritarian agency, in the case of the models presented below, theseparation is vital: another departure from Paine, is that I am interested less in theincentives of the former autocrat and more in the incentives of the democrats whofollow in his footsteps. Yet, the tradeoff between loyalty and expertise I explore is verysimilar. Moreover, since the new democrats inherit a state staff potentially loyal to theautocrat, their decision is better framed as one not about whom to appoint, but aboutwhether or not to purge the authoritarian state apparatus.

Here, the distinction between my understanding of purges and that of the broadertransitional justice literature is relevant. The TJ literature treats administrative purges(such as de-communization, de-Baathification and denazification) and lustration asthe same type of mechanism for dealing with the past (Elster 2004b, Binningsbø et al.2012, Pinto 2008). In this book, I agree that all of these institutions are forms of per-sonnel transitional justice in that they aim at eliminating members and collaboratorsof the previous authoritarian regime from the democratic state’s apparatus. However,I also draw a sharp distinction between dealing with secret collaboration (throughlustration) and known collaboration (through purges).

Before elaborating on my own classification of personnel transitional justice sys-tems, it is worth summarizing some alternative classifications used in the literature.For the most part, as remarked above, the literature pools all categories of personneltransitional justice—collaboration with known and unknown collaborators—underone heading. For instance, Olsen, Payne, and Reiter write that lustration events areoften referred to in terms of the group that is banned from public office, as in “denaz-ification,” “deommunization,” and “de-baathification.” These authors use the term“lustration” to refer to “official state policies to purge individuals from positions theycurrently hold or to ban them from holding specific positions in the future” (Olsenet al. 2010, p.38). This use of the term “lustration” is at odds with the theory presentedin this book, which rests on the distinction between clandestine and open collabora-tion with the ancien regime.

The authors of the widely recognized Post-Conflict Justice Database (Binningsbøet al. 2012) have also pooled personnel transitional justice events into one category.These researchers, however, refer to all personnel forms of transitional justice usingthe term “purges,” which describes “the acts of removing politicians, armed forcesmembers, judiciary or other members of society for their (alleged) collaboration withor participation in a conflict and limiting their influence accordingly” (Binningsbøet al. 2012, p.736). In part because their data collection effort is limited to societiesrecovering from conflict and only covers the first five post-conflict years, these authorsonly locate 15 post-conflict episodes that are followed by purges thus defined. One ofthe contributions of my Global Transitional Justice Database, described in Chapter


4, is to offer a time series of personnel transitional justice events that begins with acountry’s transition to democracy, does not end after some fixed period, and spansthe entire democratic period. In fact, this chapter draws on examples from periodsthat long postdate the immediate aftermath of a transition, presenting cases from thede la Rua administration in Argentina, for example.2

Finally, while some authors, including Roman David (2011) and Cynthia Horne(2017a), disaggregate personnel transitional justice mechanisms, they do so in a dif-ferent way than I do in this book. David defines lustration as a “special public em-ployment law that stipulates the conditions for the access of persons who worked foror collaborated with the political or repressive apparatus of socialist regimes to cer-tain public positions in new democracies.” At the same time, he limits the applicationof this term to Eastern Europe. He justifies this decision with the fact that prior to1990, lustration was not used to describe transitional justice procedures (David 2011,p.67). While it it true that the term “lustration” has rarely been used to describe dis-qualification for public employment of secret and clandestine collaborators of a for-mer regime, such procedures have been implemented both before 1990 and beyondEastern Europe, as illustrated by the Athenian example of dokimasia in the previouschapter.

Cynthia Horne, meanwhile, tries to “back out” the concept of lustration from waysin which policymakers have used it. Taking a disaggregating approach, she presentsan overlapping categorization of personnel transitional justice events that includesvetting, lustration, and purges. She limits the use of the term “purge” to describingblanket bans extending collectively to members of certain organizations linked to theancien regime. In contrast, a lustration procedure considers each case individually. Shestipulates lustration to be part of a broader category of “vetting” which can ban fromholding office members of the ancien regime based on criteria other than participationin or collaboration with the former authoritarian regime.

Both David’s and Horne’s approaches limit the scope of lustration to EasternEurope. Critically, neither David nor Horne distinguish between procedures basedon revealing new information about secret collaboration (which I call “transparencyregimes” and which include both lustrations and truth commissions) and bans thatrely on open membership in ancien regime organizations (which I call “purges”).

Being devoted to purges, this chapter touches upon all three key themes of thebook: (1) selection into political office, (2) behavior while in office, and (3) delega-tion to bureaucrats. It is organized as follows. In the next section, I present some illus-trative examples of purges in the “thorough,” “leadership,” and “perfunctory” cate-gories. A thorough purge denotes the disbanding of an entire organization of the an-cien regime without discriminating between leaders (those issuing orders) and rankand file (those following orders). Thorough purges include the shutting down of theStasi in East Germany or the closing down of Słuzba Bezpieczenstwa in Poland in 1990,for example. A leadership purge is limited to the top echelon in the hierarchy of theenforcement apparatus, thus discriminating between the leadership of the organiza-tion and the rank and file. This case is illustrated with the Panev Law from Bulgaria.Finally, a perfunctory purge is a vetting process that only applies to the rank and fileor to the lowest echelons of the hierarchy of the enforcement apparatus. Such purges

2 While Argentina is not one of the archetypical cases used in this book, I choose itover Spain for this illustration because Spain’s efforts were so delayed that no oneremained to be purged.

3.1 Illustrative examples of purges 47

sometimes occur while the authoritarian regime is still in office. Under such circum-stances, they should be interpreted as a preemptive move by the authoritarian regime,which shields itself from more severe transitional justice in the hands of the incomingdemocratic regime (Kaminski & Nalepa 2014).3 A perfunctory purge can also be in-stituted by an incoming democratic government when it cannot conduct a leadershippurge because its hands are tied by a peace agreement with the outgoing military au-tocrats (Nalepa 2010b). Under such circumstances, forgoing purges altogether is notfeasible because of third party or international pressures. As the data presented inChapter 4 show that both types of perfunctory purges are very rare, they have beenomitted from the formal model presented here.

Section three develops a formal theoretic framework to explain the concept ofpurges—that is, transitional justice procedures that limit the presence in office ofknown collaborators. Recall that known collaborators are members of the ancien regimewho served the regime in an official capacity as party members, bureaucrats, or uni-formed personnel, and not as secret informers. I develop the theory in three stages.

The first (baseline) model contrasts the tradeoff between uncertainty and loy-alty. It models the decision to purge as the reverse of a delegation problem (Epstein& O’Halloran 1999, Callander et al. 2008). Building on the intuition that firing staffmembers of a security or enforcement agency comes at the cost of losing potentiallyvaluable expertise, this baseline model implies that democratic representation is notalways boosted by thorough purges. The second model analyzes whether if addinga tool to the principals’ toolkit will help control the agent: discretionary limits havebeen introduced to the delegation and agency literature by comparativists (Huber &Shipan 2002) to model the range of actions allowing the principal to better control theagent. Finally, the third model relaxes the assumption of perfect expertise on the partof the agent. Section four concludes.

3.1 Illustrative examples of purges

There are three types of purges: thorough, leadership, and perfunctory. A thoroughpurge denotes the disbanding of an entire segment of the ancien regime institu-tion without discriminating between leaders (those issuing orders) and rank andfile (those following orders). The dissolution of communist secret police agencies inpost-communist Europe exemplifies thorough purges. The purging of the East Ger-man Stasi (from the German Ministerium für Staatssicherheit) is now legendary anddescribed by multiple historians and political scientists (Koehler 1999, Childs & Pop-plewell 2016, Miller 1998). Initially, following Erich Mielke’s resignation, the East Ger-man Council of Ministers renamed the Stasi to the “Office for National Security.” How-ever, less than two months later, the new Prime Minister of GDR, Hans Modrow, or-dered the dissolution of this new office. The Ministry of Internal Affairs inherited thebuildings and facilities of the former Stasi, but none of the employees were rehired bythe new agency.4 The Ministry took over some of the tasks performed by Stasi (notably,the ones that did not involve spying on the opposition). This thorough purge came at3 Preemptive perfunctory purges do not satisfy the definition of transitional justice.4 Childs & Popplewell (2016) report that “most of the Stasi employees had to turn to

some other means of earning their living. However, a significant number did findreemployment in the policy or private security world. In Saxony, it was reported


a cost, however. Numerous journalistic accounts document the gainful employment offormer Stasi officers in the business holdings of Martin Schlaff, an Austrian business-man; in the 1980s, Schlaff had made a small fortune by supplying senior Stasi officerswith products that were precluded from trade under “CoCom,” the embargo imposedon the Soviet bloc by the West (Tillack n.d., Borchert 2006).

Of course, thorough purges need not be limited to the secret intelligence and po-lice apparatus. A Human Rights Watch report describes a purge of the judiciary inPanama: “from top to bottom, judges who held posts under Noriega resigned or werepurged and have been replaced by new ones, almost all of whom lack prior judi-cial experience: all nine of the Supreme Courts judges resigned and were replaced;the newly-constituted Supreme Court then dismissed or had to replace 13 out of the19 judges of the Tribunales Superiores, the intermediate appellate courts; and approxi-mately two-thirds of the 48 trial-level circuit judges, were, in turn, removed or replacedby the newly appointed appellate judges”( 1991).

According to the New York Times, the Argentinian President de la Rua purged theintelligence apparatus of over 1500 agents responsible for involvement in the so-called“dirty war” only two months after taking office (Krauss 2000). Purged agents were ei-ther dismissed or forced into retirement. Instead of releasing the list of names of thosepurged, entire sections of the agency were let go, suggesting that no discriminationwas made between those giving or following orders or based on the level of involve-ment. According to the report, this “housecleaning ... mean[t] nearly a 50% reductionin military intelligence personnel, and officials said they would leave nonmilitary in-telligence work to civilian agencies” (Krauss 2000).

The story from Poland, involving the disbanding of the begrudged Słuzba Bez-pieczenstwa (SB), is less well-known, but gives a useful illustration of the complexitiessurrounding thorough purges.

As of July 1989, the SB employed 24,107 officers. After an ordinance of the Minis-ter of Internal Affairs was implemented, this number dropped to 6681. Entire depart-ments making up the SB—specifically II, III, IV, V and VI—were liquidated.5 Thesereforms, however, did not constitute a formal purge. According to Leskiewicz (2016),the 1989 Roundtable Agreement put Czeslaw Kiszczak (the former SB chief featuredin the previous chapter) in charge of the Department of Internal Affairs. As Minister,he was left responsible for reorganizing the SB. Taking advantage of this privilege,he reassigned the most compromised employees from the dissolved departments intothe Citizens’ Militia units (the Citizens’ Militia, Milicja Obywatelska, was the formercommunist state’s police force).6 It was not until two members of the former oppo-sition proposed legislation creating completely new police and security agencies that

that more than 500 ex-Stasi operatives had been taken over by the police. This in-cludes 161 former full time Ministry for State Security employees and 262 unofficialcollaborators. In addition, 370 ex-members of the DDR criminal police were in em-ployment in 1994” (195).

5 These departments were responsible for counterintelligence, combating anti-stateactivities, infiltration of religious organizations, protection of the state economy,and protection of the state agriculture, respectively.

6 The minister made use of two regulations, “Ordinance number 890 MSW of January22, 1990 changing the ordinance about the status of MO and SB and ordinance” and“Ordinance number 075/89 of August 24, 1989 on the liquidation and reconstruc-tion of some organizational units of the Ministry of Internal Affairs.”

3.1 Illustrative examples of purges 49

the prospect of actual purges in Poland’s security apparatus became a reality. As aresult of the bill, any new operations, including those on existing cases under SB in-vestigation, would be terminated. Officers conducting espionage, counter-espionage,and investigations into economic crimes would have to stop working immediately. Per-mitted to continue were only anti-terrorist activities and a few general service depart-ments such as the bureau for deciphering, the communications bureau, the passportdivision, and the population registry division.7

The leadership of what remained of SB hoped to be awarded positions in the newOffice for State Protection, Urzad Ochrony Panstwa (UOP), even as the activities of theUOP, which included combating international terrorism, white collar crime, and espi-onage, were quite different from those that had been the focus of the SB. Despite thesepreferences for continued employment, the bill passed in the legislature in early 1990.A critical part of the act addressed the fate of former SB employees. Not only were theexisting 6681 SB staff to be let go, but the purge extended to all officers who were em-ployed by the SB as of late July 1989. As a result, the purge included staff that Kiszczakhad intended to shield through transfers to the Citizens’ Militia. The Citizens’ Militiawas disbanded and a new police force was created in its place.

Implementing the new legislation was complicated by the fact that Kiszczak wasstill at the helm of the Department of Internal Affairs. In June 1990, Prime MinisterTadeusz Mazowiecki indicated to him that “there cannot be peace” while he remainedMinister. The following day, Kiszczak resigned, and his deputy, Krzysztof Kozlowski,took over. Later in his memoirs, Kiszczak expressed deep regret “that he was not in aposition to protect the staff that had worked under him for so many years especiallysince despite agreement from the Roundtable and peaceful power transition, calls forrevenge and accountability intensified” (Beres n.d.).

With Kiszczak ousted, Kozlowski was also in a position to purge the leadershipof the organization. This purge started with the dismissal of the second vice-minister,five generals, and 16 department directors as well as their immediate deputies, totaling202 staff members. Once the leadership had been purged, he was able to implementthe newly-passed legislation, which called for the firing of all employees of the SBand their rehiring only after they had been vetted by stringent verification commis-sions that had been provided for by the new legislation. Verification was coordinatedby a Central Verification Committee (Centralna Komisja Weryfikacyjna), chaired by Ko-zlowski himself. The central committee oversaw the work of 50 regional committees.The regional committees were made up of the UOP regional chief, the regional policechief, the head of the regional police, a police trade union representative, and personswho “had earned the trust of the local community and boasted high moral authority.”In practice, the additional verification committee members included two members ofparliament, one senator, a regional “Solidarity” trade union activist and a few personsnominated by the UOP chief.

Any former member of the SB (defined by his workplace prior to Kiszczak’sreshuffle) was allowed to apply for a position in the UOP as long as he or she wasunder 55 years of age. In the application, the candidate had to explain how his or herservices would be useful under the new democratic regime. Only candidates clearedof any suspicions were admitted to service in the new UOP and police force. Veri-

7 The very fact that these departments had been subsumed under the secret policein the communist state only indicates how extensive the reach of the secret policehad been under communism in Poland.


fication committees were featured in Wladyslaw Pasikowski’s 1992 film “Psy.” Thefilm, quoted in the beginning of this chapter, opens with the Chair of the verifica-tion committee reading out loud the file of the main protagonist, Franz Mauer. Sincethe warnings and sanctions in his file far outnumbered the honorable mentions, theChair asked if Mauer felt obligated to “faithfully serve the new Polish Republic.” Franzreplied: “I do, to the very end, be it mine of hers.” And he was rehired.

According to Slawomir Dudek, of all SB employees as of mid-1989, only 14,500—that is, 60 percent—applied to be vetted. Initially, only 8681 convinced the verificationcommissions of their usability, but 4,500 more appealed the regional decisions; amongthose, 1,800 had their initial decisions reversed. This brought the total number of for-mer employees positively vetted to be rehired by the UOP to 10,439 and the numberof those who failed to 3,595. As one of Kozlowski’s deputies pointed out, successfullypassing verification merely made a candidate eligible for reemployment. It is not clearexactly how many of the over ten thousand were ultimately rehired (Dudek & Gryz2003).

The reason this Polish example qualifies as a complete purge is that the entirestructure of the SB was dismantled, with all of its employees collectively fired. Allofficers old enough to remember the repression of the 60s were forced into retirement.The remainder of SB staff was only permitted to reapply for employment in the newagencies after each case received separate individual consideration.

A leadership purge, in contrast to a thorough purge, is limited to the top echelon ofthe hierarchy of the enforcement apparatus: it discriminates between the leadershipof the organization and the rank and file. A good illustration of a leadership purgeis the Bulgarian Panev Law, passed on December 9, 1992 by the Bulgarian NationalAssembly.8 Among its many provisions, the law prohibited from holding positionsin “Executive Bodies of Scientific Organizations and the Higher Certifying Commis-sion” people who had taught at the Communist Academy for Social Sciences and So-cial Management and those who had taught History of Communist Parties, Leninist orMarxist Philosophy, Political Economy, or Scientific Communism. All persons coveredby the law had to provide written statements regarding their prior employment andparty activities. A refusal to provide such a statement was regarded as an admissionof guilt. According to its author, Georgi Panev, the underlying idea behind the purge“was to bar persons of the higher totalitarian scientific structures and former collab-orators of the former State Security from academic and faculty councils and from thesupreme academic awards commission, awarding scientific degrees and other aca-demic qualifications.” As this ban did not extend to all academicians, but rather onlyto those who had trained the communist state’s tight leadership, this ban can be con-sidered a leadership purge. Most academic staff who had worked under the previousregime retained their jobs.

Finally, a perfunctory purge is a vetting process that only extends to the rank andfile or to the lowest echelons of the hierarchy of the enforcement apparatus. Suchpurges may occur while the authoritarian regime is still in office, in which case, theyare a preemptive move by the outgoing authoritarian regime. Such preemption shieldsthe regime from more severe transitional justice at the hands of the incoming demo-

8 The full name of the bill was “Law for Temporary Introduction of Additional Re-quirements for Members of Executive Bodies of the Scientific Organizations andthe Higher Certifying Commission.”

3.2 The Model 51

cratic regime (Kaminski & Nalepa 2014).9 A perfunctory purge can also be institutedby an incoming democratic government when it cannot conduct a leadership purgebecause its hands are tied by a peace agreement with the outgoing military autocrats(Nalepa 2010b). Under such circumstances, forgoing purges altogether is not feasiblebecause of third party or international pressures. Such a purge occurred in Nicaragua.Nicaragua’s autocracy ended in 1979, when the former Nicaraguan president, Anasta-sio Somoza Debayle, was defeated by Sandinista National Liberation Front (FSLN orthe“Sandinistas”). The new government had to manage a civil war that broke out asthe CIA-backed Contras staged terrorist attacks over the course of the next eight years(Hoekstra n.d.). Fighting finally came to a close when Violeta Chamorro emergedvictorious in the 1990 elections. Although preferred by the US to the Sandinistas, shestill faced considerable pressure from the United States to reform the security forces.10

These pressures notwithstanding, Chamorro decided to forgo a purge of military com-manders and only purge troops, reducing their number from 80,000 to only 20,000.Notably, while Chamorro had long been a Sandinista critic, she retained General Or-tega, the former president and FSLN leader, most probably in order to preserve peace.Three months later, when she did fire a long-time Sandinista Chief commander tomake good on her promise to lead a “government of national reconciliation,” she re-placed him with another Sandinista (Otis 1992).

A second perfunctory purge example comes from El Salvador. Like Nicaragua,El Salvador had gone through a party-based authoritarian government, classified bysome as party-military (Geddes et al. 2014), followed by a civil war that left 75,000people dead. The first competitive elections free from political violence were held in1994, though the peace agreements were signed earlier, in February of 1992 (Archiven.d.). On January 4, 1993, despite pressure from the UN to carry out a leadershippurge in the military, President Cristiani, to demonstrate “good faith over the peaceaccords ..., announced reductions in the armed forces from 63,000 to 31,500 troops.”However, the top echelon of the leadership was mostly preserved (Renewed threats topeace process Purge of army officers 1993).

3.2 The Model

The lustration and truth commissions model presented in the previous chapter showedhow transparency regimes enhance democratic representation by preventing black-mail of current politicians by former authoritarian elites. Lustration, for example, ex-poses potentially embarrassing information about collaboration with the authoritar-ian regime’s enforcement apparatus, making it impossible for former authoritarianelites to extort policies in exchange for keeping skeletons in politicians’ closets hid-den. A crucial theoretical prediction was that democratic representation improves indirect proportion to the amount of lustration implemented.9 Technically, preemptive perfunctory purges do not satisfy the definition of transi-

tional justice; for this reason, they are omitted from the Global Transitional JusticeDataset.

10 At one point, the US Congress threatened to withhold the release of 100 milliondollars to Nicaragua unless some personnel changes took place. This pressure wasin part due to charges that US funds had been funneled to Sandinista groups; thiswithholding of aid may also be interpreted as pressure for a purge.


This chapter is devoted to modeling purges. The models I present here yield pre-dictions for the effect of purge severity on democratic representation that differ starklyfrom the predictions I make on the effect of lustration. While the quality of politicalrepresentation increases monotonically with the severity of lustration, the severity ofpurges may be associated with worse effects for democratic representation. Moreover,the very feasibility of implementing purges depends on conditions present in the au-thoritarian regime itself: following a purge, new democratic leaders must replace thefired agents with new staff who share the leaders’ preferences but lack the skills nec-essary to implement policy. Whether this is feasible depends on two conditions: (1)uncertainty at the time of the transition and (2) the expertise of the former authori-tarian bureaucrats.

Consider the following general dilemma: following a transition to democracy, anincoming government inherits a state apparatus that is staffed by employees of theformer authoritarian regime. These employees possess expertise that the incominggovernment lacks. Yet, these bureaucrats or uniformed members of the enforcementapparatus may well have residual loyalties to the ancien regime.11 Thus, their pref-erences diverge from those of the incoming democratic government. In light of this,they may use the expertise they possess from their service in the ancien regime to sab-otage the policy program of the new democratic government. Given the risks posedby these bureaucrats, the incoming government faces the choice to purge the stateadministration—all personnel employed by the former autocrats or to retain most ofthe state employees and only focus on the leadership.

The ideal bureaucracy possesses three characteristics. First, bureaucrats ought tohave more expertise than rulers or lawmakers themselves. One way in which statesensure that this happens is by mandating entry exams and requiring that minimumqualifications be satisfied before a person can become a civil servant (Simon et al.1950, Wilson 2019). Second, bureaucrats are supposed to share a “commonality ofinterests with the rulers” (Weber 1968) to ensure that the policy they try to implementclosely or exactly matches the ruler’s ideal outcome. To the extent that a gap existsbetween the preferences of rulers and their bureaucrats, a “principal-agent” problemarises. Third, bureaucrats should not shirk their responsibilities. In other words, whennot monitored, they should work exactly with the same intensity as when they aremonitored (Brehm & Gates 1999). Shirking is the only feature of the three listed abovethat is not accounted for in the models I present.

It is easy to see why a bureaucracy in the aftermath of a democratic transition mayfall short of these three ideals. Bluntly put, the inherited bureaucracy may be incom-petent and disloyal to the new regime. First, the state of the ancien regime may havebeen staffed via patronage networks and hence lack the expertise necessary to imple-ment policy. Indeed, authoritarian regimes are often brought down by poor policy

11 Note that in this section, I assume that generically, persons working under the pre-democratic regime depart from the Weberian ideal type of a bureaucrat in that theydo not automatically acquire the preferences of the “ruler” or establish what We-ber refers to as a “community of interests” (Weber 1968). On the contrary, sinceworking for an ideologically tainted authoritarian regime requires at least someideological commitments on the part of the administrative staff, those loyalties aredifficult to shed immediately.

3.2 The Model 53

outcomes and subpar economic performance(Gasiorowski 1995). 12 Second, employ-ees of the previous authoritarian regime’s state apparatus frequently have loyalties tothe outgoing regime: working in the authoritarian state was a moral choice made onlyby those who, in one way or another, accepted the authoritarian ideology and the lackof competitive democratic elections that would hold their political bosses accountable.Third, regardless of his or her ideological leanings, the autocrat’s departure can be as-sociated with “orphaned bureaucracies” (Ang 2019), which obviously contributes tothe preference gap between the new democratic politicians and state apparatus staffmembers inherited from the previous regime. “Orphaned” bureaucrats are also likelyto be shirkers, though, as remarked above, I will not be modeling this feature below.13

On the other hand, the bureaucrats of the ancien regime may be equipped in exper-tise that translates well into the conditions of the new democratic polity. In the wordsof Anna Grzymala-Busse (2002), they may have usable skills that the new democraticpolitician desperately needs.

My framing of the purge problem as a principal-agent dilemma draws on modelsof delegation from the institutions literature in American and comparative politics(Callander et al. 2008, Huber & Shipan 2002, Epstein & O’Halloran 1999, Bendor &Meirowitz 2004). The classical analysis of delegation was offered according to Bendor& Meirowitz (2004), by Alexander Hamilton, who defended transferring control fromless informed officials to more informed agents in anticipation that because they arebetter informed, they will make better choices.14

Delegation models assume a single principal who can either act on his own or dele-gate authority over policy to another actor, referred to as the agent. The principal in thedelegation literature is typically a legislature, and the agent is typically a bureaucratwho, equipped with expertise, can implement policy better suited to accommodatestates of the world that are unfamiliar to the legislature. However, delegating policyauthority to the bureaucrat comes at a risk, as he may use his expertise to move the ul-timate policy outcome away from the legislature’s ideal point; this is especially likelyif the preferences of the legislature and the bureaucrat are misaligned. The delegationliterature refers to this phenomenon as bureaucratic drift (Gehlbach 2013).

Building on this framing, I treat a thorough purge as analogous to a refusal to del-egate to possibly unfaithful agents. In doing so, I assume the new democratic regimelacks expertise to implement policy as efficiently as employees of the ancien regime.Thus, even though the democratic successors know the policy outcomes they want,and I assume that this policy corresponds to voter preferences, they cannot anticipatethe policy distortions that arise at the time of implementation. Forgoing a thoroughpurge in favor of a leadership-only purge, on the other hand, can be interpreted as adecision to delegate policy choice to agents of the former authoritarian regime.

12 However, Remmer (1990) shows how economic vulnerability is the consequencerather than the cause of democratization.

13 For a comprehensive treatment of bureaucracies that accounts for shirking and sab-otage, seeBrehm & Gates (1999).

14 According to this traditional interpretation, expressed by Hamilton in FederalistNo. 23, choosing optimal policies is state contingent. Since principals are obliviousto the true nature of the state, they are best served by delegating to agents whoknow the real state of the world.


In the next section, I illustrate the application of thedelegation model to the prob-lem of purges using Gehlbach (2013)’s the standard presentation of the problem. Iuse the first two models from Chapter 5 of his Formal Models of Domestic Politics. In thefirst of the two models, the new democratic politician considers two options: to purgeor not to purge. In the second model, which is an interpretation of Huber & Shipan(2002), the politician is able to set discretion limits for the old regime’s administrativestaff if he refrains from a purge. After adapting these two classical delegation mod-els to the purge-specific context, I present a third (original) model, where I relax theassumption of perfect expertise on the part of the agent.

3.2.1 The decision to purge as a simple delegation model

I begin by with considering a one-dimensional policy space sketched out in Figure3.1. The model features a Politician, P , with ideal point 0, and a former enforcementAgency Officer, A, with ideal point xAε(0, 1). In the first stage, the Politician deter-mines whether to conduct a purge. This action corresponds to the decision to dele-gate in the classic principal agent framework. Conducting a purge is represented asP and is the equivalent of not delegating power to the agent of the former authori-tarian regime. Alternatively, in an action corresponding to delegation from the classicmodel, the new democratic politician may choose to retain the staff of the authoritarianagency (this action is labeled as ∼ P ). Following the Politician’s decision, Nature de-termines the policy shock to be applied following the choice of policy. This shock cor-responds to the inherent uncertainty under which policy decisions are being made inthe transition aftermath. I represent this shock as ωε{ε,−ε}. To keep things tractable,there are just two possible shocks: one negative and one positive. I will assume that ei-ther policy shock is equally likely: that is, Pr(ω = ε) = Pr(ω = −ε) = 1

2. This policy

shock is observed by the Agency Officer, but not by the Politician. In the third and ter-minal stage of the game, the Politician (in the event of a purge) or the Agency Officer(in the event of refraining from a purge) chooses the policy. Of course, in reality, eventhe Politician who purges the administrative apparatus does not herself implementpolicy. However, given that following the purge she can only appoint a loyal formerdissident with no expertise, it is as if she were implementing policy herself.

The preferences of the players are determined by the distance between their idealpoints and the policy that is the final outcome of the game. This final outcome, x, is thejoint product of the policy choice in the second stage and Nature’s determination ofthe policy shock, according to the expression x = p+ω. The strategies and preferencesare defined formally in the appendix, but the game tree below summarizes the players,strategies, and the timing of play.

Fig. 3.1: Issue Space (S) and Players’ Preferences

SUP UA

0 xA 1

3.2 The Model 55

Since this is a game of complete, albeit imperfect, information, the solution conceptis Subgame Perfect Nash Equilibrium. The game tree representing the baseline modelis presented below.

Fig. 3.2: Baseline Purges Model

P

Nature

ωε{−ε, ε}

P

pεS

−|p+ ω|,−|xA − (p+ ω)|

P

Nature

ωε{−ε, ε}

A

pεS

−|p+ ω|,−|xA − (p+ ω)|

∼ P

This game can be solved by backward induction. I first consider the right hand sideof the the game tree. Note that since the Agency Officer observes ω, he can implementpolicy to “perfectly absorb” the exogenous policy shock and bring the final outcometo his ideal point. That is, when the shock is positive, he chooses xa − ε and when itis negative, he chooses policy p∗ = xA + ε. In light of this, the politician’s expectedoutcome from a leadership purge is −xA.

Consider now the left hand side of the game tree in Figure 3.2. Since the Politiciancannot observe the policy shock, his best response, given the symmetry of ω, is p = 0.Note that actuall,y any pε[−ε, ε] is a best response; however, p = 0 is robust to changingthe utility functions from “tent” to quadratic.15 As a result,the expected utility to thePolitician from purging is given by −ε.

15 Bendor & Meirowitz (2004) show that changing the players’ preferences in this waydoes not change the end result of the analysis.


In Figure 3.3 below, I compare the realized policy outcome (in green) under the nopurge scenario with the realized policy under the purge scenario (in red) as a functionof the magnitude of the shock, ε.

The green line in the upper quadrant corresponds to the realized outcomes fol-lowing a purge when the policy shock is positive (hence the increasing slope), whilethe green line in the lower quadrant corresponds to the negative policy shock scenario(hence the decreasing slope).

Both lines have non-zero slopes because when the Politician purges the formerstate apparatus of members of the ancien regime, the price he pays is proportional tothe degree of uncertainty. Recall that the Politician is trying to minimize the distancebetween this realized policy and his ideal point, which is at 0 and so correspondssimply to the x-axis. The extent to which the final outcome departs from the Politician’sideal point increases, intuitively, with the size of the shock.

The outcome under the purge scenario, represented by the red line in Figure 3.33.3,corresponds to the realized outcome without a purge. This outcome does not dependon the value of ε. The two lines cross at the point where ε = xA, implying that for val-ues where ε is below xA, the Politician is better off purging the administrative appa-ratus, but for ε > xA, retaining the authoritarian administrative apparatus producesa policy closer to the Politician’s ideal point.

Backward induction leads us immediately to the solution to this baseline model:the Politician will purge if ε < xA and refrain from doing so if xA ≤ ε. In other words,the Politician will refrain from conducting a purge when the uncertainty associatedwith the policy shock is greater than the preference divergence between the Politicianand the Agency Officer. This result is merely an application of Bendor & Meirowitz(2004) to purges.

While the parameter xA is easy to interpret as the divergence in preferences be-tween the new democratic Politician and the Agent of the former authoritarian regime,the interpretation of ε is perhaps somewhat less intuitive. It is best thought of as a mea-sure of uncertainty following democratic transitions: transitions that are more abruptand lead to considerable economic, social, and political unpredictability can be in-terpreted as high ε environments, while smooth protracted transitions, offering morepredictability, can be interpreted as low ε circumstances. Alternatively, one may thinkof ε as the skills of the new democratic politicians and their ability to appoint and trainnew administrative staff.

3.2.2 Discretionary limits

The next model I consider not only allows the Politician to decide between purgingand refraining to do so, but also allows him to set “discretion limits” if he choosesthe former. Discretion limits are interpreted as lower and upper bounds on where theAgency Officer may implement policy p. This allows the Politician to retain some con-trol, while also drawing on the agent’s expertise at the same time. Empirically, thesecan be directives on how a specific policy is to be implemented, what the chain of re-porting and monitoring policy outcomes is supposed to be, or how far policy changescan reach. In their path-breaking book, Huber and Shipan operationalize discretionarylimits with the number of words contained in each bill from a given jurisdiction. Thisoperationalization relies on the assumption that less discretion implies more wordsHuber & Shipan (2002).

3.2 The Model 57

It is important to note that whether discretionary limits can be imposed at all de-pends on the institutionalization of the new democratic regime, which in turn dependson the institutionalization of the ancien regime, a topic I model explicitly in section3.2.3.

Circumstances where discretion limits may be useful when a standard delegationmodel fails are easy to imagine. Consider any case where the preference divergence be-tween the new democratic politician and outgoing administrative staff of the formerregime is considerable, but where the staffers’ expertise is so valuable that purgingthem could pose a dangerous threat to the new democracy. As an example, considerthe decisions facing the post-apartheid regime in South Africa. The preference of theoutgoing Apartheid regime and the ANC led government clearly diverged. At thesame time, with over 30 years of tenure, the South African regime had trained skilledbureaucrats who were no doubt better at implementing policies than any new work-ers would be. All these bureaucrats were white. Yet the ANC refrained from purg-ing them and instead attempted to set clear discretion limits to compel these agentsnot to exploit their knowledge to the new regime’s disadvantage. Had such tools ofcontrol been unavailable, the ANC government would have had to purge apartheid-era bureaucrats. Indeed, at the time of the transition, such a purge was the subject ofuniversal fear as memories of Rhodesia, now Zimbabwe, and the “white flight” werereferenced more frequently than any other example Southall (2020).

The first goal of solving this model is to see whether discretionary limits, whenpresent, change the relationship between uncertainty (represented by ε) and prefer-ence divergence (represented by xA) that was identified in the baseline model.

In this model, I retain all assumptions about players and preferences from thebaseline. Hence, the Politician has an ideal point at 0, and the Agency Officer has anideal point at xA.

The strategy set of the Politician is modified to reflect that he sets the discretionlimit when he forgoes a purge. However, given the preferences of the Politician andAgency Officer (at 0 and xA, respectively), it is clear that there is no need for thePolitician to set a lower bound for the discretion limits: the Agency Officer would neverwant to implement policy to the left of the Politician’s ideal point. Since A’s ideal pointis to the right of that of P , he always prefers a realized policy at the Politician’s idealpoint to a policy at any point to the left of this. Thus, the left bound on the discretionarylimits need not be set at all, and the action corresponding to the discretion limit canbe represented by just one parameter, rε<+.

A direct consequence of imposing discretion limits is that the Agency Officer nolonger has perfect shock absorption. In the model, this is represented by setting thediscretion limit so that r < xA + ω. With this setting, the Agent can never achieve hisideal point, so the Politician may refrain from purging him without fearing that thepolicy will swing too far.

The game tree of this version of the model is provided in Figure 9.1 of Appendix9.2.2., where it is also solved for Subgame Perfect Equilibrium through backward in-duction.

The reasoning behind finding the optimal discretion limit, r can be summarizedas follows: first, note that if P conducts a purge, the optimal policy is, as before, p = 0,with an expected payoff of −ε. If the Politician chooses to forgo a purge, he has to setthe discretion limit, r, to maximize his utility. In Appendix 9.2.2, I focus on examiningfocal discretion limits such as r = ε. For this specific discretion limit, the Politician’sexpected utility is −xA

2, as shown in the appendix.


To see that the discretion limit r = ε is the unique best choice, consider two pos-sibilities: (1) that r is set higher than ε, and (2) that it is set lower than ε.

If the discretion limit is set higher, to some r = ε + δ, A would have to set policyto min{ε + δ, xA − ε} for ω = −ε and to min{ε + δ, xA − ε} for ω = ε. This wouldresult in P ’s expected utility −xA+δ

2. If, on the other hand, the discretion limit is set

lower, to some r = ε − δ, A would propose p = min{ε − δ, xA + ε} for ω = −ε andp = min{ε− δ, xA − ε} for ω = ε, leading to P ’s expected utility − δ

2− xA

2.

Since in both cases, P ’s utility is lower than in the instance where r = ε, I concludethat the Politician’s optimal discretion limit is r = ε. P ’s expected utility with this dis-cretion limit leads to the conclusion that P will forgo a purge as long as xA < 2ε. Thisresult can be contrasted with the baseline model. It is clear that when the Politiciancan set discretion limits, he will stop short of a purge even if the administrative appa-ratus inherited from the ancien regime diverges in its preferences from those of thePolitician twice as much as in the baseline model (where discretion limits were notavailable).

outcome

ε

Purge if ω = +ε

Purge if ω = −ε

no PurgexA

xA

(a) Policy Outcomes without Dis-cretion Limits (baseline model)

outcome

ε

Purge if ω = +ε

Purge if ω = −ε

no PurgexA

2

xA

2

(b) Policy Outcomes with Discre-tion Limits

Fig. 3.3: Policy Outcomes in Baseline and Discretion Limits Model of Purges

The rationale behind P ’s choice is captured in Figure 3.3 , where the red line cor-responds to the expected outcome from the leadership purge and the green line to theexpected outcome from the thorough purge. As soon as ε exceeds xA

2, a leadership

purge generates higher utility for the Politician. To see this, note that given the Politi-cian’s ideal point at zero and the fact that the utility function is the negative absolutevalue of the distance between the ideal point and the policy outcome, we only needto focus on comparing the upper green line to the red line. The left panel of Figure3.3 allows for a direct comparison of the model with discretion limits to the modelwithout. It is clear that discretion limits allow politicians to refrain from purges in

3.2 The Model 59

a wider spectrum of situations—including even those that are not characterized bymuch uncertainty—than when implementation of discretion limits is not feasible.

3.2.3 Modeling the value of expertise

The purge models above both assume that the Agent of the ancien regime has com-plete and perfect expertise in the sense that he can always, in the discretion model—ifgiven the sufficient latitude—would move the outcome exactly to his ideal point16. Inthis section, I relax this assumption and instead allow the expertise of the agent to vary.This reflects the fact that not all autocrats institutionalize their regimes and offer theirbureaucrats the opportunity to become skilled technocrats. Such instituionalizationincludes creating tenure in the state administration, offering clear paths of advance-ment within the ranks, and attracting the most talented and skilled professionals byoffering attractive wages or job security.

In order to formalize the expertise that the Agency Officer has to offer, I will relaxthe assumption that the Agency Officer gets to perfectly observe the exogenous shock.Note that in some circumstances in the model with discretionary limits, A did not getto implement the policy that would completely absorb the distortion stemming fromthe move of Nature. Now, I will assume that instead of observing the direction of theexogenous shock directly, the Agency Officer is given a signal of the shock; this signalcan be of high or low quality. When the signal reflects the true nature of the policyshock, the Agent is better able to absorb the shock and bring policy towards his idealpoint; when it does not, he cannot absorb the shock as well. This is an intuitive way ofmodeling expertise: bureaucrats with poor expertise are often thought of as bad whenit comes to “reading signals” from their area of specialization.

Where do employees of the ancien regime’s state apparatus acquire expertise? Inother words, what determines whether they have better or worse capacity for shockabsorption? A natural interpretation would be the level of institutionalization of theprevious authoritarian regime. As Rauch & Evans (2000) point out, regimes, includ-ing authoritarian ones, vary considerably in the extent to which their bureaucraciesrely on meritocratic recruitment and internal promotion instead of on nepotistic ap-pointments and the promotion of agents for loyalty as over skills.

Herein lies the main departure of my model from existing models that extendthe traditional approach described in section 3.3, such as those of Huber & McCarty(2004) and Bendor & Meirowitz (2004). Huber & McCarty (2004) allow for agentsto have imperfect expertise, but model it as additional noise, µ, that accompanies thepolicy implementation of the agent. As the magnitude of this noise goes to zero (rep-resenting an agency better skilled at policy implementation), the condition for dele-gation reduces to the one from the baseline model. Correspondingly, as µ increases,any benefits from delegation become outweighed by the uncertain performance of theagent. Translating these insights to the case of purges would mean that even after ifno purge is implemented, the agent of the ancien regime can still make mistakes inhis implementation of policy, and his capacity to overcome these mistakes is reflectedin µ. The question then becomes: when does the new democratic politician retain alow-capacity agent of the ancien regime? The answer boils down to the severity of thetradeoff between the divergence in preferences between the new and old regime andthe capacity of the former bureaucrats of the new regime.16 The literature refers to this as “perfect shock absorption”(Gehlbach 2013)


An alternative approach to modeling bureaucratic capacity or expertise is to intro-duce uncertainty to the way that the signal about the exogenous shock is transmitted.This is the approach of Bendor & Meirowitz (2004), who assume that with a certainprobability, π, the agency receives a perfectly informative signal of the value of therandom shock, ω, with probability 1 − π the agency receives a perfectly uninforma-tive signal, adding nothing at all to what the principal already knew. After solvingthis model, the authors find that the original tradeoff between preference divergenceis directly proportional to the informativeness of the signal; that is:

πε2 > x2A (3.1)Next, Bendor & Meirowitz (2004) go on to model π as an endogenous choice of

the agent. In a democratic context, this is a very sensible extension, as it is indeed upto the bureaucrats themselves to determine whether or not to acquire expertise. Theextension of this is less apt in the context of post-authoritarian purges, where all theexpertise to be spoken for was either acquired in the past or not at all. Another sensiblealteration is to change the signaling technology to reflect the fact that expertise canvary on a continuum and need not be symmetric with regard to detecting positive ornegative signals.

In light of the discussion above, the new institutionalization parameter that I in-troduce here,s, will allow me to account for the fact that the expertise of the agentis highest (that is, the probability of a correct signal is much higher than the prob-ability of an incorrect signal) when the preceding authoritarian regime was highlyinstitutionalized, and lowest when the preceding authoritarian regime was poorly in-stitutionalized.

The model starts with a move of Nature expressed as sε(0, 1), where 1 representsthe least possible expertise (and so a very poorly institutionalized ancien regime),and 0 represents the highest possible expertise. The parameter s is then related to theprecision with which the Agency Officer can read the signal that the policy shock wasε and not −ε. For this purpose, I define

Pr(s = i|ω = j as the probability that A receives a signal i when the signal is i∀i, jε{−ε, ε}.(3.2)

As a result,

Pr(ε|ε) is the probability that the signal the officer receives is high when the policyshock is indeed positive (correct signal).Pr(−ε|ε) is the probability that the signal the officer receives is low when the pol-icy shock is positive (incorrect signal).Pr(−ε| − ε) is the probability that the signal the officer receives is low when thepolicy shock is indeed negative (correct signal).Pr(ε| − ε) is the probability that the signal the officer receives is high when thepolicy shock is negative (incorrect signal).

Next, I assume that ∀i, jε{ε,−ε}:

1. Pr(i|j)ε(0, 1)2. Pr(i|i) + Pr(j|i)) = 13. Pr(i|i) > Pr(j|i)

These three conditions formalize intuitive expectations. Condition 1 states that there isno perfectly correct or perfectly incorrect signal. Condition 2 states that given a policy

3.2 The Model 61

shock, either a low or high signal must be issued. Condition 3 states that the signalsare not perverse. That is, the probability of an incorrect signal for a given policy shockis never higher than the probability of a correct signal.

The probabilities defined above are next used to introduce the institutionalizationparameter, s as follows:

s =Pr(−ε|ε)Pr(ε|ε) =

Pr(ε| − ε)Pr(−ε| − ε) (3.3)

Therefore, s = 0 when A never receives an incorrect signal (perfectly informative),and s = 1 when the likelihood of a correct and incorrect signal are equal (no informa-tion). I bind s between 0 and 1 so that the signal is always imperfectly informative.

After observing s, the Politician decides whether to implement a purge. In theevent of a purge, he is forced to implement policy by himself without knowing if thepolicy shock will be positive or negative. As in the two previous models, he imple-ments his ideal point, 0. As a result, his expected utility from a purge is −ε. On theother hand, if P decides to forgo a purge, then the Agency Officer gets to observe aninformative signal about the quality of the shock. As explained above, the informative-ness of this signal is determined by the institutionalization parameter, s, as follows:P (ε) represents a signal that the policy shock is high, and P (−ε) represents a signalthat the policy shock is low. After observing the signal, but not the shock, the officerdecides which policy to adopt. Since his decision is contingent on the kind of signal hereceived, two actions are needed to define his strategy. p represents his action whenthe signal is high (that is, the policy shock, ω, is equal to ε), while q represents hisaction when the signal is low (that is, the policy shock, ω, is equal to −ε).

Preferences could be Euclidean. However, for ease of solving maximization prob-lems, I switch the first of the players to quadratic preferences; that is, the Agent’s payoffis given by:

UA(s; p, q) =

{−(xA − (p+ ω))2 if signal is ε−(xA − (q + ω))2 if signal is− ε.

The Politician’s payoff is defined analogously but for a corresponding ideal pointof 0:

UP (s; p, q) = −(p+ ω)2

The strategy sets of both the Politician and Agency Officer are defined in Appendix9.2.3, but they can also be easily gleaned from inspecting the game tree in Figure 3.4below.

Note that the graphic representation (for aesthetic reasons) is missing the moveof Nature determining s, the level of institutionalization in the very first stage of thegame. Also, in keeping with convention, all moves of Nature take place in the begin-ning of the game; the sequence of the game has the Politician moving directly afterNature’s determination of s. The game tree is accurate—note that all four nodes atwhich the Politician moves are in the same four-element information set. The AgencyOfficer knows s but does not know the realization of ω.

Since this is not a game of incomplete information,17 the solution concept is Sub-game Perfect Equilibrium. As before, the game can be solved by backward induction.In the appendix, I first calculate the optimal strategy of the Agency officer as a function17 Although the Politician determines whether the Officer must make a decision in

conditions of uncertainty, the signal itself is issued by Nature.


Fig. 3.4: Leadership Purges with Imperfect Shock Signals

NatureNature

Politician

Agent

( −(p+ε)2

−(xA−p+ε)2)

−ε2,−(xA − ε)2

Politician

Agent

( −(q+ε)2

−(xA−q+ε)2)

−ε2,−(xA − ε)2

Nature

Politician

Agent

( −(p−ε)2−(xA−p−ε)2

)

−ε2,−(xA + ε)2

Politician

Agent

( −(q−ε)2−(xA+q−ε)2

)

−ε2,−(xA + ε)2

High signal

Low signal

ε −ε

∼ purge ∼ purge

∼ purge ∼ purge

purge purge

purge purge

p

q

p

q

of s, xA, and ε, and then show the circumstances under which the Politician prefersimplementing a purge over refraining from one. The optimal policy choices p∗ and q∗are given by:

p∗ = xA −ε(1− s)1 + s

, q∗ = xA +ε(1− s)1 + s

(3.4)

Table 3.1 shows the Agent’s best responses in terms of p∗ and q∗ if the Politician re-frains from a purge. Recall that in the event of a purge, the policy outcome will alwaysbe−εwhen the realization ofω is low and εwhen the realization ofω is high. Building

3.2 The Model 63

Table 3.1: Equilibrium policy outcome as a function s

Institutionalization BRA(s) = p∗ q∗ outcome when ω = ε outcome when ω = −εs = 0 xA − ε xA + ε xA−ε

2xA+ε

2s = 1

4 xA − 35ε xA + 3

5ε xA − 45ε xA + 4

5εs = 1

2 xA − 13ε xA + 1

3ε xA − 89ε xA + 8

9εs = 3

4 xA − 17ε xA + 1

7ε xA − 4849ε xA + 48

49ε

on the formulas used to produce this table, the Politician’s utilities from deciding topurge or not can be calculated and compared. Notice that given the Politician’s idealpoint is 0, the utility from refraining from a purge can be calculated as:

EUP (∼ P, ω = ε) = −(xA−ε(1− s)1 + s

+ ε)21

s+ 1− (xA+

ε(1− s)(1 + s)

+ ε)2s

s+ 1(3.5)

when the shock is positive and

EUP (∼ P, ω = −ε) = −(xA−ε(1− s)1 + s

−ε)2 s

s+ 1−(xA+

ε(1− s)(1 + s)

−ε)2 1

s+ 1(3.6)

when the shock is negative. Because the shock is negative half of the time and positivehalf of the time, the Politician will prefer to refrain from a purge when:

EUP (L; p∗, q∗;ω = ε)

2+EUP (L; p

∗, q∗;ω = −ε)2

> −ε2 (3.7)

or

(xA−ε(1− s)1 + s

+ε)21

s+ 1−(xA+

ε(1− s)(1 + s)

+ε)2s

s+ 1+(xA−

ε(1− s)1 + s

−ε)2 s

s+ 1−(xA+

ε(1− s)(1 + s)

−ε)2 1

s+ 1< 2ε2

(3.8)The decision of when it is beneficial to refrain from purges is presented graphically

as a function of s in Figure 3.5 and discussed below. Figure 3.5 illustrates the expectedpolicy outcome (left panel) as well as the Politician’s utility outcome (right panel) asa function of s. Consider first the left hand side panel. The “whiskers” in blue and redoriginate at parameter values for xA1 = .2 and xA2 = .65, which I have chosen ascases. Since ε = .6, the first case (marked in blue) has been chosen so the preferencedivergence is lower than the exogenous policy shock, while the second case (markedin red) has been chosen so the preference divergence is higher than the policy shock.

The top line for each case reflects the policy outcome following a purge in theevent that the shock is positive; the bottom line reflects the policy outcome followinga negative shock. The dashed grey lines marked with−ε and ε represent where policywould be located were the Politician to forgo a purge (recall that his optimal strategywhen he forgoes a purge is to implement policy at his ideal point). We see that in thecase of divergent preferences, the policy associated with forgoing a purge is better forthe Politician only if the shock is positive (see the top red line). When preferences arealigned more closely, a purge is worse for the politician regardless of the direction ofthe shock. Using these two cases and simply looking at the policy outcomes does notsuggest that the optimal decision of the politician depends on s.


0.0 0.2 0.4 0.6 0.8 1.0

s

ω = ε

ω = − ε

ω = ε

ω = − ε

ε

− ε

− ε

−0.9

−0.6

−0.3

0

xA1

0.3

xA2

0.9

0.0 0.2 0.4 0.6 0.8 1.0

s

x_A = 0.2

x_A = 0.65

− ε2

−1.8

−1.55

−1.15

−0.9

−0.6

ε2

0

0.36

Fig. 3.5: Policy outcomes (left) and payoffs to the Politician (right) resultingfrom purges (blue and red) and non-purges (dashed grey)

Recall however, that the Politician does not observe the shock direction. For thisreason, to illustrate the Politician’s optimal decision, I turn to the right hand side panel,which compares the utility associated with both actions for the exact same values ofthe Agency Officer’s ideal point, xA1 = .2 and xA2 = .65. The payoff from purging isrepresented by the grey dashed line at ε2 = .36. Now, a more complex picture emerges.First, for divergent preferences, a purge is always associated with higher utility. This isshown by the fact that the grey dashed line is always above the red line (correspondingto xA = .65). For the case of aligned preferences (xA = .2), however, the action asso-ciated with the higher payoff depends on the insitutionalization parameter, s. Thus,for low levels of s, corresponding to high levels of institutionalization, the Politicianis better off refraining from a purge and harnessing to his advantage the skills of theancien regime’s administrative apparatus. Yet, as the level of institutionalization de-creases, the benefits from forgoing a purge declines and at some intermediary level ofs, the Politician is better off purging the ancien regime. The exact level of institution-alization at which a purge is more beneficial than retaining members of the formerregime is the intersection of the blue and grey lines. This is the point which solves:

xA =ε(1− s)2

(1 + s)2(3.9)

or equivalently:

ε =xA(1 + s)2

(1− s)2 (3.10)

Intuitively, for low levels of professionalization, the benefits of retaining the pre-vious regime’s employees—even those with preferences that are proximate to thoseof the new democratic politician—are outweighed by a purge. With little expertise tooffer, workers of the ancien regime cannot count on keeping their employment.

3.3 Discussion and comparative statics 65

3.3 Discussion and comparative statics

The discussion above implies that when the shock is positive (a factor that the Politi-cian cannot anticipate ex ante) and when s is lower—corresponding to a highly institu-tionalized prior regime—refraining from a purge is better than purging. Yet, when theshock is positive but s is higher—corresponding to low institutionalization—purgingmay be better. It is also clear from the blue line in Figure 3.5 that decreasing the mag-nitude of the shock, ε, moves the cutoff point to the left, requiring more institutional-ization before a forgoing a purge is preferred. If uncertainty over the shock is lowered,making policy decisions without any expertise at all should be cheaper.

Recall that the empirical interpretation of s is that the expertise of the agent ishighest (the probability of a correct signal is much higher than the probability of an in-correct signal) when the preceding authoritarian regime was highly institutionalizedand lowest when the preceding authoritarian regime was poorly institutionalized.

First, I expect that following highly institutionalized authoritarian regimes, politi-cians should refrain from purges. This is because the substantial expertise of agentsof the former regime is likely to help them choose policies that absorb the kinds of ex-ogenous shocks that plague regime transitions. Forgoing any type of purge altogethershould improve the quality of democratic outcomes relative to those in countries thatundertake purges when institutionalization is high.

On the other hand, following poorly institutionalized environments (Boix & Svo-lik 2013), forgoing purges is detrimental to the quality of democracy. Agents with lowexpertise are just as bad at correcting for policy shocks as the new politicians them-selves, but instead try to reach their own successor-authoritarian ideal point. Hence,a purge should bring about a better quality of democracy than doing nothing at all.

In a nutshell, the effects of both kinds of purges depend heavily on the degree towhich the previous authoritarian regime was institutionalized. This is because insti-tutionalization translates into higher levels of expertise among former agents of au-thoritarian agencies. Purging bureaucrats and officers equipped with skills usable formanaging the new democratic regime (Hicken & Martınez Kuhonta 2011) will likelysabotage policies of the new democracy and weaken the quality of representation.

In contrast, a poorly institutionalized authoritarian regime is staffed with agentswhose level of expertise is too low to warrant trading off their potential to skew pol-icy outcomes for precision of implementation; increased purging in formerly under-institutionalized regimes should lead to improving the quality of democracy. Theseeffects ought to be more prominent in the case of thorough purges than in that ofleadership purges because in the former, entire agencies are being disbanded.

These empirical implications will be illustrated with data from the Global Transi-tional Justice Dataset in Chapter 7.

3.3.1 Contrast with Representation under the Lustration Blackmail Model

We can now compare the transitional justice procedures that have been captured in myformal models; the key comparison from the point of view of the project of this book iswhether purges, like lustration, contribute to a higher quality of representation. Twocontrasts are immediately apparent. First, in the case of purges, the quality of repre-sentation is sensitive (and in some cases even critically so) to the distance betweenthe Politician’s and the Agency Officer’s ideal points. This can be seen for instance in


the comparative statics on the equilibrium policy outcome presented in Figure 3.5 ,where this outcome is highly dependent on how closely his preferences align withthose of the Agency Officer and the level of authoritarian institutionalization. Hence,unlike in the case of lustration, more purges are not always better, other factors heldconstant, for the quality of representation. Indeed, as the discussion in section 3.3 in-dicates, the optimality of the decision to purge is highly dependent on the amount ofuncertainty following the democratic transition and the disparity in preferences be-tween the Politician and Agency Officer. In some instances—most notably, when thepolicy shock is high—purging more is better from the point of view of the Politician.However, when uncertainty is relatively low, the quality of representation will not beserved by an extensive purge.

3.4 Conclusion

This chapter has discussed purges as a personnel transitional justice mechanism thatfunctions differently from lustration and produces different effects for the quality ofdemocratic representation. As in the previous chapter, the quality of democratic rep-resentation is operationalized as the politician’s ability to get his ideal point imple-mented. I assume that this ideal point corresponds to the preferences of the voters. Ofcourse, this assumption does not require agreeing with the dubious claim that electedrepresentatives always represent voters’ interests. Instead, I seek to uncover the dis-tortion to representation that originates specifically in not dealing with legacies of theancien regime, be these legacies secret (as in the case of lustration) or open (as in thecase of purges).

I have argued that unlike vetting of secret collaborators, administrative purgesmay easily be understood in the framework of a delegation model. Purging memberswho ran agencies of the former authoritarian regime is akin to purging bureaucrats;thus, administrative purges are the reverse of a delegation problem. If a new politicianexecutes an effective and thorough purge ridding the agency of all civil servants asso-ciated with the ancien regime, he or she must implement policy in inherently uncer-tain conditions. Without the expertise of prior administrators who ran the agencies,the new politician is hamstrung evaluating and implementing policy. On the otherhand, only purging the leadership (and not all lower-level servants) may be analo-gous to delegating to an agent with subject-matter expertise and thus the ability toimplement the new politician’s post-authoritarian policies.

Building on existing models of delegation, I developed a nuanced theory of purgesthat showed that the very feasibility of purges in the aftermath of a transition todemocracy depends on the uncertainty associated with policy implementation, theloyalty of former agents to the ancien regime, and the past regime’s institutionaliza-tion. The empirical implications of this model led to several hypotheses about the re-lationship between purges and the quality of democratic representation. Specifically,the effects of purges depend heavily on the degree to which the previous authori-tarian regime was institutionalized: institutionalization translates into the expertiseof former authoritarian agents. To the extent that this expertise is usable in the newdemocratic regime (Grzymala-Busse 2002), purging these agents will come at a cost tothe new democracy and negatively affect indicators of democratic quality. In contrast,under-institutionalized authoritarian regimes lack experts. Purging their bureaucra-cies should lead to improving the quality of democracy.

3.5 Conclusion to Part I: Normative implications 67

Although the models presented here make an argument against purges that relieson the information deficit created by the departure of agents of the state, one can eas-ily construct other reasons for which carrying out purges is a risky endeavor in a newdemocracy. One reason is suggested by the illustrative case from section 3.1, whichdescribe how former Stasi officers found permanent employment in Martin Schlaff’sholdings and engaged in money laundering and tax evasion operations across Eu-rope. The film Psy by Wladyslaw Pasikowski, which provided the opening quote forthis chapter, illustrates an even more gruesome dynamic from Poland: some of thefired SB officers joined organized crime groups, where their former bosses had alreadyestablished themselves in leadership positions and engaged in trafficking weapons,drugs, and minors. This effect of thorough purges—rank and file being forced to seekemployment with their former bosses—is not unlike the dynamic described by BenLessing in an article devoted to the unintended consequences of mass incarceration(Lessing 2017). According to Lessing’s argument, the inevitability or close inevitabil-ity of being confronted with the gang’s leader in prison makes rank and file gangmembers more likely to obey the orders of prison gangs when they are still on “theoutside.” The parallel argument for the case of purges would run as follows: deprivingrank and file SB officers of legal employment in the enforcement apparatus of the newdemocratic state forces them to seek employment with their former leading officersoutside of the state’s official agencies. Consequently, a thorough purge pushes formersecret police officers with usable skills into organized crime led by former leaders ofthe secret political police.

The Tunisian example offers further illustrations of this phenomenon. Again, inBouguerra (2014)’s account, purges of the Tunisian political police sparked fears ofinternational terrorism. Would the agency and its brand new staff be capable of ro-bust counterterrorism and counter-extremism measures? Already in October 2014, theHarriri Center for the Middle East of the Atlantic Council feared that “the weakenedstate of the Tunisian government following the revolution, combined with the generalamnesty decree for and political polarization between secularists and Islamists, al-lowed for a resurgence of jihadist groups.” The thousands of casualties that have sinceresulted from terrorist attacks (Masri 2017), and the vastly successful ISIS recruitmentcampaign carried out on Tunisian soil, illustrate the unintended consequences purgescan have on crime.

3.5 Conclusion to Part I: Normative implications

This chapter concludes the contribution that this book makes to theorizing who is se-lected into office in a new democracy, how such representatives behave, and how theydelegate policy-making decisions to bureaucrats. The two formal models presented inPart I of this book fall into the literature that also includes such work as Dragu (2014)and Dragu & Polborn (2013) who show that norm-based behavior such as humanrights protections and legal limits on executive behavior can be self-enforcing. Like-wise, my contribution has broad normative implications. For instance, it speaks tothe effects of transparency policies on the long-term quality of representation in newdemocracies, an area that has sparked relatively little scholarly attention. With a fewexceptions, which include Yarhi-Milo (2013) and Felli & Hortala-Vallve (2015), the useof secrets and blackmail to affect policy has not received much scrutiny from politi-cal scientists. Yet, these concepts are eminently relevant to current events: in 2017, the


term kompromat made front page news as it was used to describe Russia’s attempts tointervene in the US elections. Generally, kompromat (“compromising materials”) refersto either embarrassing information or evidence of illegal activity which, if revealed,could damage a person’s career or open her up to prosecution. As Keith Darden pointsout, this kind of data was routinely collected by authoritarian secret services in theSoviet Era and used to control people through blackmail by threatening the “compro-mised” with the release of damaging information to the public—or worse, to prose-cutors (Darden 2001). Such embarrassing or damaging information, even if collectedby authoritarian security forces, may be put to use long after the authoritarian regimeitself has expired.

A crucial theoretical result of this chapter in particular is that while democraticrepresentation improves in direct proportion to the amount of lustration implemented,it does not improve in proportion to the intensity of purges. Purges, although they mayseem to placate a backward-looking public, can have dire consequences down the line.Especially if those who staffed the ancien regime’s administrative apparatus are ex-perts with preferences that are not excessively misaligned with those of leaders of thenew democratic regime, purging them may come at great cost to the new democracy.

For the last few decades, the question of whether transitional justice helps or hurtsdemocracy has been of central interest to scholars of normative transitional justice,many of whom are skeptical that transitional justice can help successful democratiza-tion without jeopardizing the rule of law. Authors such as Holmes (1994) and Osi-atynski (1994, 1992) even singled out lustration as undermining the principle of non-retroactivity: “Nullum criminen sine lege” (“no punishment without a crime”). Theseskeptics have argued that lustration cannot possibly offer a legal foundation for newdemocratic states because it violates principles of rule of law by discriminating againstcitizens who were acting according to the law of the times. Halmai et al. (1997) arguedthat “living well is the best form of revenge.” Others, in the same vein, have advocatedfor the so-called “Spanish Model of Transitional Justice” (Elster 1998, Rosenfeld 1997,Elster 2005). These authors used Spain’s reserved way of dealing with former author-itarian collaborators (sealing off the archives of Franco’s secret police) to build theircase that “doing nothing” is the best approach for new democracies to deal with pastauthoritarian regimes (Elster 2004b). Misuses of lustration, de-communization, andde-Ba’athification corroborate this endorsement of the Spanish model (Kritz 1995).For examples, one need look no further than Iraq’s policy aimed at purging new demo-cratic institutions of former Ba’athists. De-Ba’athification prevented 185 members ofSaddam Hussein’s party, mostly Sunnis, from running for the legislature in 2003. De-spite its promise to promote societal reconciliation, it ignited ethnic tensions (David2006). France’s policy of epuration, banning former Vichy collaborators from holdingoffice after WWII, had very similar effects (Elster 2004b).

Most of these theories castigating personnel forms of transitional justice, however,have focused on the immediate aftermath of the transition. In the first part of the book,I posited the theoretical possibility that the Spanish model of “doing nothing” may notproduce immediate negative consequences, but over time may strengthen the powerof authoritarian networks and particularly the networks involving secret legacies ofthe authoritarian regime. Damaging information collected by the former authoritariansecret police for the benefit of authoritarian elites may turn elected politicians intoclients of agents who threaten to reveal their “skeletons in the closet” (Nalepa 2010b).

Lending credence to these normative theories, note however, that the model oftransparency regimes presented in Chapter 2 does corroborate some of their findings.


Note that I find there that ooling equilibria occur more frequently when the cost tothe politician of having his skeletons in the closet exposed is greater. Empirically, thismight happen in the immediate aftermath of the transition from authoritarianism,when the salience of who was and was not a collaborator with the ancien regime isat its highest. This may imply that lustration implemented shortly after the transitionworks less well than when lustration is implemented later after the transition. Thus, ifthese scholars comparing lustration to “witch hunts” and “ritual sacrifices” are usingthe early experiences with transparency regimes as their inspiration 18, it is entirelypossible that they are picking up on the bluffing-rich pooling equilibria.

Forgiving and forgetting may sabotage the capacity of elected politicians to repre-sent voters. In the second part of this book, I subject this theory to empirical testing.

18 None of this early work is based on systematic data collection

Introduction to Part II

The main argument presented in this book thus far can be summarized in two points:First, revealing the truth about collaboration, and the associated bans on holding

public office imposed on secret collaborators, can have a different effect than pun-ishing former elites and perpetrators of human rights whose collaboration in the oldregime is known. In a world without lustration, unknown collaborators of the authori-tarian regime (or perpetrators who committed atrocities on its behalf in secret) can beblackmailed with “skeletons in the closet” Nalepa (2010b). So long as the public stillpays attention to what happened in the past, the revelation of such skeletons can termi-nate political careers. In return for their silence, individuals in possession of credibleevidence of skeletons in the closet can demand rents or policy concessions. The pos-sibility of such blackmail reduces the quality of democracy. In light of this dynamic,I expect transparency regimes to enhance the quality of democracy more consistentlythan purges—that is, personnel transitional justice procedures that sanction knownforms of collaboration.

Second, the mechanism behind the vetting of known collaborators—which I defineas purges—is different. Purges reveal no new information. In fact, purging state insti-tutions of elites who maintained the former authoritarian regime, when this purgingreaches a sufficient depth, can be conceptualized as the reverse of a delegation prob-lem. If a newly elected politician takes office and vets the administrative apparatus sothoroughly that he removes all bureaucrats with policy expertise, he has to implementpolicy himself in the inherently uncertain conditions of a post-transition environment.Without the expertise of thepeople who ran the agencies under the ancien regime, hecannot know how policy implementation will be affected by unknown states of theworld. Thus, purges, and especially thorough purges, whereby entire agencies of theancien regime are disbanded, are the equivalent of refusing to delegate to an informedagent. Forgoing purges, on the other hand, can be thought of as the equivalent of dele-gation to an agent who is equipped with expertise and so able to adjust policies to thestate of the world. In light of the uninformed post-transition bureaucracy that resultsfrom thorough purges, I expect increases in thorough purges to decrease the qualityof democracy more than increases in leadership purges, which remove only the chiefautocrats or perpetrators of human rights violations.

In this book, I am interested in the implications of these different transitional jus-tice mechanisms for the quality of democracy. I evaluate the quality of democracy


through three questions: (1) who goes into politics (selection into office), (2) how dopublic officials behave while in office, and (3) to whom does the state delegate.

I begin this empirical part of the book by presenting a research design imple-mented in the next four chapters of this book.

I first address the questions of who is selected into office and how they behaveonce in office.

Chapters 2 and 3 of the first part of the book presented formal models of trans-parency regimes and purges to explain how the mechanisms behind these two formsof transitional justice differ.

The theory presented in Chapter 2 could be interpreted as the“forward-looking”argument (in contrast to the“backward-looking” or retroactive argument) in supportof transparency regimes: by revealing information about former secret collaboratorsof the authoritarian regime, lustration prevents former authoritarian elites from ex-tracting policy concessions from elected politicians. Absent lustration, former agentsof the secret police could blackmail collaborators who have assumed political officeand threaten to reveal skeletons in their closets in the event of refusal to implementdesired politicians. Hence, implementing lustration laws should allow all politiciansto become responsive to their constituencies.

The theory also implies that when a country is capable of scrutinizing false infor-mation about skeletons in the closet (for example, by having a free media or a com-prehensive and effective right to information law), blackmail occurs only in instanceswhere evidence of collaboration actually exists. In other words, transitional justiceworks as intended only when damaging information can be objectively verified. Thelogic driving this finding is as follows: former collaborators-turned-politicians cannotknow with certainty if files against them survived the transition, so former agents canbluff politicians into thinking such evidence exists if this bluff cannot be easily chal-lenged. Hence, if such information cannot be properly scrutinized, former agents willbe able to extort (via bluffing) policy concessions from politicians even when they donot actually have skeletons to reveal.

In equilibria that allow for bluffing (in pooling and semi-pooling equilibria),blackmail is effective regardless of lustration. I contrast such equilibria with the sepa-rating equilibrium, where blackmail is only effective insofar as lustration proceduresleave evidence behind. However, even if lustration does not perform as normative the-orists would like it to—that is, in the pooling and semi-pooling equilibria—the formalmodel predicts that the bluffing former secret police officer will adjust his demanddown.

Demonstrating that lustrations have the predicted effects is the subject of Chapter5. Even given the aforementioned caveats, the severity of transparency regimes hasa positive and monotonic effect on the quality of representation. I thus expect truthcommissions and lustration to outperform purges and “doing nothing.” I will alsodemonstrate this to be the case in Chapter 5.

Another important parameter to control for in testing the model from Chapter2 is the distance between the blackmailer and the new democratic politician. This iscomplicated by the fact that in democratic polities, politicians can occupy many dif-ferent positions in the policy space. Because of this, in Chapter 5, therefore, the unitof analysis is a party within a country. This allows me to account for the fact thatdemocratic parties’ ideal points are different from one another and some are closer tothe blackmailer than others. This also calls for a multilevel (or hierarchical) modelinganalysis—that is, one that accounts for the fact that variation in the quality of parties


within countries ought to be broken down into two levels: the country level (as partiessubjected to the same electoral and party systems have elements of variation in com-mon) and the within-country level (as parties vary within the same country). Thiswill also call for a dependent variable capable of measuring the quality of democraticrepresentation at the level of political parties within a country. In chapter 4.1, I explainwhy party programmaticness—that is, their ability to represent voters programmati-cally, as opposed to clientelistically—is a suitable variable for my analysis.

Truth commissions have a similar effect to lustration but, with the exception of ageneral comparison of mechanisms at the most macro level, are analyzed in a separatechapter of the book.19 The reason for this is that although both constitute transparencyregimes, truth commissions have a naturally broader reach: they are capable of reveal-ing the skeletons of anyone in society, and not only politicians. As a result, I expect tobe able to detect the effects of truth commissions after shorter periods of time than lus-trations. This suggests that while an HLM design is suitable for analyzing of the effectsof lustrations, which take longer to disseminate their effects, a design that takes timeseriously, such as a difference-in-difference design, is more appropriate for studyingthe effects of truth commissions. Such a design allows us to “borrow the trajectoryof the untreated” (with truth commissions) countries to construct, for each countrythat experienced truth commission events, a counterfactual of what the quality of itsdemocracy would have been had if the truth commission events not taken place. Truthcommissions constitute the universe of cases in Chapter 8. HERE

Chapter 2 addressed who is selected into office and how they behave in office,but did not address when new democratic politicians delegate power to agents. Thisis the subject of Chapter 3, with known or open collaborators. Since purges are thereverse of a delegation problem, I showed that under most circumstances, the sever-ity of purges does not have a monotonic effect on the quality of representation. Infact, whether purges work at all depends on three factors. The first is how distant thepreferences of the former authoritarian elites are from the preferences of new demo-cratic politicians (preference divergence).20 The second critical factor is the magnitudeof the exogenous shock that the new democratic politician faces when implementingpolicy. To some extent, every post-transition environment is ridden with uncertain-ties, so we may be tempted to ignore this parameter and assume that ε is always large.However, I will distinguish between polities where these conditions are more acute be-cause the transition took the form of a revolution (such as Tunisia) and those wherethe transition was negotiated over many months or years (such as Poland or SouthAfrica). I argue that in the latter, the shock is smaller. Finally, the third determinant ofwhen politicians delegate is institutionalization of the former authoritarian regime’senforcement and bureaucratic apparatus.

The predictions from Chapter 3 on whether purges are feasible are summarizedin Tables ?? and ?? below:

Several caveats are in order here. First, these are clearly not normative predictions,but rather predictions of what a new democratic politician motivated by bringing pol-

19 Hence, Chapter 5 referred to above, on the programmatic representation of parties,uses only lustration severity among the transitional justice independent variables.

20 As a side note, this is a second key difference between transparency regimes andpurges because, as I showed in Chapter 2, the effectiveness of lustration does notdepend at all on the misalignment of the preferences of former authoritarian elitesand newly elected democratic politicians.


Institutionalizeduncertainty xA → 0 xA >> 0ε high ∼ Purge Maybe

Purgeε low ∼ Purge Probably

not Purge

∼ Institutionalizeduncertainty xA → 0 xA >> 0ε high either Definitely

Purgeε low Maybe

PurgePurge

Table 3.2: Model’s Predictions of whether a purge is expected following thetransition as a function of uncertainty (ε), preference divergence (xA), andprofessionalization (s)

icy as close to his ideal point as possible would do. Obviously, newly democratic politi-cians in South Africa were reacting to normative pressures from the public, such asthe urge to punish members of the Apartheid era officers. After all, the essence ofpurges is that they deal with known members and collaborators of the ancien regime.Hence, it is to be expected that purge events, for all of these normative reasons, willbe implemented in contexts diverging from these predictions.

Second, some of the parameters modifying our expectations on purges are harderto measure than others. Proximity of preferences between the outgoing regime’s ad-ministration and the new democratic politician is hard to approximate for at least tworeasons. First, these ideal points cannot be recovered from expert surveys or roll callvotes because they are nested within authoritarian regimes that have ended. Second,even in contemporary democratic regimes, political scientists lack routines for plac-ing bureaucrats on policy dimensions. Assessing uncertainty, or the magnitude of theshock, is similarly difficult.21 The length and abruptness of the democratic transitioncan offer some approximation, but such evaluations can at most be conducted for ahandful of cases.

It is much easier to use existing data to evaluate the level of professionalizationof the ancien regime, as I will elaborate on below. One can also place the archetypalcases discussed in this book into cells of the table above, as follows:

The distribution of the cases across the uncertainty dimension matches the lengthof their transitions. The level of institutionalization for this handful of cases is based onmy reading of secondary sources. For instance, according to Masri (2017) and Perkins(2014), the Tunisian military refused to carry out Ben Ali’s orders in defense of theregime. In addition, the military had been historically weak. Masri calls Tunisia an“Arab anomaly” in part due to the fact that labor unions in the country had tradi-tionally been strong, while the military had been chronically under-institutionalized.Hence, the Tunisian military can plausibly be assigned to the cell with aligned pref-erences (xA → 0) and very low professionalization. Yet, even within Tunisian en-forcement agencies, there are differences in institutionalization. The secret police re-sembled a praetorian guard and was in contrast to the military, very loyal to Ben Ali,

21 A final statistical consideration, as I explain below, is that this divergence betweenoutgoing autocrats and incoming democrats is not easily measured over time; con-sequently, controlling for these conditions can only be done within random effectsmodels. Recall that in the fixed effects framework, country effects include any po-tential constant characteristics of different units.


Institutionalizeduncertainty xA → 0 xA >> 0ε high Tunisia

(mili-tary)

Tunisia(secretpolice)

ε low Spain Bolivia

∼ Institutionalizeduncertainty xA → 0 xA >> 0ε high S. Korea

(othermilitary)

S. Korea(Hana-hoe)

ε low Poland S. Africa

Table 3.3: Predictions of whether a purge is expected following the transitionas a function of uncertainty (ε), preference divergence (xA), and profession-alization, (s)

though equally lacking in expertise when it comes to effective quelling of the oppo-sition. At the same time, the transition in Tunisia was so rapid that it seems justifiedto evaluate the level of uncertainty to be high. As I described in earlier chapters anddemonstrated in Figure 1.4, Tunisia was indeed a country with a high intensity ofpurge activity even though it has only been a democracy for six years in our dataset.

Another country where disaggregating in the analysis of ancien regime agenciesseems justified is South Korea. Chung Hee Park, during his 18 year rule, created twoparallel career tracks within the military, with separate but credible rules of advance-ment and promotion. The first were the armed forces, whose main goal was deterringNorth Korea. The second arm included the security forces who, in the words of Joo-Hong Kim, “traded honor and prestige for power” (Baik et al. 2011, p.171). The latterbranch’s promotion path was slower, but they were argued to have more influenceon domestic politics and acted as Park’s personal militia in addition to providing himwith much needed security. Park would ultimately authorize all promotions himself.

Following the transition, Korea’s transitional justice focused on truth commis-sions, as illustrated in Figure 1.4. However, March 8, 1993 saw the beginning of athorough purge of the so-called, “Hanahoe group” (the “Group of one”), an unofficialprivate group of military officers whose members were graduates of the eleventh classof the Korean Military Academy of 1955. The Hanahoe had maintained great influencein the security branch of Park’s military, as Park himself was a graduate of that class.Had the organization been left intact, it would have remained a bulwark of authoritar-ian influence. Key members of Hanahoe were forced into retirement, but in additionto this, in January of 1994, the National Assembly revised the laws on major intelli-gence agencies, forcing the Agency for National Security Planning (the equivalent ofSouth Korea’s CIA) and the Military Security Command to limit political influence.This was accomplished by placing both the military and the security apparatus undercongressional oversight (Sang-Hun 2015).

South Africa’s state administration was professionalized but was also restrictedunder the apartheid to Afrikaners. Despite this misalignment of preferences, uncer-tainty is estimated to have been quite low because more than five years prior to thesigning of the transition pact, the leadership of apartheid regime had come to the real-ization that “brute force could keep the Africaaner in power until the next century, butit was not worth the cost in economic development”(Waldmeir 1997, p.53). With theCold War’s end, the international community’s attention turned to sanctioning statesviolating human rights regardless of whether they claimed to be protecting the worldfrom communism, as the apartheid frequently did (Price 1991). An additional imped-


iment to implementing purges was that school boycotts had become one of the mainforms of protest against the apartheid regime. Consequently, the ANC did not havea well-trained bureaucratic elite ready to replace that of the apartheid regime. Therewere no purges at all in the aftermath of the transition in South Africa. Instead, thecountry invested all of its resources in an extensive transparency regime. The trend ofthe process of truth commission events in South Africa from Figure 1.4 in Chapter 1reflects predominantly events associated with the creation, hearings, and final reportof the Truth and Reconciliation Commission.

Bolivia had an abrupt transition, as the dictator wielding the longest period ofdictatorship, Barientos, died while in office. However the transitions between succes-sive regimes showed remarkable continuity Malloy & Gamarra (1988). Hence the un-certainty at the time of transition is judged to be low. The military, while similarlyUS funded to the South Korean military lagged behind it in military success, its onlyclaim to fame being the assassination of Che Guevara, the Cuban secret police chiefand revolutionary. Moreover, what kept Barientos in power was a pact between thepeasant movement and military against the urban trade unions, suggesting the mili-tary would not be able to survive on its own. After considering all these factors, theBolivian case is classified as one of low institutionalization. As figure 1.4 indicatesthere were practically no purges—or for that matter any kind of transitional justice—in Bolivia following the transition to democracy.

Finally, it is worth commenting on the classification of Spain. Spain presents a com-plicated case because, as described in the introduction, the pact of forgetting blockedthe possibility for transitional justice. It is also complicated by the nature of the Francodictatorship itself, which on the one hand offered some members of the state admin-istration vast autonomy.22 At the same time, however, Payne writes that almost up tohis death, Franco personally met with members of his cabinet and prided himself on“not trusting anyone.” This attitude of trusting but controlling23 is clearly illustrated ina comment of Salgado Arawjo’s (who was also a member of Franco’s inner circle):“Surely, His Excellency does not want to have ministers with independent personal-ities who could create difficulties” (Payne 2011). It would similarly be tempting torefer to Spain’s military as professionalized by virtue of the fact that Franco himselfwas a military general. Nevertheless, beginning with his victory over Republicans inthe Civil War and throughout World War II (the percentage of the budget devotedto the military decreased between 1940 and 1945 from 41 to 34 %) and its aftermath,Franco slashed the military budget. According to Payne, this led to the marginaliza-tion of at first the junta and later the military in general. According to Payne (2011),“during the first phase of the regime down to 1945, military personnel would hold45.9 percent of the ministerial appointments and 36.8 percent of all government po-sitions (236).” By 1974, the percentage of the military in the top state administrationwas down to 15.8 %.

In light of these considerations, I classify Spain under Franco as an under-institutionalizedauthoritarian regime. This classification is in line with criteria put forward by recentresearch on authoritarian institutionalization from Anne Meng (2020). Meng uses as

22 According to (Payne 2011, 398), “cabinet ministers and principal subordinates werealmost always given great latitude in running their departments” and backs thiswith a quote from one of his former ministers, Lequerica, “being a cabinet minister isthe only serious thing you can be in Spain. To be a minister of Franco is to be a little king.”

23 A maxim attributed to Lenin.


one criterion of institutionalization whether the regime managed to outlive its found-ing leader. Despite Franco’s best efforts to nominate a successor who could lead Spainunder Franco’s constitution beyond his death, the Generalissimo failed at this task, asKing Juan Carlos initiated transition negotiations promptly following Franco’s death.

As mentioned above, for purges to be optimal, the institutions of the former au-thoritarian state must have been developed enough to allow elites to gain expertisevaluable in the new democracy. This calls for a way to measure institutionalizationof the preceding authoritarian regime. Political scientists have, for a long time now,been developing measures of state capacity that could be used for such purposes. Theoverwhelming problem with these existing measures, however, is that they smuggle innormative evaluations of regime type. Consider the dilemma of evaluating which au-thoritarian regimes are better at extracting tax revenue, which is a question squarely inthe domain of state capacity. Clearly, some authoritarian regimes are better than oth-ers at tax extraction even with lower coercive apparatuses. Hence, a desirable mea-sure should take into account both aspects without adding anegative score for theregime for being authoritarian. This is very important for my purposes also becauseprofessionalization of state agencies can but need not, go hand in hand with a state’srepressive apparatus. Authoritarian regimes have to distribute their coercive capac-ity across tax extraction and dealing with anti-authoritarian resistance. The strongerthe opposition’s challenges are, the more the state must transfer away from extractionand towards repression. Moreover, neither of these aspects touch upon an authoritar-ian state’s ability to administer scarce resources. For this reason, I take advantage ofa new measure of state capacity proposed by Hansen & Sigman (2020), who decom-pose state capacity into three dimensions: coercive capacity, extractive capacity, andadministrative capacity. Each dimension is measured separately with a set of existingindicators and next, Bayesian latent variable analysis isapplied to all to recover howmany actual dimensions to state capacity there are.

The availability of this measure and its precision are vital for the purpose of testingthe Chapter 3 model because as Table 3.2 indicates, whether a new democratic polityshould purge or not is contingent on the professionalization of its administrative appa-ratus. We must first, therefore, account for former authoritarian regimes selecting intopurges before we can test the model’s predictions regarding the quality of democracybrought about by purges.

Note that Figure 3.5 from Chapter 3 not only predicts when the new democraticpolitician should conduct a purge, but also makes predictions for how far the policyoutcome is from the new democratic politician’s ideal point. In light of this, the dis-tance of this policy outcome corresponds to the quality of democratic representationand can be operationalized with a measure of the quality of democracy.

This calls for a two-stage model (equivalent to an instrumental variable approach),with the first model predicting whether a post-authoritarian purge occurs and usingthe measure of state capacity as a predictor, and the second model using the estimatedvalue from the first model as an explanatory variable to predict the quality of demo-cratic representation. This is the task of Chapter 6. Hence, Chapter 6 will simultane-ously address dilemmas (2) and (3) of this book: ‘how do politicians behave once inoffice?’ and ‘when do politicians delegate to bureaucrats?’

Concerning the former question, to address the question of how politicians behaveonce in office empirically, I have to choose a way of measuring political output in waythat captures intuitions of high democratic quality. I will propose a few measures inthe next chapter, while the subsequent chapters will be devoted to measuring transi-


tional justice. Another complexity of the research design presented here is associatedwith the very nature of the inquiry into the relationship between mechanisms of tran-sitional justice and the quality of democracy. Answering whether the enactment oftransitional justice improves the quality of democracy is made hard by the fact thatthe same factors that cause states to embark on transitional justice may lead them tohave a more robust democracy down the line.

The next chapter is devoted to developing suitable dependent variables for testingmy theory. It also includes an extensive section on how to operationalize and measuretransitional justice. The last three substantive chapters of this book will develop andtest the theoretical predictions about the differential effects of the transitional justicemechanisms to which this book is devoted.

4

Measurement

This book is devoted to asking the broad question of whether transitional justice en-hances the quality of democracy. I have broken down answering it into three comple-mentary dilemmas: (1) who is selected to serve in office? (2) how do these politiciansbehave in office? and (3) how do they delegate political power? In order to addressthis system of questions, it is necessary to measure both the quality of democracy andtransitional justice. The first part of the chapter is focused on measuring the dependentvariable: quality of representation. I develop three distinct measures, called for by thefact that, as the three parts of my question are examined, the unit of analysis changesfrom a party within a country to a country at a point in time to a country over theentire course of the post-transition period. The second part of this chapter is devotedto measuring transitional justice. There, I will lay out a data collection strategy thatboth takes time seriously and disaggregates transitional justice to an unprecedentedlevel.

4.1 Measuring the quality of representation

Over the last decade, scholars have created datasets that span transitional justice im-plementation across the globe and over time. The dependent variables these scholarshave looked at range from trust in governmental institutions (Horne & Levi 2004)to peace (Binningsbø et al. 2012) to democratic stability (Olsen et al. 2010). Schol-ars analyzing these data on transitional justice have found little evidence that thesemechanisms make a difference in preventing former autocrats from reasserting theirpolitical dominance (Olsen et al. 2010, Binningsbø et al. 2012, Van der Merwe et al.2009, Thoms et al. 2010).

Programmaticness

None of the variables used thus far in the literature capture the quality of democracyas expressed by the three questions above. For instance, in order to operationalize theparameter that characterizes the equilibrium prediction from Chapters 2 and 3, weneed a measure of the degree to which politicians are able to implement the policiesthey prefer for purely ideological reasons; that is, the degree to which participants, free

80 4 Measurement

from patronage, pressures of blackmail, and unburdened by a lack of expertise, canimplement policies voters desire. Indeed, this conceptualization is very close to whatis considered in the comparative politics literature as programmatic representation(Carroll & Nalepa 2014, 2020), according to which parties are programmatic when

1. Parties run on policy platforms2. Voters are aware of these platforms and elect parties on the basis of these plat-

forms3. Parties aspire to implement these platforms upon election.

There is a vast literature on how clientelism interferes with politicians and votersengaging in these three tasks (Stokes et al. 2013, Auerbach & Thachil 2018, Kitschelt& Wilkinson 2007, Kitschelt 1999). However, legacies of authoritarian rule create ad-ditional mechanisms that interfere with programmatic representation. Notably, theblackmail of politicians by members of the previous authoritarian regime, if success-ful in extracting concessions from the politicians, constrains their ability to faithfullyrepresent voters. To the extent that programmatic representation involves deliveringpolicies that represent broad wishes of the electorate, the theory from Chapter 2 makespredictions about how transparency regimes enable such representation by eliminat-ing the possibility of using compromising information to extract policy concessions.

Thus, as a first measure of the quality of representation, I propose a variablethat builds on the measure of programmaticness constructed by Herbert Kitscheltand his co-authors from the Democracy and Accountability Linkages Project (DALP)(?Kitschelt & Freeze 2010).

In DALP, programmaticness is conceptualized as a combination of three partycharacteristics: (1) the salience of party platforms; (2) parties’ cohesion around theseplatforms; and (3) the polarization of parties in the party system. Kitschelt et. al con-structed an index based on responses to questions tapping into each of these com-ponents. Given the conceptual proximity between their polarization and my model’sdistance between the ideal points of the former authoritarian elites and the new demo-cratic elites, I created a new measure that leaves out questions pertaining to polariza-tion.1 This is essential, as distance ought to appear in the model on the right hand sideof the equation and using it as a component of the dependent variable would producebiased parameter estimates. The result is a score between 0 and 1 that comprises justthe cohesion and salience dimensions of the original variable. The graph below showsthe histogram of this measure using the 315 parties for which this measure was cre-ated. This dependent variable is used in the hierarchical modeling analysis of the nextchapter. The added value of DALP is that it offers a party-based measure; this yieldsmultiple measures per country—as many as there are parties in each country—andalso allows me to better approximate the model’s predictions. Hence, this measurewill be used in a hierarchical modeling setup. In contrast, the next variables I proposeoffer multiple measures per country because they are measured over time.

For the next set of dependent variables I propose, I use V-Dem, the Varieties ofDemocracy Project. V-Dem is a dataset spanning 221 countries and 116 years for mostpolities2 The dataset was created by country experts (approximately 5 independent

1 This measure was prepared with the valuable assistance of Milena Ang.2 Some polities did not survive as independent states for the entire covered period.

For a detailed list and explanation of country unit selection, see Coppedge et al.(2017).

4.1 Measuring the quality of representation 81

Fig. 4.1: Histogram of measure of party cohesion and salience, based on DALP,by taking polarization questions out of the cosalpo3 measure

0.5

11.5

22.5

Density

0 .2 .4 .6 .8 1cosal_3

experts per country) who subjectively coded a host of regime characteristics that arenot directly observable over time. V-Dem is exceptional among expert surveys in thatit corrects for how differences of opinion or mistakes cause experts to diverge in theirevaluations. Traditionally, datasets report expert-coded data with means and standarddeviations, ignoring the fact that expert reliability and the way in which experts ap-ply an ordinal scale to ratings may vary systematically. V-Dem, on the other hand usesitem response theory to model and adjust for differences in how experts apply scales(Pemstein et al. 2015). To allow for scaling the independent coding by country experts,V-Dem scholars also encouraged experts to “bridge code” a second or third country.While experts have less expertise in evaluating these second and third countries thanthey do at evaluating their first countries, this effort allowed V-Dem methodologiststo compare the use of the ordinal scales across coders and correct for systematic dif-ferences.

V-Dem researchers asked approximately three thousand country experts hun-dreds of questions to arrive at five indices of democracy: electoral, liberal, participa-tory, deliberative, and egalitarian. Each of the indices uses some variable to whichtransitional justice is potentially endogenous. Thus, in lieu of using any of the 5 V-Dem indices, I use specific expert evaluations that tap into the concept of quality ofdemocratic representation. Among V-Dem variables that stand out as particularly use-ful for this purpose are Power Distributed by Socioeconomic Status (v2pepwrses) andthe Political Corruption index (v2xcorr), which I discuss in turn below.

Power Distributed by Socio-economic Status

It is challenging to pick a dependent variable for a research design aiming to uncoverthe relationship between transitional justice and the quality of democracy because theimplementation of transitional justice is endogenous to phenomena that are used asbuilding blocks for so many democratic indicators; these range from rule of law tofreedom from discrimination to freedom to run for office. For this reason, such a vari-able should rather focus on the authoritarian characteristics of leaders than on the

82 4 Measurement

democratic characteristics of institutions. I therefore turn to elite reproduction mea-sures.

Research on the reproduction of authoritarian elites is related closely to the lit-erature on authoritarian survival, spearheaded by Adam Przeworski and JenniferGandhi. In their seminal 2009 article, they show that political institutions that “so-licit the cooperation of the opposition” by giving it a voice can aid autocratic survival.In their empirical analysis, they single out one particular strategy—giving oppositionparties representation in the legislature—as most conducive to prolonging the life ofdictatorships (Gandhi & Przeworski 2007).3

Milan Svolik (2012) takes an inside-the-regime approach, attributing authoritar-ian survival to the way in which dictators share power with their ruling coalitions.Since the typical way of deposing an autocrat is through a coup d’etat, sharing powerwith an elite made up of potentially threatening competitors not only keeps a dicta-tor in check, but also simultaneously ensures the survival of the regime beyond thelife of the leader. Although they focus on different mechanisms, both Svolik’s andGandhi’s work point to the same by-product of securing authoritarian survival: insti-tutionalization of the regime. The expectation implicit in Svolik and explicit in Gandhiis that when equipped with power-sharing institutions, authoritarian regimes survivelonger.

In some very recent work, Anne Meng adds more nuance to this argument. Shepoints out that a legislature and elections alone are no more an indication of powersharing than formal constitutional rules are of actual constraints on executive power.True authoritarian institutionalization manifests itself in staffing key regime positionswith competitors to power and not figureheads. Only dictators that place de facto andcredible institutional limits on their survival succeed in allowing their regimes to out-live them. Yet, not all dictators have the same incentives to institutionalize. Becausesome autocrats are strong enough to stave off challenges to their power without insti-tutionalization, they may stay in power just as long as weaker autocrats who do placesuch limitations upon themselves. Consequently, since the decision to institutionalizeis endogenous, when measured properly, the relationship between institutionalizationand survival rates should disappear (Meng 2020).4

While fascinating in its own right, the primary interest of this scholarship lies inthe survival of autocrats under the existing authoritarian regime. My objective, how-ever, is to examine autocrats’ endurance prospects beyond their regime of origin. Thisquestion has been addressed, for instance, by Alexandre Debs (2016). In search ofan answer to the question “what makes certain dictatorships more likely than oth-ers to democratize?,” he argues that since military dictators are most threatening totheir successors unless they are chosen by democratic elections, military dictatorshipsshould be the most likely to transition to democracy swiftly. Also, because former mil-itary dictators are less likely to face punishment at the hands of democratic successorsthan at the hands of authoritarian successors, they are less reluctant to step down.

3 For more examples of institutional reforms that extend the persistence of autocratssee Blaydes (2010, 2018), Weeks (2014), Svolik (2012).

4 For a different conclusion about the decision to institutionalize under authoritari-anism, see Myerson (2008), who shows that only strong autocrats—that is, thosewho can credibly commit to making transparent the rewarding ofefforts made ontheir behalf by agents who work for them, will institutionalize. For a more tractableversion of Myerson’s model, see Gehlbach (2013).


An alternative interpretation to the question of authoritarian survival has beenprovided more recently by Albertus and Menaldo Albertus & Menaldo (2018), whoargue that dictators in general only concede to democratize on conditions that arebeneficial to them. However, Albertus and Menaldo are predominantly interested inthe terms under which autocrats concede to a democratic transition. They researchneither the specific career trajectory of post-authoritarian elites nor the nature of thetransitional justice process that may have been used against them. Indeed, all of theirindependent variables predate the transition itself.

Anna Grzymala-Busse (2002) has examined how successor authoritarian partiesas fare organizations under new democratic conditions. Although her research is lim-ited to the post-communist world, she finds that the degree to which a party’s organi-zation was technocratic—as opposed to ideological—prior to the transition is a verystrong predictor of its success. Grzymala-Busse’s newer work extends this argumentto a global universe of cases (Grzymala-Busse 2017), but finds that although the strat-egy of reinvention leads successor authoritarian parties to do better in the short runthan non-reformed successor parties, eventually their advantage dwindles and theydie out.

Finally, my earlier book, also restricted to post-communist Europe, developed theidea that former authoritarian elites may be harder to punish in the aftermath of tran-sitions to democracy if while negotiating the terms of transition, they succeed at con-vincing the opposition that transitional justice may reveal skeletons in the opposition’scloset by exposing collaborators who worked for the communist secret police and thusstraining the fragile reputation of the new democratic elites (Nalepa 2010b).

This literature has convincingly established that authoritarian elites have a bettershot at surviving the transition when the opposition is inclined to tolerate them fortheir skills. This toleration may originate in the autocrats’ and opposition having sat inthe same legislative assembly, or came about because the opposition committed itselfto respecting the former autocrats when negotiating the terms of the transition (pos-sibly after skeletons in the opposition’s closet were revealed). The power of formerauthoritarian elites extends beyond the life span of an authoritarian regime. Autocratsmay be well-positioned to capture state resources at the time of democratic transition,which they can then use in a clientelistic fashion to stay in power (Brun & Diamond2014, Haggard & Kaufman 2016, Albertus & Menaldo 2014). The outgoing autocrats’access to resources can be cut off if they or their successors are voted out of officefollowing the transition to democracy. Various cases from around the world demon-strate, however, that this removal may only be temporary (Kitschelt 1999). Grzymala-Busse (2002), for instance, attributes the revival of successor authoritarian parties tothe organizational advantage authoritarian parties hold over parties that are new tothe party system. This organizational advantage allows them to make better use ofstate resources when they eventually do find themselves in government.

Effective transitional justice institutions are often portrayed as the last resort tocurb autocrats’ unfair advantage. Indeed, scholars of transitional justice have arguedthat its mechanisms should undercut the privileged position of members of parties ofthe former autocrats, their collaborators, or their enforcement apparatuses (Stan et al.2009, David 2011, Vinjamuri & Snyder 2004, Escriba-Folch & Wright 2015).

In light of this discussion, personnel transitional justice may plausibly be inter-preted as a mechanism preventing former authoritarian elites from holding on tosuch economic resources. Therefore, a variable measuring the association betweeneconomic wealth and political power is an ideal candidate for a dependent variable

84 4 Measurement

operationalizing the effects of transitional justice on who is selected to be in political office.Additionally, given the temporal nature of my data, an ideally suited dependent vari-able also measures this association over time. Fortunately, the Varieties of DemocracyExpert Survey (V-Dem) contains such a measure.

Called “Political Power Distributed by Socio-economic status” (PdSES), the vari-able I ellect to use is based on the following question posed to V-Dem experts: “ispolitical power distributed according to socio-economic position?”(Coppedge et.al.2017b).5 In his clarification note, John Gerring elaborates that the measure was de-signed to gauge the extent to which inequalities translate into political power (Coppedgeet.al. 2017b).6 If the goal of personnel transitional justice is to undermine the privi-leged position of authoritarian elites, this score should increase with the severity ofthe transitional justice mechanism in question.

Political Corruption index

To measure how politicians behave in office, I use the V-Dem index of political corruption.This index is created out of six different questions regarding corruption in three differ-ent branches of government—the legislature, judiciary, and executive.7 Its advantagesare that within the executive, it treats bribery and embezzlement separately and thatit distinguishes between corruption taking place in high levels of government (“grandcorruption”) and lower in the public sector (“petty” corruption). The index takes onvalues between 0 and 1.8

While the index question does not get directly at the extent to which politiciansare influenced by blackmail with secret police files, it does measure how much theysuccumb to pressures that impede their ability to deliver on voters’ wishes once in

5 Answers to the question were distributed along a 5-point scale. The possible an-swers included (0)“Wealthy people enjoy a virtual monopoly on political power.Average and poorer people have almost no influence”; (1)“Wealthy people enjoy adominant hold on political power. People of average income have little say. Poorerpeople have essentially no influence”; (2)“Wealthy people have a very strong holdon political power. People of average or poorer income have some degree of influ-ence but only on issues that matter less for wealthy people”; (3)“Wealthy peoplehave more political power than others. But people of average income have almost asmuch influence and poor people also have a significant degree of political power”;and (4)“Wealthy people have no more political power than those whose economicstatus is average or poor. Political power is more or less equally distributed acrosseconomic groups” (Coppedge et.al. 2017b).

6 PdSES is also a particularly reasonable measure of quality of democracy for mypurposes because while it measures an important aspect of democracy, it is unlikelyto be correlated with rule of law, which could also affect the implementation oftransitional justice.

7 For example, the question about executive corruption is“How routinely do mem-bers of the executive, or their agents grant favors in exchange for bribes, kickbacks,or other material inducements, and how often do they steal, embezzle, or misap-propriate public funds or other state resources for personal or family use?”

8 Specifically, it is made up of questions coded as variables v2xpubcorr v2xexecorr,v2lgcrrpt, v2jucorrdc, v2excrptps, and v2exthftps Coppedge et al. (2018).


1990

2000

2010

0.00 0.25 0.50 0.75 1.00

Corruption Index in Argentina

Year

Corruption Index in Argentina since Transition

Fig. 4.2: Political Corruption IndexSince Argentina’s Transition

1990

2000

2010

0.00 0.25 0.50 0.75 1.00

Public Corruption in Argentina

Year

Public Corruption in Argentina since Transition

Fig. 4.3: Component: Public Theftand Corruption

office. Figure shows the trends since Argentina’s democratic transition of the politicalcorruption index (lower left) and the public theft measure (upper right), respectively.

1960 1970 1980 1990 2000 2010

0.0

0.2

0.4

0.6

0.8

1.0

Political corruption

Year

Ind

ex

Argentina

1960 1970 1980 1990 2000 2010

−3

−2

−1

01

23

Power distributed by social status

Year

Argentina

Fig. 4.4: Trend lines for dependent variable measures—Political Corruptionand Political Power Distributed by Socio-economic Status

Figure 4.4 presents, for all countries in the Global Transitional Justice Dataset,trend lines for both of my dependent variable measures—Political Corruption and

86 4 Measurement

Political Power Distributed by Socio-economic Status. The Argentinian case is in boldto allow the reader to focus on one case that offers especially rich variation. The fig-ure shows considerable variation both across and within countries, which is a keyadvantage of the V-Dem dataset over other datasets offering indicators for quality ofdemocracy, such as Polity IV.

Note that the third question—addressed in this book—when and how do politiciansdelegate power has been interpreted as when and how to purge. In light of this, it is thesubject of the subsequent section, which deals with measuring transitional justice, theset of independent variables in this book.

4.2 Measuring Transitional Justice

The previous section was devoted to operationalizing the dependent variables thatemerged from the models in Chapters 2 and 3. In order to conduct empirical tests ofthese models, an important prerequisite task is the operationalization of transitionaljustice. The previous chapter explained how extant quantitative studies of the effects oftransitional justice have not found a strong significant effect of personnel transitionaljustice (that is, purges and lustrations) on the quality of representation. This non-finding is puzzling as one would expect personnel transitional justice to particularlystand out as critically important for the objective of limiting the influence of formerthese elites because it deals with authoritarian elites most directly.

I believe that the ambiguities in these findings stem from data collection strategiesthat (1) are not sensitive to the temporal dimension of transitional justice and that (2)do not account for whether the type of collaboration with the former authoritarianregime was secret or open.

This section introduces my Global Transitional Justice Dataset. This dataset is thefirst to disaggregate TJ events in three novel ways. First, it breaks down transitionaljustice events into a time series that spans the years between the date of a country’s de-mocratization and 2016. Second, it disaggregates transitional justice mechanisms intothose, such as purges, that deal with open forms of collaboration and human rightsviolations, and those, such as lustration and truth commissions, that deal with secretforms of collaboration and human rights violations. It complements existing datasetson events related to transitional justice by focusing not only on post-conflict societies(as the PCJ does) and not only on post-authoritarian societies (as the TJDB does), buton both. Third, the transitional justice events themselves, the building blocks of mymeasures, are divided into positive and negative events, which for any given countrycan be visualized as a time series.

I demonstrate the advantages of my disaggregation strategy by using these data toconstruct the main measures of transitional justice intensity used in this book: sever-ity measures. The granular structure of the data allows researchers to construct ad-ditional measures depending on their theoretical questions of interest. In Chapter 5,I use the severity measures to predict “political power distributed by socioeconomicstatus,” the V-Dem variable (Coppedge et. al 2017a) described above. In subsequentchapters, I use variations of severity, modifying it to better fit the statistical models Iemploy. Following, Bates et al. (2020), I explain for illustrative purposes how to con-struct additional measures of urgency, and volatility. I will also compare and contrastmy measure of transitional justice severity with the ones proposed by the authors ofPCJ (Binningsbø et al. 2012) and TJDB (Olsen et al. 2010).

4.2 Measuring Transitional Justice 87

4.2.1 Why do we need a new Transitional Justice dataset?

Data on personnel TJ mechanisms are notoriously hard to collect, in no small part be-cause policy makers in different parts of the world refer to these mechanisms usingdifferent general terms ranging from “vetting” and “purging” to “house-cleaning,” orspecific terms such as “denazification,” “decommunization,” or “de-Baathification.”This variation most likely explains why many researchers and most databases poolall personnel TJ events under one heading of “purges” or “lustration.” Researcherssuch as Jon Elster (2004b), who have contributed theoretical accounts of TJ, and schol-ars who have developed datasets of personnel TJ treat lustration and administrativepurges (such as decommunization, de-Baathification, and denazification) as the sametype of mechanism for dealing with the past.

Authors of the widely recognized Post-Conflict Justice Database (Binningsbø et al.2012) refer to all personnel forms of TJ with the term “purges,” which describes “theacts of removing politicians, members of the armed forces, judiciary ... for their (al-leged) collaboration with or participation in a conflict and limiting their influence ac-cordingly” (Binningsbø et al. 2012, p.736). In part because their data collection effort islimited to societies recovering from conflict and only covers the first five post-conflictyears, these authors only managed to locate 15 post-conflict episodes that are followedby purges thus defined.

Olsen et al. (2010) similarly pool personnel TJ mechanisms into one category, call-ing it “lustration.” They define lustration as “the mechanism that occurs when thestate enacts official policies denying employment in public positions to individualsbecause of their former political acts or identity” (Olsen et al. 2010, p.38). Using thisdefinition, these authors were only able to record 54 instances of lustration, mostly inEastern Europe.9 Table 4.1 will demonstrate that I find many more lustration eventsthan the Olsen et.al. team was able to. The vast discrepancy is attributable in part tothe fact that these authors code the presence or absence of lustration with a dummyvariable while I differentiate across time between distinct lustration events, both posi-tive and negative. Second, the Olsen et. al. team’s collection was biased to only look forlustrations in Eastern Europe. This caused them to miss numerous lustration events(whether one uses Olsen et. al.’s definition or mine) outside of East Central Europe.

In a recent book, Cynthia Horne (2017a) underscores a failure to distinguish be-tween lustration and purges. She defines the former as “a form of vetting—the setof parliamentary laws that restrict members and collaborators of former repressiveregimes from holding a range of public offices, state management positions, or otherjobs with strong public influence (such as in the media or academia) after the collapseof an authoritarian regime”(Horne 2017a). “Purges,” meanwhile, extend collectivelyto members of organizations linked to the ancien regime. Thus, whereas “lustration”denotes a procedure that considers each case individually, purges presume collectiveresponsibility. Horne notes further that lustration is part of a more general categoryof “vetting,” which she defines as any ban on holding office; vetting does not have tobe limited to members of or collaborators with the ancien regime. Aligning with thecriticism laid out by Horne, I believe that the negligible effects of personnel TJ mech-

9 Indeed, most contributions to the lustration literature suggest that this kind of TJpolicy is limited to post-communist Europe (De Greiff & Mayer-Rieckh 2007, Ellis1996, Closa Montero 2010, Letki 2002, Stan 2013, Stan & Nedelsky 2015).

88 4 Measurement

anisms on the quality of democracy identified by Olsen et al. (2010) and Binningsbøet al. (2012) may have at their roots the failure to separate purges from lustration.10

Roman David (2011) takes a similar disaggregating approach, but limits the scopeof lustration to Eastern Europe. He justifies this decision with the fact that prior to1990, this term was not used to describe transitional justice procedures (David 2011,p.67).11 While it is true that the term “lustration” was rarely used outside of EasternEurope to describe the disqualification for public employment of secret and clandes-tine collaborators with the former regime, my data reveal that such procedures wereimplemented both before 1990 and beyond Eastern Europe. Consider as an examplethe Portuguese decree from March 22, 1975, which stipulates that any member of thearmed forces who was found to be “incompetent” or not prepared to make a decla-ration of loyalty to the MFA could be placed on reserve. Under the same decree, allmilitary members involved in the coup attempt of March 11 of that year were expelledfrom the forces, and their property was confiscated (Pinto 2001).

The failure to distinguish purges from lustration is one problem with existing data.Another is the difficulty scholars face when trying to pinpoint when exactly a mecha-nism of a certain type was implemented. This has led many researchers to aggregateall TJ activity of a certain type into a single variable that signifies the presence or ab-sence of that transitional justice mechanism in a given country. Noteworthy in theirattempt to surmount these problems are efforts by the Transitional Justice Database(TJDB) team (Olsen, Paine, & Reiter) and the Transitional Justice Research Collabora-tive.12 Both record the year in which a transitional justice mechanism of a given typewas implemented, allowing a country to have multiple events. While these datasetsoffer a considerable improvement over previous TJ datasets, they still condense muchof the process and assume away the possibility of reversals (Bakiner 2016). Further-more, particularly in the case of personnel TJ, it is difficult to exactly pinpoint a decisivemoment when these mechanisms are implemented. Instead, legislation is frequentlyfirst proposed, then amended, and eventually passed. Following this, it is sometimesstruck down by a constitutional court or presidential veto only to be reintroduced onthe legislative floor again. Moreover, with the passage of time, the scope of transi-tional justice mechanisms can be expanded or curtailed, or the legislation can be com-pletely revoked. Data projects trying to identify a specific year in which a procedurewas finally implemented reduce transitional justice to a “one-shot event.” Ignoringthe temporal nature of these processes assumes away their nonlinearity. My methodof documenting the entire process of personnel TJ as it unfolds over time requiresfewer judgment calls and is able to account for regressive changes in personnel TJ im-

10 Olsen et al. (2010) also include truth commissions in regressions as explanatoryvariables, but even these are marginally significant for only one type of outcomevariable: variously constructed terror scales.

11 David defines lustration as a “special public employment law that stipulates theconditions for the access of persons who worked for or collaborated with the polit-ical or repressive apparatus of socialist regimes to certain public positions in newdemocracies.”

12 The TJDB includes data on five transitional justice mechanisms includingamnesties, trials, truth commissions, lustrations, and reparations. The TransitionalJustice Research Collaborative covers amnesties, trials (including domestic, foreign,international, and civil), vetting, truth commissions, reparations, and customaryjustice.


plementation. Furthermore, a byproduct of the panel structure of my data is that itallows scholars to employ research designs that are not available to them with justcross-sectional structures. At the same time, all research designs suitable for cross-sectional formats can be used as well: one can do with disaggregated data anythingthat is possible with aggregated data, but not the other way around.

In sum, existing transitional justice datasets either pool complex events, leadingto the collapsing of transitional justice into a simple dummy variable, or fail to disag-gregate personnel transitional justice into mechanisms that deal with secret and opencollaborators.

Recall, however, from the previous chapters of this book, that although lustrationand purges are forms of personnel transitional justice in that they aim to eliminatefrom the state apparatus members and collaborators of the previous authoritarianregime, lustrations actually have more in common with truth commissions becausethey uncover information that was kept secret before the transition. Further, the effectof lustration severity on democratic representation is very different from the effect ofpurge severity on democratic representation. While the quality of political represen-tation increases monotonically with lustration severity, this is not necessarily the casewith purges. Moreover, purges are not feasible in all new democratic polities to thesame extent.

Recall the key difference between lustration—dealing with unknown collaborators—and purges, that is, dealing with known members and collaborators of the former an-cien regime. Politicians who in the past were clandestine collaborators or committedatrocities in secret, can be blackmailed with threats to reveal such information to thepublic by those with credible access to information on such skeletons in the closet(Nalepa 2010b). Needless to say, if the public still pays attention to what happened inthe past, the revelation of such skeletons could end a politician’s career. In return fortheir silence, individuals in possession of credible evidence of skeletons in the closetcan demand rents or policy concessions. Regardless of the currency in which the ran-som is paid out by the blackmailed politician, the quality of democracy suffers.

As I explained in Chapter 3, the mechanism behind purges is different: since get-ting rid of members who ran the agencies of former authoritarian regime is similar tofiring bureaucrats of a government agency, one can think of administrative purges asthe reverse of a delegation problem. If a newly elected politician takes office and car-ries out a thorough purge, he is forced to implement policy in conditions with whichhe is unfamiliar. Without the expertise of people who used to run the agencies un-der the ancien regime, he cannot know how policy implementation will be affectedby states of the world unknown to him. Conversely, a decision to forgo a purge or set-tle on implementing a leadership-only purge can be thought of as the equivalent ofdelegation to an agent who is equipped with expertise and thus able to adjust poli-cies to the state of the world. The dilemma facing the new politician is obvious to anystudent of principal agent models. The agent—in this case, the staff member of theauthoritarian agency—may have preferences that are so misaligned with those of theprincipal (the new politician) that he will use his expertise to implement policy hehimself prefers. On balance, this policy outcome may be worse for the principal thanhis own implementation, ridden with lack of expertise as it is.

These two intuitions already suggest that the mechanism behind purges is verydifferent from the mechanism of lustration, particularly when what is of interest is theability of new democratic politicians to represent the interests of their voters. Thus,these two categories of personnel transitional justice ought to be disaggregated. The

90 4 Measurement

remainder of this chapter describes the dataset that accomplishes this task and usesthis dataset to test the intuitive distinction between the effects of purges and lustrationon the quality of democratic elites.

4.2.2 Taking Time Seriously: Transitional Justice as a time series

Mechanisms for righting wrongs committed by former ruling elites, and commentaryon these wrongs, date back to the ancient Athenian democracy and its attempts to dealwith crimes committed by the Thirty Tyrants (Todd et al. n.d.). For obvious reasons,my dataset cannot cover a time span stretching back to the 5th century B.C. Adheringinstead to conventions in the conflict literature (Olsen et al. 2010, Binningsbø et al.2012, Sikkink & Walling 2007), my research team and I document the occurrence ofthese mechanisms in all countries that experienced civil war or transitioned to democ-racy between 1946 and 2016. Globally, there are 83 such states.

My country selection criteria build on the existing Autocratic Breakdown andRegime Transitions (GWF) dataset (Geddes et al. 2014) and the Post-Conflict Justice(PCJ) database (Binningsbø et al. 2012). I select countries that, as indicated by theGWF dataset, transitioned from a military, personalist, or party-based authoritarianregime in the post-1946 period. We include countries that are no longer democratic,but that experienced a democratic spell after 1946.13 My dataset also encompassescountries with multiple indicators of regime type, such as Burundi, which transi-tioned in 1993 from a military-personal regime, or Indonesia, which transitioned froma party-personal-military regime. We exclude all countries that are currently authori-tarian and have remained authoritarian for most of the post-1946 era as by my defini-tion, these countries are unable to implement transitional justice.

As far as democracies having experienced civil conflict are concerned, I begin byfollowing exactly the same criteria as the PCJ database; that is, I include all armedconflict episodes that have ended in the 1946-2006 time period. While the PCJ databasehas a conflict-episode structure, I aggregate the conflict episodes identified in the PCJdataset to the country level and remove all long-term authoritarian regimes that neverbecame democracies in order to remain consistent with the rest of my data.

If a country dissolved into a collection of smaller countries as a result of successfulsecession efforts, as Czechoslovakia and Yugoslavia did, I include any relevant tran-sitional justice events from the original country as the transitional justice events forthe most relevant successor country—that is, the one that actually inherited the tran-sitional justice institutions. All additional countries are coded as having transitionedat the date of independence. For example, we code the Czech Republic as the suc-cessor country following the 1993 dissolution of Czechoslovakia. The Czech Republicthus inherited the post-1989 transitional justice events attributed to Czechoslovakia.Slovakia is then coded as having transitioned in 1993. Serbia is coded as the successorcountry to Yugoslavia, while Slovenia, Bosnia and Herzegovina, Croatia, Macedonia,Kosovo, and Montenegro are all coded as having transitioned at their respective datesof independence.

Finally, I include information on countries and conflict periods that led to transi-tional justice but that had been excluded from previous datasets.14 I chose explicitly13 Examples of such countries include Russia, Egypt, and Thailand.14 Examples include Cyprus, which is excluded from GWF based on size, and Kenya,

which is excluded from PCJ despite its post-election violence in 2007-2008. This


to include small countries because such countries are immune neither to periods ofauthoritarian rule nor to periods of conflict, and as seen in the cases of East Timor orKosovo, can implement all forms of personnel transitional justice.

To create my dataset, I start with two major electronic databases: Keesings’ Recordof World Events and the Lexis Nexis Academic Universe.

The raw data include chronologies of events pertaining to purges and truth rev-elation procedures for all countries satisfying the selection criteria outlined above. Isearched them for information about events related to purges, TJ vetting, and truthcommissions in all relevant countries, beginning from the date of the transition todemocracy, the start of the post-conflict period, or both (in the case of conflicts whichoccurred in democracies), and ending in either 2016 or the year in which the countryreverted to authoritarianism.

There are, of course, serious limitations to relying on electronic databases alone.For one, they are biased towards large countries and countries with authoritarianregimes and conflicts that are better penetrated by the global network of human rightsoriented NGOs and journalists. For this reason, I supplement my database searcheswith numerous secondary sources ranging from historical accounts in monographsand chapters in edited volumes to articles. A bibliography of these secondary sourcesis provided in the Online Appendix.

Based on these searches, we created a chronology document for each country. Thechronology includes relevant information about the final authoritarian regime andtransition, conflict and post-conflict period, or both. Then, the team and I providerecords of each TJ event in chronological order, noting the date, a brief identificationof the event, the relevant state and non-state actors, a more detailed description of theevent, and the source where the information was obtained. In order for a personnel TJevent to be relevant, it must include an actor in his or her governing capacity enabling(in a positive event) or disabling (in a negative event) the pursuit of personnel TJ.

To be more specific, I define a positive TJ event as the submission of a TJ pro-posal to the floor of the legislature, the passage of such legislation, the upholding ofsuch legislation as constitutional by a supreme court, or the overturning of a presiden-tial veto against such legislation. In the case of truth commissions, the publication ofthe commission’s report(s) and the extension of the commission’s mandate are alsoconsidered to be positive TJ events. I define a negative transitional justice event, incontrast, as the voting down, vetoing, or striking down by the constitutional court ofa transitional justice proposal or law. Expanding the set of persons targeted by TJ orbroadening the set of “offenses” (where “offense” is defined in light of the TJ pro-cedure in question) to include more past or present positions constitutes a positivetransitional justice event, whereas attempts to narrow the set of targets or “offenses”are coded as negative TJ events.

The guiding principle in determining if an event is negative or positive is whetherit advances the TJ event forward or backward. Consider Onur Bakiner (2016)’s ob-servation that some truth commissions were disbanded before they could finish theirwork (examples provided by Bakiner include Indonesia, Turkey, Mexico, and Nepal).The temporal organization of our data, which accounts for progressive as well as re-gressive events, allows us to account for such setbacks in the work of truth commis-sions.

post-election violence in Kenya produced numerous domestic transitional justiceevents, including a truth commission.

92 4 Measurement

After each event’s category was determined (purge, lustration, or truth commis-sion), it was coded as positive or negative. Events that were not relevant for the datasetwere labeled as such, with an explanation of why they were excluded.

The number of positive and negative TJ events was then aggregated to create anannual panel, with countries as the cross section and time since transition as the tem-poral dimension. A panel assembled in this way allows for the creation of many dif-ferent measures of transitional justice. In addition, the raw chronologies themselvesallow researchers to experiment with different systems of disaggregation.15

A time series of positive and negative transitional justice events can also producetrends for individual countries, such as the ones offered in Chapter 1 (figures ).

Recall that these trends should not be interpreted as periods when personnel tran-sitional justice was implemented, though they may be correlated with those periods.Instead, the blue line represents the number of events in a given year since the tran-sition advancing the lustration process, while the red line represents the number ofevents delaying lustration.

These trends can be thought of as building blocks out of which scholars can con-struct measures pertaining to various aspects of personnel TJ. In the next section, Ioffer examples of three such measures and illustrate how one of them may be used toaddress the question motivating this book.

4.3 Constructing measures of Transitional Justice

Measures are never constructed in complete abstraction from research questions (Go-ertz 2006, Coppedge 2012). My dataset collection and this specific book are motivatedby the question of how dealing with former authoritarian elites impacts the quality ofdemocracy. However, “dealing with former authoritarian elites” could mean dealingwith them harshly or mildly or not at all; dealing with them immediately or later afterthe transition; or being consistent in pursuing a certain type of personnel transitionaljustice process.

In order to capture these three aspects of dealing with former authoritarian elitesand their collaborators, I develop three measures that use my yearly positive and neg-ative personnel TJ events as building blocks: (1) Severity, (2) Urgency and Delay, and(3) Volatility. Below, I introduce each in turn.

15 For instance, among the events included in the chronologies are ones that couldnot be classified as purges, lustration, or truth commissions. These were labeled“non-events” and include trials, amnesties, apologies, and victim compensation.Other researchers may want to create their own categories out of these events, anundertaking made possible by our organization of the chronologies. Moreover, mytechnique of labeling events as positive or negative could be fruitfully applied tocriminal trials. The initiation of an investigation, along with an indictment, couldbe the first progressive event in a trial proceeding. Reducing the number of countson which a defendant is charged would be a regressive event, as would acquittal orthe commuting of a sentence.

4.3 Constructing measures of Transitional Justice 93

Fig. 4.5: Positive and Negative lustration events in 75 post-authoritarian coun-tries from regime transition to 2018

0

Albania

0 10 20 30 40 50

00

Argentina

0 10 20 30 40 50

0

0

Bangladesh

0 10 20 30 40 50

0

0

Benin

0 10 20 30 40 50

0

0

Bolivia

0 10 20 30 40 50

0

0

Bosnia

0 10 20 30 40 50

0

0

Brazil

0 10 20 30 40 50

0

0

Bulgaria

0 10 20 30 40 50

0

0

Burundi

0 10 20 30 40 50

0

Cape Verde

0 10 20 30 40 50

0

0

Chile

0 10 20 30 40 50

00

Colombia

0 10 20 30 40 50

0

0

Comoros

0 10 20 30 40 50

0

0

Crotia

0 10 20 30 40 50

0

0

Cyprus

0 10 20 30 40 50

0

0

Czech Rep

0 10 20 30 40 50

0

0

Dominician Rep

0 10 20 30 40 50

0

0

East Timor

0 10 20 30 40 50

0

Ecuador

0 10 20 30 40 50

0

0

Egypt

0 10 20 30 40 50

0

0El Salvador

0 10 20 30 40 50

00

Estonia

0 10 20 30 40 50

0

0

Georgia

0 10 20 30 40 50

0

0

Germany

0 10 20 30 40 50

0

0

Ghana

0 10 20 30 40 50

0

0

Greece

0 10 20 30 40 50

0

0

Grenada

0 10 20 30 40 50

0

Guatemala

0 10 20 30 40 50

0

0

Guinea Bissau

0 10 20 30 40 50

0

0

Guyana

0 10 20 30 40 50

0

0Haiti

0 10 20 30 40 50

0

0

Honduras

0 10 20 30 40 50

0

0

Hungary

0 10 20 30 40 50

0

0

India

0 10 20 30 40 50

0

0

Indonesia

0 10 20 30 40 50

0

0

Kenya

0 10 20 30 40 50

0

Kosovo

0 10 20 30 40 50

0

0

Latvia

0 10 20 30 40 50

0

0

Lebanon

0 10 20 30 40 50

0

0

Liberia

0 10 20 30 40 50

0

0Lithuania

0 10 20 30 40 50

0

0

Macedonia

0 10 20 30 40 50

0

0

Mali

0 10 20 30 40 50

0

0

Mexico

0 10 20 30 40 50

0

0

Moldova

0 10 20 30 40 50

0

Mongolia

0 10 20 30 40 50

0

0

Montenegro

0 10 20 30 40 50

0

0

Nepal

0 10 20 30 40 50

0

0

Nicaragua

0 10 20 30 40 50

0

0

Niger

0 10 20 30 40 50

0

0

Nigeria

0 10 20 30 40 50

00

Northern Ireland

0 10 20 30 40 50

0

0

Pakistan

0 10 20 30 40 50

0

0

Panama

0 10 20 30 40 50

0

Paraguay

0 10 20 30 40 50

0

0

Peru

0 10 20 30 40 50

0

0

Philippines

0 10 20 30 40 50

0

0

Poland

0 10 20 30 40 50

0

0

Romania

0 10 20 30 40 50

0

0

Russia

0 10 20 30 40 50

0

0

Sao Tome and Principe

0 10 20 30 40 50

00

Serbia

0 10 20 30 40 50

0

0

Sierra Leone

0 10 20 30 40 50

0

Slovakia

0 10 20 30 40 50

0

0

Slovenia

0 10 20 30 40 50

0

0

South Africa

0 10 20 30 40 50

0

0

South Korea

0 10 20 30 40 50

0

0

Spain

0 10 20 30 40 50

0

0

Sri Lanka

0 10 20 30 40 50

0

0

Taiwan

0 10 20 30 40 50

0

0Thailand

0 10 20 30 40 50

00

Tunisia

0 10 20 30 40 50

0

Turkey

0 10 20 30 40 50

0

0

Ukraine

0 10 20 30 40 50

0

0

Zambia

0 10 20 30 40 50

Years since transition

Positive events Negative events

94 4 Measurement

Severity

Among those evaluating personnel transitional justice, there is a high demand foran instrument to capture its intensity. Such a measure would place the severity ofa personnel transitional justice procedure somewhere between the two extremes ofminimal and severe. One way to approach this is by taking the cumulative effect ofrelative changes in the law since the transition as follows:

S =

∑t(P

t)∑t(P

t +Rt) + 1

where t is the subscript over time, Rt is the number of regressive events in a countryin period t, and P t is the number of progressive events in period t. This measureis simply the total number of positive events over the total number of events. In acountry with no personnel transitional justice, S will obviously take on 0; the “+1” inthe denominator ensures that the measure is well-defined. The more positive events acountry has among its total events, the closer its score is to 1. The figure below appliesthis measure to thorough purges.

Fig. 4.6: Severity of Thorough Purges

Latvia

India

Mali

Czech Republic

Kenya

Bulgaria

Hungary

Greece

Albania

Argentina

Bangladesh

Georgia

Guinea Bissau

Macedonia

Panama

Paraguay

Poland

Tunisia

Ukraine

Lithuania

Portugal

Romania

Comoros

0.00 0.25 0.50 0.75 1.00

Rate of Severity

Cou

ntry

Severity Measure for Purges

Even though existing datasets treat all purges as equal and the same, my measuredemonstrates a fair degree of variation. Figure 4.6 ranks thorough purges from leastsevere (Comoros) to most severe (Latvia). In Latvia, post-transition governments ini-tiated a series of purges that banned anyone who had been active in the Soviet Com-munist Party or its affiliate organizations from running for office. These bans were up-held despite numerous court challenges throughout the 1990s and 2000s (Stan et al.2009). This differs starkly from Comoros, where a weak purging of the military was


initiated after its 2006 transition to democracy. Several years later, the purge in Co-moros was undone, and the “militia” loyal to the previous regime were reintegratedinto the military (UN Integrated Regional Information Networks (Nairobi) 2011).

As further insight into the usefulness of my severity measure, consider Table 4.1below, which compares its application to “lustration” from TJDB and “vetting” fromthe Transitional Justice Research Collaborative. While the event counts implied byTJDB and TJRC equate countries like Albania and Argentina—with two lustrationevents for TJDB, and two and three vetting events, respectively, for TJRC—my measureof severity shows a clear distinction between the two countries: TJ vetting in Argentinawas significantly more severe (.83) than TJ vetting in Albania (.55). Albania, in fact,has much more in common with Bulgaria (.59), which has three lustration events ac-cording to TJDB. As shown in Table 4.1, the nature of my disaggregated data allowsfor a more nuanced understanding of the severity of personnel transitional justice.

Other illustrative measures: Urgency and Delay and Volatility

While severity is probably the TJ measure most needed by the research design I pro-pose in this book, TJ scholars around the globe have long called for a measure to cap-ture how long a country waits for a TJ event. The most straightforward measure mightcount the number of years lapsed before the first positive transitional justice event as aproportion of years lapsed since the transition itself. Herein, however, lies the problemwith the censored nature of my data. Should a country that transitioned five years agoand implemented TJ vetting in its fourth year of democracy be equivalent to a countrythat transitioned twenty years ago and implemented TJ vetting only four years ago?The naive measure of delay proposed above would give both a score of .8. Yet, thelatter country seems to be more of a latecomer to the TJ vetting process than does theformer.

Importantly, it is hard to vet elites for collaboration with the authoritarian regimein the immediate aftermath of transition. In contrast to purges, vetting may rely onuncovering and organizing files of the secret political police, an activity that, depend-ing on the level of organization of these files, may take years. Similar delays caused bythe very nature of unearthing evidence plague truth commissions. In order to accountfor such startup costs, I propose to measure urgency/delay for each country with:

U =TN − P 1

(TN − T 1)(P 1 − T 1)

16 where TN is 2016 or the last year of the democratic spell before the country’s rever-sal to authoritarian rule,17 P 1 is the first year following the country’s transition thatfeatured a progressive personnel transitional justice event, and T 1 is the first year fol-lowing the country’s transition.18 To understand how this measure works, considerFigure 4.7 below:

16 I clarify that in all three measures, subscripts do not represent exponents but timeindices.

17 TN need not be 2016, as illustrated in the case of Thailand, which experienced amilitary coup in 2014.

18 In countries like Thailand, T 1 will be subtracted from the year of the authoritarianreversal rather than from 2016.

96 4 Measurement

Table 4.1: Comparison Table

Countries TJDB (Lustration) TJRC (Vetting) Severity (TJ Vetting)Albania 2 2 0.545

Argentina 2 3 0.80Bangladesh 0 0 0.50

Benin 0 0 0Bolivia 0 0 0Bosnia 3 5 0.667Brazil 0 0 0.80

Bulgaria 2 4 0.588Burundi 0 0 0

Cape Verde . 0 0Chile 0 0 0.667

Colombia 0 0 0.50Comoros . 0 0.50Croatia 0 0 0Cyprus . 0 0

Czech Republic 2 2 0.81Dominican Republic 1 1 0.50

East Timor 0 0 0Ecuador 0 0 0Egypt 0 0 0.50

El Salvador 0 0 0Estonia 0 1 0.762Georgia 0 1 0.556

Germany 1 2 0.80Ghana 0 0 0Greece 2 5 0.842

Grenada . 0 0Guatemala 0 0 0.75

Guinea Bissau 1 0 0.60Guyana . 0 0

Haiti 0 0 0Honduras 0 0 0Hungary 1 2 0.684

India 0 0 0.75Indonesia 0 0 0

Kenya 0 1 0.667Kosovo . 0 0Latvia . 1 0.778

Lebanon 0 0 0Lesotho 0 0 0Liberia 0 2 0

Lithuania 1 4 0.75Macedonia 0 1 0.636Madagascar 0 0 0

Malawi 0 0 0Mali 0 0 0.833

Mexico 0 0 0.667Moldova 0 0 0.50Mongolia . 0 0

Montenegro . 0 0Nepal 0 2 0.50

Nicaragua 0 0 0.50Niger 0 0 0

Nigeria 1 1 0.50Northern Ireland . 0 0.50

Pakistan 0 0 0.50Palestine . 0 0Panama 0 0 0.667

Paraguay 1 1 0.833Peru 0 0 0.50

Philippines 0 1 0Poland 1 2 0.625

Portugal 1 2 0.706Romania 0 2 0.692Russia 0 0 0.30

Sao Tome and Principe 0 0 0Senegal 0 0 0Serbia 0 1 0.833

Sierra Leone 0 0 0Slovakia . 1 0.714Slovenia . 1 0.429

South Africa 0 0 0South Korea 0 0 0.50

Spain 1 0 0.50Sri Lanka 0 0 0

Sudan 1 0 0Taiwan . 0 0

Thailand 0 0 0Tunisia 0 0 0.667Turkey 1 1 0Ukraine . 0 0.692

Venezuela 0 0 0Zambia 0 0 0.50


Fig. 4.7: Urgency Measure

P 1T 1 TN

TN − P 1

P 1 − T 1

TN − T 1

In the numerator, TN - P 1 corresponds to how soon the first positive TJ eventoccurs. The larger its value, the more urgent is the implementation of personnel TJ.The denominator contains two expressions: (1) the length of the democratic spell,TN−T 1, which is used to to account for the fact that countries with longer democraticspells have more opportunities to engage in transitional justice, and (2) (P 1 − T 1), aweight capturing the number of years lapsed before the first progressive TJ event. Bothof these expressions lower the value of the urgency measure.

U assumes the value of zero when a country has no positive TJ events. Given thaturgency and delay are a function of time, a country that has not yet implemented aparticular type of TJ cannot possibly have a non-zero value for the measure. The mea-sure will approach the value of 1 when P 1 approaches T 1—that is, when the firstpositive event occurs very soon after the transition. Because T 1, TN , and P 1 can onlytake positive integer values and T 1 < P 1, the measure will never actually assume 1.The measure is also well-defined (the denominator cannot be zero).19 For a substan-tive illustration of the urgency measure, Figure 4.8 below presents its values for mydata’s universe of TJ vetting events.

This exercise captures the tremendous simplification that occurs when scholars tryto capture the implementation of transitional justice with a dummy variable. Clearly,the extremely delayed vetting in Spain cannot be treated as equivalent to the urgentlyimplemented vetting in Greece. While Spain transitioned to democracy in 1976, it didnot begin a TJ process until nearly forty years later, when a proposal to declassify mil-itary reports from the Franco era was made in the legislature in 2013 (Congreso delos Diputados, Comision de Cultura 2013). Greece, however, began its vetting pro-cess almost immediately after its July 1974 transition, removing the leadership of thearmed forces, police, and banking and public sector by mid-August of that same year(Alivizatos & Diamandouros 1997, Keesing’s Record of World Events 1974).

19 I code the data beginning with the first year after the transition, which is consistentwith the fact that the first positive event has to take place after the transition.

98 4 Measurement

Fig. 4.8: Urgency of Truth Commissions

PortugalArgentina

BoliviaPhilippines

ParaguayMalawi

Sri LankaIndonesia

NigeriaEast Timor

ZambiaEgyptChile

Czech RepublicEstonia

South AfricaGuatemala

Northern IrelandGhanaMexico

PeruSerbiaKenyaNepal

BangladeshTunisia

GermanyLatvia

LithuaniaSierra Leone

LiberiaThailand

BrazilSloveniaGeorgiaLesotho

PalestinePolandBosnia

LebanonHonduras

PanamaSouth Korea

EcuadorColombiaGrenadaBurundi

MoldovaGuyana

SpainVenezuela

0.00 0.25 0.50 0.75 1.00

Rate of Urgency

Cou

ntry

Urgency Measure for Truth Commission Events

Finally, TJ scholars may be interested in a measure that would capture the stabilityof transitional justice. In this illustrative measure, I focus on the magnitude of policyswings regarding a given transitional justice mechanism in a given country. I interpretpolicy swings as dramatic shifts from progressive to regressive events or vice versa.Consequently, my volatility measure only applies to countries that have experiencedboth progressive and regressive events. Polarization could be measured simply withthe number of years separating the year with the maximum number of positive eventsand the year with the maximum number of regressive events. Such a measure, how-ever, would fail to account for the fact that a country can experience both positive andnegative events over the course of the same year. Thus, in order to capture this netvalue of positive and negative events, I propose:

V =maxt{P t −Rt} −mint{P t −Rt}

(|mediant {Tmax} −medianT { Tmin}|+ 1) ∗maxc{maxt{P t −Rt} −mint{P t −Rt}}

The value of (P t−Rt) is highest in the year in which a country records the maximumnumber of positive events net of negative events and is at its lowest when a countryrecords the maximum number of negative transitional justice events net of positiveevents. Thus,maxt{P t−Rt} is the maximum of net events, andmint{P t−Rt} is theminimum of net events. The difference between these two expressions, which consti-tutes the numerator in my measure, captures the policy swing in personnel TJ that isreferred to above.


The denominator of this measure is made up of two parts. The first, subtractingmedianT { Tmin} frommedianT {Tmax}, accounts for the number of years the coun-try in question took to experience such a policy swing. The volatility measure shouldincrease when these years are fewer. At the same time, because the maximum andminimum net values could be associated with more than a single year in a country’spost-transition history, I take the median year of all maximum net values and the me-dian year of all minimum net values. Since it does not matter whether the uptick inpositive events precedes or succeeds the uptick in regressive events, I take the abso-lute value of the difference between the two medians. The second expression in thedenominator is a weight ensuring that my measure does not exceed 1. The denomi-nator represents the largest possible swing across all countries in my data (hence thesubscript “c” in the maximum expression:maxc{maxt{P t −Rt}−mint{P t −Rt}}.

The operation of this seemingly complex measure is conveniently illustrated inFigure 4.9 below. To make the figure more transparent, I refrain from including mul-tiple years with the maximum and minimum scores of (P t − Rt), which are labeledat Tmax and Tmin, respectively, in the figure.

Fig. 4.9: Volatility Measure

t = Tmaxt = 0 2016

or end ofdemocratic

spell

t = Tmin

min(P t −Rt)

max(P t −Rt)

Progressive Events

Regressive Events

|Tmax − Tmin|

100 4 Measurement

P takes the value of zero when a country does not experience any policy swingin transitional justice and approaches 1 when a country exhibits a significant policyswing in a relatively brief time period. Note that whether a country has a measure ofpolarization at all will vary from one TJ mechanism to another, as the measure can onlybe applied to countries that exhibited both positive and negative personnel TJ events.Countries that did not implement any transitional justice of a given type, and thosethat had only progressive or regressive events, do not have polarization measures asthey cannot experience policy swings. This explains why Figure 4.10, illustrating theapplication of volatility to the case of truth commissions, has fewer cases than thefigures illustrating severity and urgency. Figure 4.10 illustrates the operation of thismeasure using lustration as an example.

Fig. 4.10: Volatility of Lustration

Estonia

Romania

Lithuania

Bulgaria

Slovenia

Hungary

Czech Republic

Macedonia

Poland

Russia

Slovakia

Moldova

Georgia

Latvia

Germany

Albania

Spain

Ukraine

0.00 0.25 0.50 0.75 1.00

Rate of Polarization

Cou

ntry

Polarization Measure for Lustration Events

Volatility, by far the most complex of the three measures presented here, exposesthe potentially greatest transitional justice coding procedures relying on dummy vari-ables that only capture whether a country has experienced TJ and possibly providea date for this TJ. First, consider cases such as Ukraine, Spain, and Albania, wherevolatility is very low. These three countries’ volatility values correspond to situationswhere transitional justice was implemented once and for all; thus, the coding is notaffected by the time at which data were collected. Contrast this with highly polar-ized countries such as Lithuania, Romania, and Estonia, where transitional justicewas implemented only to be revoked shortly after. In such countries, the coding ofthe presence or absence of the transitional justice mechanism is highly dependent onthe moment in time when data collection took place.

4.4 Conclusion

This chapter introduced readers to my operationalization of quality of democracy andtransitional justice, the dependent variable and independent variable used to address

4.4 Conclusion 101

the three interwined questions of this book: who is selected into office, how do theybehave in office, and when do they delegate to bureaucrats.

Scholars of democracy and democratization rely overwhelmingly on expert sur-veys for measures of the quality of democracy. This is understandable as many of theconcepts they strive to measure are inherently unobservable. I use readily availabledata from two such surveys, each of which presents some advantages and some chal-lenges.

My first measure focuses on the programmaticness of political parties and is basedon surveys. In these surveys, country experts are asked to rate the quality of repre-sentation of specific post-transition parties. This measure allows me to most directlyexamine the impacts of truth revelation procedures on programmaticness, the extentto which parties deliver policies desired by the electorate; by extension, this allowsfor a direct examination of the mechanisms behind the role of these procedures inpreventing blackmail with information about past collaboration, which may impedeprogrammaticness. Since the quality of representation and the policy positions of par-ties are measured at the party level, this measure is ideally suited for operationalizingparameters from the second chapter of this book, where both representation qualityand policy positions are critical.

Recall that the frequency of the separating equilibria depends on the proximitybetween the position of the new democratic politician and that of the former author-itarian elite. Although a measure of the spatial location of a party does not map onto my theory as well as a measure of the ideal point of a specific politician would, itis better than cruder measures at the country level. Of course, the cost of using thismeasure is that it was taken only at one point in time and therefore does not allow meto unleash the power of my transitional justice time series data.

To remedy this, I use two categories of measures that have been collected overtime. Both come from the Varieties of Democracy project. The first is the Political Cor-ruption Index, which is composed of variables based on several questions pertainingto corruption in different branches and at different levels of government. I will useboth the general index, which provides measures for all countries and post-transitionyears in the Personnel Transitional Justice Dataset, as well as the components of thisindex (which are missing values for some country-years).

The second V-Dem measure, also collected as a time series, is Political Power Dis-tributed by Socio-economic status. The advantage of this measure is that it is veryunlikely to be endogenous to the same factors that are responsible for the onset oftransitional justice program implementation. At the same time, it seems to tap verywell into the observable characteristics of authoritarian elite reproduction by captur-ing the failure to dissociate political power from economic dominance.

The employment of three different measures allows me to create a variety of re-search designs, using techniques such as difference-in-difference for analyzing paneldata and hierarchical modeling for analyzing nested data when the unit of analysis isa specific party in a country.

In measuring transitional justice, however, I could not rely as comfortably on exist-ing scholarship and was propelled to collect and organize my own data and developnew measures of TJ. Hence, this chapter also introduced my original Global Transi-tional Justice Dataset. This dataset is innovative in a number of ways. First, it recordstransitional justice as events unfolding over time following the year of transition. Thisallows me to account for instances of delayed transitional justice. It also allows me todesign innovative ways of measuring aspects of transitional justice such as volatility

102 4 Measurement

and, critical to this book, severity. I encourage scholars to use my transitional justiceevents data as building blocks for constructing new measures motivated by their spe-cific theoretical interests.

A second innovation of my dataset is that it parses out similar yet distinct ways ofdealing with personnel of the former authoritarian regime—namely, purges and lus-tration. Whereas purges remove from office elites whose actions under the former au-thoritarian regime were known or perpetrators who committed crimes that are com-mon knowledge, lustration and truth commissions reveal information that was keptsecret. I argue that due to this, lustrations have more in common with truth commis-sions than they do with purges.

5

Lustration and Programmatic Representation

In the previous chapter, I explained the data collection strategy that I adopted to an-swer the theoretical question posed in the first part of the book: how does transitionaljustice enhance democratic representation in a country recovering from authoritarianrule? I broke this question down into the selection of politicians, politicians’ behaviorin office, and delegation to bureaucrats. My strategy for measuring the independentvariable in this study involves disaggregating transitional justice into transparencyregimes (lustration and truth commissions) on the one hand and purges (of the thor-ough and leadership variety) on the other. Following this coding, I proposed threeways of measuring the intensity of transitional justice with severity measure that takesthe ratio of progressive transitional justice events to regressive events. This measurecan be modified to provide values that vary over time, thus offering considerable flex-ibility when assessing the effects of transitional justice mechanisms. In this chapter, Iuse this measure to examine the effect of transitional justice on democratic quality. Ifirst focus on lustration; the next chapter examines the effects of truth commissions,and the final chapter is devoted to purges. In the conclusion of this book, I again usean HLM model to summarize the effects of all three transitional justice mechanismson the quality of democracy measured as power distributed by socio-economic status.

5.1 How does lustration affect who wins office and how theybehave?

I begin with an examination of the implications of the model from Chapter 2, which Iused to make predictions about the quality of representation in response to lustration—that is, legislation dealing with secret forms of collaboration.

The model in Chapter 2 described the behavior of individual politicians and of-ficers of former authoritarian regimes. Since data on political preferences of individ-ual officers and the quality of representation at the level of individual legislators isvery difficult to obtain, I rely on slightly more aggregated information to evaluate themodel. Namely, I test the implications of my model using a measure of party pro-grammaticness, defined in the literature as a situation in which parties run on policyplatforms that are recognizable to voters, voters support parties because of these plat-forms, and voters can later hold parties accountable for implementing those platforms

104 5 Lustration and Programmatic Representation

(Kitschelt & Freeze 2010). Thus, programmaticness is associated in the literature withhigh-quality democratic representation.

The original measure made by the creators of the Democratic Accountability andLinkages Project (DALP) was constructed from a large-scale survey of experts in 88different countries answering questions about the cohesion, salience, and polarizationof 506 parties. I restricted this dataset of partisan programmaticness to countries thattransitioned from party-based authoritarian rule after 1946. This resulted in over 300parties in 61 countries, as not all countries from the Global Transitional Justice Datasetare represented in DALP. The subsection below derives hypotheses from Chapter 2’sempirical implications and tests them with a hierarchical model.

5.1.1 Hypotheses

The model’s main implication is that the probability that information from secret po-lice files has been exposed by lustration (parameter π in the model) raises the qualityof representation because it prevents blackmail of politicians with these secret files.This can be stated as the following hypothesis:

Main Hypothesis: All else kept constant, the quality of representation will in-crease as the severity of lustration increases.

Lustration is operationalized with the severity measure developed in the previouschapter.

The model yields several additional implications that justify the inclusion of ad-ditional variables as controls in the statistical model presented. For instance, a sec-ond empirical implication of the model is that quality of representation decreases asthe cost of having compromising information revealed increases (parameter F in themodel). This means that as voters, and consequently politicians, become more con-cerned with having skeletons in their closet revealed, blackmailers with access to se-cret police files can distort policy more. In order to operationalize this concern over ex-posed skeletons in the closet, I make use of the fact that in democracies recovering fromauthoritarian rule, the salience of who was and who was not a collaborator diminishesover time.(Stan 2006, Bernhard & Kubik 2014, Wilde 1999, Cohen 1995, De Brito et al.2001) This claim finds support in the literature on historical memory,Bernhard & Ku-bik (2014) and is also corroborated by the fact that in a young democracy, more voterswill have experienced (and will remember) life under autocracy (Pop-Eleches 2007,Pop-Eleches & Tucker 2011, 2012) than in a seasoned democracy. Hence, I operational-ize F as the years lapsed since the transition. 1 This leads to my second hypothesis:

Additional Hypothesis All other things held constant, as the time lapsed since thetransition to democracy increases, the quality of representation will increase.

This operationalization is somewhat coarse because the cost of revealing collabo-ration is not only a function of how voters react to allegations of collaboration but alsoof other institutional features. For example, it could be that revelations about collabo-ration are costlier in electoral systems with the personal vote (such as Single MemberDistricts or Open List Proportional Representation) than in systems where voters can-not directly specify which candidate gets elected into officeCarey & Shugart (1995).Alternatively, voters may be more interested in elections in some countries than in

1 The date of transition is taken from the Authoritarian Regimes Database (Geddeset al. 2014), which my research assistants and I have updated to 2017.

5.1 How does lustration affect who wins office and how they behave? 105

others; this would imply that revelations are mediated by an overall interest in elec-toral participation or turnout. Nevertheless, I believe that the political cost of exposedskeletons in a politician’s closet should be higher, on average, in recently transitionedcountries than in countries that transitioned a long time ago.

The third implication of my theory is that the ideological distance between Politi-cians and Officers does not impact the quality of democracy. However, as mentionedearlier, this finding is not robust to changing the functional form of players’ utilityfunctions to quadratic. Nevertheless, it is important to control for this distance in myregression. Like the dependent variable, quality of democracy, my operationalizationof distance relies on the Democratic Accountability and Linkages Project. It is alsomeasured at the party level, as different parties in a democratic polity have differentdistances from the successor autocrat. Specifically, I take advantage of the fact thatDALP reports average ideological placements of political parties in countries frommy sample. Since some of these parties are successor authoritarian parties, I use theirplacement to proxy for the ideological placement of blackmailing Officers. Therefore,for each non-successor party p in country k, I define party distance as the absolute differ-ence in left-right placement between party p and the country’s successor authoritarianparty.

Some countries in my sample do not have a successor party to the authoritarianregime. For these countries, I replaced the missing distance with the average distancebetween successor and non-successor parties across all remaining countries, and I in-cluded an indicator variable to account for these cases.2

Finally, consider the cost of bluffing, c. Importantly, in the model from Chapter 2,the cost of bluffing does not appear in the predicted misrepresentation. However, anincrease in the costs of bluffing to the Officer increases the chances of a non-separatingequilibrium, an equilibrium where transitional justice does not work as “it should”(in the sense that even with lustration, bluffing takes place). Thus, in addition to thequality of representation the cost of bluffing affects whether politicians are bluffed orblackmailed with actual evidence. Hence, I will also control for the cost of bluffing.

I propose to operationalize this cost with the Press Freedom Index as reported byReporters Without Borders.3 The reasoning behind this operationalization is as fol-lows: when an officer of the authoritarian regime tries to bluff a politician into thinkingthat he is in possession of evidence against the politician, such a bluff can be easily ex-posed in an environment with free media. Put differently, claims regarding collabora-tion between a politician and the former secret police will likely be reported, and if thepress is free, claims will be investigated and eventually denounced if they are foundto be false. A wide literature discusses the thoroughness of journalists in revealingevidence of former secret police conduct(Laplante & Phenicie 2009, Pinto 2010, Chap-man 2009). Although the scrutiny offered by journalists is not perfect, the higher thedegree of media independence, the more likely it is that a Officer’s bluff will be called.

Taking together these empirical implications leads to the following linear specifi-cation:4

2 See Empirical Appendix for a replication of my analysis limited to the subsampleof countries for which I do have information on the successor party.

3 To avoid confusion, I linearly transformed this variable to make lower rankings cor-respond to lower press freedom in a given country.

4 See Empirical Appendix for a derivation of this specification.


programmaticnessp,k =βk + βd ∗ party distancep,k + βs ∗ severityk+

βn ∗ years since transitionk + βc ∗ press freedomk+

βX∑

Xk + ep,k + ek

where programmaticness is operationalized as cosal3 for party p in country k.5 Inaddition to the main variables of interest described above, the model incorporates aseries of country-level control variables (

∑Xk), as well as a series of country-specific

intercepts (βk).As implied by the equation above, the outcome of interest (programmaticness) and

one regressor (party distance) are party-level attributes, while the remaining variablesare measured at the country-level. To account for this structure, I use multilevel re-gression. Given that parties are nested within countries, estimating this model usingordinary least squares regression (OLS) would risk biased and inefficient results be-cause characteristics of parties within the same country are not independent of oneanother.6 Multilevel regression is a compromise between assuming all parties are in-dependent from each other and assuming that all variation across parties can be ex-plained by their country. A multilevel approach structures variation in the dependentvariable according to party-level and country-level characteristics. Specifically, I willestimate the model above with a series of country-level coefficients (βj) that share anunderlying distribution with a global mean (µβj ) and standard deviation (σβj ).7 Thisequation is the best way of modeling the outcome of interest given the data available.

Before presenting and discussing the results, I address one possible concern: ifcountries where lustration was successfully implemented also have parties that arebetter at representing voters, I risk overestimating the relationship between lustra-tion severity and quality of representation. Such overestimation, however, would havebeen captured by the country-level intercepts. Put differently, if countries that imple-mented lustration were also those with more programmatic parties, this regularitywould be captured by the country-specific estimates.

The results of the estimations are shown in Table 5.1. For ease of interpretation,all variables have been linearly transformed to range between 0 and 1. All modelspresented here include country-level intercepts (estimations for βj) and an indica-tor variable controlling for missing successor parties, both of which are omitted fromthe table for space considerations. The table reports the average change in quality ofrepresentation by a non-successor party associated with an increase by one unit in thecovariate of interest.

5 The details on how this measure was derived from DALP data are provided inAppendix 9.5.1.

6 Traditional solutions to working with data of this kind (such as fixed effects orclustered standard errors) cannot be applied here because some countries only haveone party that is not successor authoritarian.

7 Assuming that all parties are independent of one another would require estimatinga fully pooled model with one global intercept. Assuming total dependency oncountry characteristics would mean a traditional fixed-effects approach that wouldestimate one intercept per country (minus a base category). The advantage of themultilevel model is that when there is very little observed variation at the country-level, the coefficient can be estimated to be close to the mean, µβj .


Table 5.1: Quality of representation and severity of lustration

Model 1 Model 2 Model 3 Model 4 Model 5party distance −0.103∗∗∗ −0.107∗∗∗ −0.109∗∗∗ −0.111∗∗∗ −0.111∗∗∗

(0.034) (0.034) (0.034) (0.034) (0.034)

severity 0.278∗∗∗ 0.255∗∗∗ 0.208∗∗∗ 0.209∗∗∗(0.062) (0.064) (0.069) (0.070)

years since transition 0.150 0.108 0.113(0.103) (0.107) (0.115)

press freedom 0.370∗ 0.369∗(0.215) (0.217)

opposition status 0.005(0.038)

Constant 0.379∗∗∗ 0.278∗∗∗ 0.217∗∗∗ −0.039 −0.044(0.032) (0.036) (0.055) (0.158) (0.163)

Country intercepts Y Y Y Y YMissing successor Y Y Y Y YObservations 313 313 313 307 307Log Likelihood 177.450 184.327 184.036 179.721 177.384Akaike Inf. Crit. −344.900 −356.654 −354.073 −343.442 −336.768Bayesian Inf. Crit. −326.169 −334.177 −327.849 −313.627 −303.226

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

The resulting estimations are generally consistent with the implications of the the-oretical model from Chapter 2. Most importantly, they show that higher scores of lus-tration severity are associated, on average, with more programmatic parties. The effectis statistically significant at the 0.05% confidence level, and robust to the inclusion ofcountry-specific intercepts and other relevant covariates. In addition, its average effectis substantial: the difference in programmaticness between a country that did not at-tempt any lustration at all (with a severity score of 0) and a country with the highestlustration severity (with a severity score of 0.82) is 0.17 points, according to the pointestimate of Model 4. Since the variable quality of representation ranges from 0 to 1, thisaverage effect accounts for one-fifth of the possible range of my outcome of interest.

The relationship between lustration severity and quality of democracy interpretedas programmaticness can be shown in yet another way. Figure 5.1 provides a visualrepresentation of the main relationship of interest in this paper: between programmat-


icness and severity of lustration. The figure shows the severity score on the horizontalaxis and the average country effects estimated in model 1 on the vertical axis. Model1 includes only the regressor that was measured at the party level—party distance—and estimates a series of country-specific intercepts that capture the average country-programmaticness. Figure 5.1 suggests a positive relationship between these two vari-ables. Furthermore, this pattern seems to be much clearer for the observations that fallin the right half of the panel: the countries that implemented more severe lustrationpolicies. This variation supports my decision to measure transitional justice as a pro-cess rather than as a dichotomous measure.

Fig. 5.1: Severity of lustration and average quality of democracy: Intercept bycountry

0.0 0.2 0.4 0.6 0.8

−0.

3−

0.2

−0.

10.

00.

10.

20.

30.

4

Severity of lustration (severity)

Cou

ntry

coe

ffic

ient

s (M

odel

1)

ALB

ARG

BGD

BEN

BOL

BRA

BGRCHL

COL

HRV

CZE

DOM

ECU

EGY

SLV

EST

GEO

DEU

GHA

GRC

GTM

HND HUN

IND

IDN

KEN

LVA

LBN

LTU

MKD

MLI

MEX

MDA

MNG

NIC

NER

NGA

NIR

PAK

PAN

PRY

PER

PHL

POL

PRT

ROM

RUS

SEN

SRB

SVK

SVN

ZAF

KOR

ESP

TWN

THA

TUR

UKR

URY

VEN

ZMB

Secondly, the evidence shows that the severity of transitional justice, and specif-ically of lustration policies, is positively related to the quality of representation. Thisresult is promising insofar as it suggests that lustration policies produce, on average, ahigher quality of representation after a transition. However, the theoretical model also


shows that lustration alone does not prevent blackmail of former dissidents-turned-politicians. The proper functioning of lustration policies requires an accompanyingfree press to prevent bluffing.

I also find a negative and statistically significant effect of the ideological distancebetween each party and its successor (party distance). This finding is intuitive, thoughnot consistent with the model’s predictions—where distance was predicted not to havean effect at all. However, as I pointed out above, the irrelevance of distance between theblackmailer and the politician from the model in Chapter 2 is not robust to modifyingthe utility functions of the players’ to quadratic (Ang & Nalepa 2019). A hypothe-sis consistent with empirical implications from the model with quadratic preferenceswould be that as the distance between each party and the successor party increases,the programmaticness should decrease. This result is not only what happens empiri-cally, but the empirical finding is robust to a number of specifications, the most notableof which uses only countries with successor parties (see Empirical Appendix 9.2).

This result could, however, be driven by preexisting dynamics between the oppo-sition and authoritarian parties. Suppose, for instance, that parties that existed duringthe authoritarian period share some ideological characteristics that make their mem-bers more likely to be blackmailed. I control for this possible confounder using a mea-sure of how active the opposition was in representative state institutions during theauthoritarian era. Cheibub et al. (2010) report data on the status of the oppositionin authoritarian legislatures.8 In my sample, only two authoritarian regimes had nolegislature at all, so I turned the variable opposition status into an indicator taking thevalue of 1 if there were multiple parties represented in the legislature and 0 otherwise.As can be seen in model 5, the inclusion of this variable does not change my results.

Models 3-5 include the variable years since transition, which captures time since thetransition to democracy as a proxy for the cost of having skeletons revealed (F ). Inthe additional hypothesis described previously in this chapter, I anticipated that overtime, politicians pay lower costs for being exposed as collaborators of the former au-thoritarian regime. As a result, they should be less prone to blackmail, and the qualityof representation should increase. However, the coefficient associated with years sincetransition is not statistically significant, so I find little evidence to support the claimthat “years since the transition took place” offers a good proxy for the costs of havingskeletons in the closet revealed. I attribute this lack of support to the coarse measureof costs of firing that I use in my analysis.

The last two models also include the variable press freedom, which operationalizesparameter c, the cost of bluffing. I find that the variable is positively related to thequality of representation, although this association is only positive at the 90% level.I also find that it does not change the estimated effects of the rest of the theoreticallyrelevant regressors, which I take as evidence supporting my hypotheses. Interestingly,the constant term is no longer statistically significant when press freedom is included. Iinterpret this as a type of omitted variable bias: in models 1-3, the average effect of afree press and time since transition is encapsulated by this constant.

The models reported here are static, meaning that they take a snapshot of a coun-try and thus cannot produce evidence of changing quality of representation over time.Nevertheless, they have many redeeming features. First, DALP extends to over 80democracies and exhibits almost perfect overlap with the set of countries included

8 Specifically, these data are on whether the authoritarian regime had no legislature,a single party-legislature, or a multiparty-legislature.


in the Global Transitional Justice Dataset. No competing datasets (CHES or the Man-ifesto Project) allow us to measure quality of representation for even half of the coun-tries in the Global Transitional Justice Dataset.

To ensure that the results presented in this chapter are not an artifact of the staticnature of the statistical model, in the concluding chapter to this book, I present a modelthat uses all the distinct transitional justice mechanisms to predict another dependentvariables—power distributed by socio-economic status that is measured over time.In the conclusion, lustration appears as the strongest predictor of higher quality ofdemocracy.

5.2 Conclusion

In this chapter, I investigated how lustration, understood as the policies that revealinformation about prior collaboration with secret authoritarian elites, influences pro-grammatic representation of political parties. The evidence presented here corrobo-rates the theoretical predictions from the formal model from Chapter 2, according towhich officers of the former authoritarian regime use secret police files to blackmailnew democratic elites. The mechanism presented in Chapter 2 outlined how evidenceof past collaboration can be used to force democratic elites to implement policies thatare amenable to former authoritarian elites and not to the voters. Based on the the-ory I developed, I predicted that former authoritarian elites’ ability to shape policyis greater when transitional justice is legislated less frequently, and that these elites’influence decreases with how much voters care about authoritarian skeletons in theirpoliticians’ closets. I also made predictions about the effects of the ideological distancebetween former autocrats and former dissidents: as this distance increases, so does theability of former autocrats to extract concessions via blackmail. Data to test these pre-dictions came from the Global Transitional Justice Dataset, originally constructed forthis purpose, and from the Democratic Accountability and Linkages Project.

The findings of this chapter by and large corroborate my theoretical predictions,especially confirming the prediction about the effects of lustration severity on pro-grammatic representation. Lustration decreases the amount of potentially embarrass-ing information that remains hidden and useful for blackmail and so prevents black-mail of current politicians by former authoritarian elites.

Yet, lustration does not always prevent blackmail: I also find that holding all elseconstant, a lack of press freedom can sabotage transitional justice efforts. Allegationsof collaboration need to be investigated and verified; in a liberal democracy, this taskis often performed by a free press. In the absence of independent media, blackmailcan occur through bluffing: that is, threatening to accuse current politicians of havingties with the former authoritarian regime even when no actual evidence exists.

This chapter concludes the contribution by this book to theorizing the effects oftransitional justice on party programmaticness.

6

Truth Commissions and the Quality of Democracy

The previous chapter tested the implications of the model from Chapter 2, but the onlymechanism that was tested was the effect of lustration on how politicians behave inoffice with a special focus on political parties’ ability to deliver policy platforms to vot-ers. This was sensible in light of the fact that lustration targets politicians. Politicians,in turn, engage in corrupt or programmatic exchanges with voters through party orga-nizations. A transparency regime, arguably with much broader scope than lustrationis a truth commissions, which can effect any member of societyZvobgo (2019b).

7

Purges and the Quality of Democracy

8

Conclusion: Beyond Ritual Sacrifices

116 8 Conclusion: Beyond Ritual Sacrifices

test moretest

9

Appendices

118 9 Appendices

9.1 Proofs to Chapter 2

In this appendix, I present the formal analysis to derive the results from the model inchapter 2.

9.1.1 Players, Preferences, strategies and Beliefs

There are two players in this game: an Officer (O) who has complete informationand a Politician (P ) who has incomplete information. The player with incompleteinformation comes in two types: those with evidence of skeletons in the closet (withprobability (1− π) and those without (with π). Recall that the players’ payoffs are alinear function of the Euclidean distance between the players’ respective ideal pointsand implemented policy as well as the two types of costs characterized above: (1) thecost to the politician of being fired as a result of revealed skeletons in his closet; and(2) the cost of bluffing incurred by the officer if he makes an empty threat.

The strategy set of the Politician can be defined as SP = {x(d) : D → [0, a]}. Inother words, a strategy for the Politician is a proposal x ∈ [0, a] that P makes in theevent that O makes a policy demand. Note that O can simply ignore the demand ifhe choses x = 0. 1 A strategy for the officer can be written as σO : T → D × {r(x)},where

T = {Evidence,¬Evidence}, D = {demand,¬demand} and r(x) ∈ {Reject,¬Reject}.r(x) denotes the action taken by O following P ’s counteroffer x. This part of O’s

strategy can best be represented as a rejection region R = {x : x ∈ R→ O reveals}.This structure resembles that of a signaling game in which O is the sender of the

message while P is the receiver. In line with these types of games, the politician mayhave a chance to update his prior belief about the type of Officer he is facing—that is,about whether it is an Officer equipped with evidence of collaboration or not—afterobserving the Officer’s action. Depending on the relative magnitudes of a, c andF , thismight be a signaling game or a cheap talk game. The solution concept we use is thatof Perfect Bayesian Equilibrium (PBE). This means that strategies are best responsesto each other, that I have to specify the Politician’s beliefs, and that those beliefs mustbe consistent with what is observed in equilibrium play.

Below is a formal Proposition that summarizes the results from the main text:

9.1.2 Proposition

Proposition 9.1. 1. Suppose F < c. There is a pure separating Perfect Bayesian Equilib-rium in which the Officer makes a demand if and only if evidence exists and in response, thePolitician makes a counteroffer x∗ ≡ F . The Officer’s rejection region in this equilibriumis defined by R∗ ≡ (0, F ), so he accepts the counteroffer and does not reveal the evi-dence in his possession. The posterior beliefs of P are described by Pr(E|no demand) =0, P r(¬E|no demand) = 1, P r(E|demand) = 1, P r(¬E|no demand) = 0.

2. Suppose c ≤ F < c(1−π) . There is a pure pooling Perfect Bayesian Equilibrium in

which the Officer always makes a demand, and the Politician responds with a coun-teroffer x′′ ≡ (1 − π)F . The officer’s rejection region in this equilibrium is defined

1 SinceP does not observe whether evidence exists or not, but only observes whethera demand was placed or not, we only need a single action to describe his strategy.

9.1 Proofs to Chapter 2 119

by R′′≡ (0, (1 − π)F ); thus, he accepts the counteroffer and does not reveal the evi-

dence is his possession. The posterior beliefs of P are described by Pr(E|demand) =(1− π), P r(¬E|demand) = π.

3. Finally, suppose c ≤ F < c(1−π) . There is a semi-separating Perfect Bayesian Equilibrium

in which the Officer always places a demand when evidence exists but if evidence doesnot exist, he places a demand with probability λ∗ and refrains from placing a demandwith probability 1 − λ∗. In response to the demand, the Politician makes a counterofferx′≡ c. Since the Officer’s rejection region is defined byR′ ≡ (0, c), the Officer accepts the

counteroffer and does not reveal the evidence in his possession. The posterior beliefs ofP aredescribed by Pr(E|no demand) = 0, P r(¬E|no demand) = λ∗, P r(E|demand) =1, P r(¬E|demand) = 1− λ∗, where λ∗ ≡ (1−π)(F−c)

(π)c.

9.1.3 Proof

I first prove the part of proposition 9.1. pertaining to a pure separating equilibrium inwhich the officer equipped with evidence always places a demand, while the Officernot equipped with such evidence never places a demand.

9.1.4 Pure separating equilibrium

Since it is never rational forOnot to make a demand when he has evidence against thepolitician,2 the only possible separating equilibrium is: (demand,R∗,no demand, x∗).The requirement for this to be a Bayesian equilibrium is that the posterior beliefsare Pr(E|no demand) = 0, P r(∼ E|no demand) = 1, P r(E|demand) = 1, P r(∼E|demand) = 0. The following four steps uncover the conditions for this equilibrium.

1. First, suppose x∗ is the proposal accepted in equilibrium. P knows that if he pro-poses x < x∗, evidence against him will be revealed with certainty. Therefore, onthe one hand, if he makes a proposal, in equilibrium, it must be accepted becauseif it were to be rejected,P should propose x = 0. On the other hand, if a demand isplaced and P makes a counteroffer x 6= a, P ’s utility must be greater from havinghis proposal accepted than from proposing his ideal point, i.e.,EUP (x|demand,no demand) ≥ EUP (d|demand,no demand), which is equiva-lent to −x ≥ −F . The last expression simplifies to

x ≤ F

2. Second, note that the lowest x∗ that the politician should be willing to issue isx∗ = F . Therefore, O’s optimal rejection region is R = (0, F ).

3. Third, note that to prevent O from bluffing (and making demands when there isno evidence), it has to be the case that that EUO(0|x∗,∼ E) ≤ EUO(1|x∗,∼ E),which is equivalent to −|a− x∗| − c ≤ −| − a|. This last expression simplifies tox∗ ≤ c, which after substituting x∗ from above gives:

F ≤ c2 Note that P cannot make a counterproposal x when 0 has not made a demand.

Thus, the worstO can do when making a demand is−a, which would be his payoffif P ’s counterproposal were x = 0—that is, if P made no concession at all. Withoutmaking any demand, O is guaranteed to receive −a and no more.

120 9 Appendices

4. Fourth, for x∗ to be a feasible proposal, it has to be the case that x∗ ≥ 0. ( that is,F ≥ 0); otherwise, P should just propose 0. Given the assumption that F > 0,this always holds.

I conclude that a pure separating equilibrium exists only when the cost of beingfired (F ) relative to the cost of bluffing c is quite low. In this pure separating equilib-rium, blackmail is effective with probability (1−π), and the level of misrepresentation(F ) is directly proportional to the cost of firing. In the next two sections, I show howthis departure from programmatic representation compares with the effectiveness ofblackmail under the pure pooling and semi-separating equilibria.

9.1.5 Pure pooling equilibrium

In pooling equilibria, agents with and without evidence will choose the same action,implying that the Politician cannot update his prior beliefs to posterior beliefs by con-ditioning on the agent’s action. In the analysis above, I established that the agent neverrefuses to place a demand when evidence is present. Thus, the only possibility of apooling equilibrium in this game is (demand,R, demand;x′′) with accompanying be-liefs: Pr(E|demand) = (1 − π), P r(∼ E|no demand) = π.3 In this equilibrium, theagent always places a demand and the dissident always offers the same counterpro-posal, x′′.

1. For such an equilibrium to hold, the politician has to prefer having his proposalaccepted to having it rejected, which may lead to his getting fired (if the pro-posal were to be rejected, he’d rather propose his ideal point, 0). Thus, it must bethe case that EUP (x′′|demand, demand) ≥ EUP (0|demand, demand), which isequivalent to−|−x| ≥ (1−π)(−|0−0|−F )+π∗0.4 The last expression simplifiesto :

x ≥ πF2. Since x′′ = (1− π)F is the lowest counterproposal the Politician will issue, given

his beliefs, the Officer’s optimal rejection region is (0, (1− π)F ].3. To ensure that the Officer always has an incentive to place a demand, it has to

be the case that EUO(demand|x′′,∼ E) ≥ EUA(no demand|x′′,∼ E), which isequivalent to −|a − x′′| − c ≥ −| − a|. This last expression simplifies to x′′ ≥ c,which after substituting x′′ from above gives

F ≥ c

π

4. Finally, as before, to be feasible, x′′ has to lie between 0 and a, i.e., (0 < πF < a).The first part of the inequality is ensured by my assumption 0 < F , and the secondis ensured by F < a. .

3 Note that paths that involve the officer not making a demand are off the equilibriumpath and we do not have to specify the beliefs there. We can assume that if thePolitician observes no demand, he assumes he is dealing with an officer with noevidence.

4 Note that since the Politician cannot tell which type of Officer—with or withoutevidence—he is facing any better than he could before the agent took an action,his expected utility from making a proposal outside of the acceptance region isweighted by his priors about the probability that evidence exists.


Summing up, a pure pooling equilibrium exists only when the cost of being fired (F )relative to the cost of bluffing c is quite high. In this pure pooling equilibrium, lus-tration blackmail is always effective (takes place with probability 1). The distortionit causes relative to the dissident’s ideal point is (1 − π)F . It is directly proportionalto the cost of firing and the extent to which evidence exists. The final subsection ofmy equilibrium analysis looks at the effectiveness of lustration blackmail under thesemi-separating equilibrium.

Semi-separating equilibrium

In addition to the pure separating equilibrium discussed above, I also derive condi-tions of a semi-separating equilibrium (and verify their plausibility). In this equilib-rium, the Officer plays a mixed strategy. He always makes a demand when evidenceis present, but he also, with some probability λ, makes a demand if evidence doesnot exist (and with probability 1 − λ does not make a demand). Consequently, anysemi-separating equilibrium must fit the format (demand,R, λ;x′).5

Note that the beliefs consistent with this semi-pooling equilibrium are:Pr(E|no demand) =0), P r(∼ E|demand) = λ∗π

λπ+(1−π) , P r(E|demand) =(1−π)

λπ+(1−π) , P r(∼ E|no demand) =1. These beliefs will be used in the calculation of the expected utilities. I proceed in sixsteps.

1. First, to find the equilibrium value of λ∗, I calculate the expected utility of thePolitician from responding with x to the agent’s demand (EUP (x′|0, λ)) and setit equal to the expected utility of the Politician’s choosing an xwhich is outside ofthe agent’s acceptance region (EUP (d|0, λ) This yields the equality (1− π)(−| −x′|) + πλ(−| − x′|) + 0 = (1− π)(−|0− 0| − F ) + πλ(−|0− 0|+ 0− (1− π)F ).

2. Next, assuming that x′ is the equilibrium proposal that falls into O’s acceptanceregion, I require that (1 − π)(−| − x′|) + πλ(−| − x′|) = −(1 − π)F , which interms of λ can be stated as:

λ =(1− π)(−x′)− (1− π)F

π[−(−x′)] (9.1)

3. To ensure that 0 < λ < 1 and that λ is a probability F must satisfy:

x′ < F <x′

(1− π) (9.2)

4. Next, to pin down x′, we make use of the fact that when evidence does notexist, O must be indifferent between placing a demand and not placing one,i.e.: EUO(demand|x′,∼ E) = EUO(no demand|x′,∼ E), which reduces to:−| − a| = | − a− x′| − c. The last equality can be written in terms of x′ as x′ = c.

5. Finally, substituting x′ into equation 9.1:

λ =(1− π)(F − c)

πc(9.3)

5 This means that the agent makes a demand with probability 1 if evidence existsand with probability λ if evidence does not exist.

122 9 Appendices

6. To get scope conditions for the semi-separating equilibrium, we can substitute x′into condition 9.2:

c < F < c(1− π)And to ensure that the equilibrium proposal is between a and 0, I need c > a = 0.

I will also prove a result about the expected misrepresentation under all threeequilibria. It is described again by the proposition below:

Proposition 9.2. The PBE outcome, interpreted as the expected level of misrepresentation, isgiven by the same formula across all three equilibria: (1− π)F .

Proof. In order to derive the expected level of misrepresentation, I weight the PBEoutcome in each equilibrium by the frequency of its occurrence. In the case of the pureseparating equilibrium, it is simply (1−π)(F )+π0, as the officer only proposes x∗ = Fwhen evidence exists, which is π of the time. The remaining π of the time, he revertsto 0. In the case of the pure pooling equilibrium, the average policy is implemented at

1 ∗ ((1− π)F ), as the officer always places a demand and the politician always re-sponds with x” = (1−π)F . In the case of the semi-separating equilibrium, the calcula-tion of the policy implemented is somewhat more complex, because the Officer alwaysplaces a demand when evidence exists (with probability1−π), and also does so λ∗ ofthe time when it does not exist. Hence, the total frequency of placing a demand is givenby (1−π)+π (1−π)(F−c)

πc. The Politician responds to this demand with x′, bringing the

expected policy outcome to (c)((1 − π) + π (1−π)(F−c)πc

+ π(1 − (1−π)(F−c)πc

)0, which,as in the previous two cases, reduces to (1−π)F . These expected policy outcomes arepresented in Figure 2.3 and described below in Proposition 2.1, which follows directlyfrom the derivation above.

9.2 Proofs to Chapter 3

9.2.1 Baseline model

Let S ⊂ < be the uni-dimensional policy space. Players are P andA, with ideal points0εS and xAε(0, 1) ⊂ S, respectively. The strategy set of P , who moves first is given bySP = {P,∼ P}. Following the move of the Politician, Nature introduces a shock thatrepresents uncertainty: ωε{ε,−ε}. I assume that either policy shock is equally likely:that is, Pr(ω = −ε) = Pr(ω = ε) = 1

2.A observes the shock and chooses his strategy,

which associates with each type of shock a policy p. Hence, SA = {sA : {ε,−ε} → S,where an example of a strategy is a pair (pε, p−ε).

The strategy set of the Politician is given by SP = {Purge,∼ Purge} × S.Since the final outcome is modeled as x = p+ ω, the utility functions are written

as a function of the policy choice, p, as follows:UP (p) = −|p+ ω|UA(p) = −|xA − (p+ ω)|.This game is solved by backward induction as follows. Assume that in the first

stage, the Politician chose “Purge.” Since the Agency Officer observes ω, he can im-plement a policy to “perfectly absorb” the exogenous policy shock and bring the finaloutcome to his ideal point. That is, for ω = ε, he will choose BRA(ε) = p∗ = xa − ε


and forω = −ε, he will chooseBRA(−ε) = p∗ = xA+ε. In light of this, the Politician’sexpected outcome from a leadership purge is given by:

UP (Leadership,BRA(Leadership)) = −xA.

Assume now that the Politician chose “∼ Purge” in the first stage. Since he cannotobserve the policy shock, the Politician’s best response, given the symmetry ofω, is p =0. (Note that actually, any pε[−ε, ε] is a best response; however, p = 0 is robust withrespect to changing the utility functions from “tent” to quadratic. Gehlbach (2013) andBendor & Meirowitz (2004) show that changing the players’ preferences in this waydoes not change the end result of the analysis.) As a result, EUP (Thorough, 0) = −ε.

9.2.2 Model with discretion

The players and preferences in this version of the model are just as before. Therefore,I use the following game tree to summarize it:

Fig. 9.1: Purges Model with Discretion Limits

P

Nature

ωε{−ε, ε}

P

pεS

−|p+ ω|,−|xA − (p+ ω)|

T

Nature

ωε{−ε, ε}

A

pε(−∞, r]

−|p+ ω|,−|xA − (p+ ω)|

rε<+

As reflected in the game tree, the only modification of this model relative to theprevious one is that the strategy set of the Politician is now SP = {P} ∪ <+) × S toreflect that if the Politician chooses to purge, he also sets the discretion limit.

124 9 Appendices

The game is solved for the optimal discretion limit by examining potential limitsthat appear to be focal. I start here with limit r = ε.

1. If ω = −ε, A’s utility is maximized at p∗ = xA + ε because xA > 0 ensures thatxA + ε > ε. Given this, whenever ω = −ε and r = ε, A will propose p = ε. Thiswill result in A’s expected utility of −|xA − ε|.

2. If ω = ε, A’s utility is maximized at p∗ = xA − ε. A will therefore propose p =xA − ε whenever xA − ε < ε or, in other words, when xA < 2ε. When xA > 2ε,he will propose p = 0. This will result in an expected utility of 0 when xA < 2ε,and of xA − ε when xA > 2ε.

3. The corresponding utilities for P will be 0 with probability 12, −xA with proba-

bility 12

when xA < 2ε, and 12(−xA)+ 0 ∗ 1

2when xA > 2ε. Thus, if−xA

2> −ε or

xA < 2ε, P will set the discretion limit to r = ε and select a leadership purge.

Now, suppose (1) that r is set higher than ε and (2) that it is set lower than ε.

1. If the discretion limit is set higher, to some r = ε+δ,Awould have to set policy tomin{ε+δ, xA−ε} for ω = −ε and tomin{ε+δ, xA−ε} for ω = ε. Consequently,P ’s expected utility would be given by:EUP (r = ε+ δ) = − 1

2|ε+ δ− ε|− 1

2|xA−

ε+ ε| = −xA+δ2

.2. If the discretion limit is set lower, to some r = ε−δ,Awould propose p = min{ε−δ, xA+ε} forω = −ε and p = min{ε−δ, xA−ε} forω = ε, leading toP ’s expectedutility of:EUP (r = ε− δ) = − 1

2|ε− δ − ε| − 1

2|xA − ε+ ε| = − δ

2− xA

2.

In both instances, the expected utility of the Politician is lower than with a discre-tionary limit set at ε6

9.2.3 Model with imperfect expertise

This game is analogous to the baseline, save for the institutionalization parameter thatcharacterizes the signal technology received by the Agency Officer. This parameter isdefined as:

s =Pr(−ε|ε)Pr(ε|ε) =

Pr(ε| − ε)Pr(−ε| − ε) (9.4)

, where

Pr(s = i|ω = j is the probability that A receives a signal i when the signal is i ∀i, jε{−ε, ε}.(9.5)

.Formally, the strategy set of the Politician is given by SP = {sP : (0, 1) →

{Purge,∼ Purge}}. That is, the Politician’s strategy must assign to every value ofsε(0, 1) one of two actions: purging or refraining from purging. The strategy set of theAgency Officer is given bySA = {sA : (0, 1)→ {(p(ε), q(−ε)) : p is response to high signal and q is response to low signal}}.6 For an alternative proof, see Gehlbach (2013)’s textbook presentation. I am grateful

to Eli Rau for pointing this out and for making comments on my proof. I note thatone additional assumption needed to complete the second step of this proof—thatincreasing the limit relative to ε only decreases P ’s expected utility—is that δ ≥xA − 2ε. Without this assumption, the Politician’s expected utility simply does notchange relative to r = ε.


In order to solve the game for subgame perfect equilibrium, I proceed as follows:First, assume that the Politician refrains from a purge. Then, in the final stage,

given the Agent’s limited shock absorption, the Agent will choose p to maximize:EUA(s, p, q; .) = −(xA − (p+ ω))2 ∗ P (ε|ε)− (xA − (p− ω)2 ∗ P (ε| − ε)and simultaneously choose q to maximize:EUA(s, p, q; .) = −(xA − (q + ω))2 ∗ P (−ε|ε)− (xA − (q − ω)2 ∗ P (−ε| − ε).Moreover,

P (ε| − ε)P (ε|ε) = s =⇒ P (ε| − ε) = sP (ε|ε) (9.6)

and

P (−ε|ε))P (−ε| − ε)) = s =⇒ P (l(ε)) = sP (−ε| − ε)) (9.7)

Using condition (2) from Section 3.2.3, we thus have:sP (ε|−ε)+P (ε|ε) = 1 leading toP (ε|ε) = 1

(s+1), and sP (−ε|ε)+P (−ε|−ε)) = 1,

leading to P (−ε| − ε) = 1(s+1)

.These transformations allow us to write A’s expected utility following a high sig-

nal as:EUA(H) = −(xA − (p+ ε))2 ∗ 1

(s+1)− (xA − (p− ε))2 ∗ s

(s+1)

argmaxpEUA|H can be found by solving the First Order Condition:

δEUA|Hδp

= 0

, equivalent to:

δEUA|Hδp

=xAs+ 1

+sxAs+ 1

− (p+ ε)

(s+ 1)− s(p− ε)

(s+ 1)= 0

After solving for p, we arrive at:

p∗ = xA −ε(1− s)1 + s

(9.8)

Similarly,

EUA|L = −(xA − (q + ε))2 ∗ s

(s+ 1)− (xA − (q − ε))2 ∗ 1

(s+ 1)

And the optimal choice of q in response to the low signal is the q that solves:

δEUA|Lδq

= 0 (9.9)

EUA|L = −(xA − (q + ε))2 ∗ s(s+1)

− (xA − (q − ε))2 ∗ 1(s+1)

Since δEUA|Lδq

= 0 if and only if xAss+1

+ xAs+1− s(q+ε)

s+1− (q−ε)

(s+1)= 0, this q is given by:

q∗ = xA +ε(1− s)(1 + s)

(9.10)

126 9 Appendices

9.3 Quantitative Appendix to chapter 5

9.3.1 Quality of Representation (cosal 3)

I begin this appendix by explaining the creation of the dependent variable, cosal 3, forthe analysis conducted in Chapter 5. Recall that it draws on Herbert Kitschelt et. al.’smeasure of programmaticness from the Democracy and Accountability and LinkagesProject (DALP). Of the three components used to produce Kitschelt & Freeze (2010)’smeasure, I use only two, cohesion and salience, skipping polarization. The reason foromitting polarization is that it is potentially correlated with distance between par-ties’ ideal points, the independent variable on the right hand side of the regression inChapter 5.

My new variable programmaticness, was constructed using Kitschelt & Freeze(2010)’s raw data recording experts’ responses to questions characterizing the featuresand behavior of 535 parties in my sample. Specifically, following Kitschelt & Freeze(2010), I rely for the dimensions of programmaticness on the following questions fromthe expert survey:

d1: Party policy position on social spending, measured on a 1-10 scaled2: Party policy position on the state role in economy, measured on a 1-10 scaled3: Party policy position on public spending, measured on a 1-10 scaled4: Party policy position onnational identity, measured on a 1-10 scaled5: Party policy position on traditional authority, institutions, and customs, mea-sured on a 1-10 scale

The different cosal measures are composed of cohesion and salience based on dif-ferent combinations of the variables above.

Following Kitschelt et. al., I measure cohesion for each of the five issues aboveby calculating the standard deviation for the responses on that issue. The standarddeviation of those issues that received less than 5 expert responses and had a higherscore than 3.5 are capped at 3.5 (to avoid outliers resulting from a low response rate).Next, the scores were transformed to translate higher standard deviations into lowervalues for cohesion. Finally, I normalized the scores to range from 0 to 1. The saliencefor issue i of party p in country k is simply the proportion of experts that gave a validanswer to the question on issue i. The specific measure cosal 3 used here is created bytaking the average of three values. The first two values are cosald4 and cosald5, and thethird is the maximum value of cosald1, cosald2, and cosald3.

There are, of course, many different ways to create the measure of programmatic-ness. I chose this one for its affinity with the original (Kitschelt & Freeze 2010). Table9.1 presents summary statistics of all variables used in the cross-sectional models—that is, all non-time varying variables.

9.3.2 Robustness checks

In this subsection, I present a series of robustness checks of the analysis in Chapter 5.The main result showing a positive effect of lustration on programmatic representa-tion is robust to these different specifications and perturbations to the statistical model.

9.3 Quantitative Appendix to chapter 5 127

Table 9.1: Summary statistics for country-level variables

Min. 1st Qu. Median Mean 3rd Qu. Max. Nseverity 0.00 0.00 0.00 0.30 0.60 0.82 59n year 0.02 0.35 0.53 0.50 0.57 1.00 59

press freedom 0.54 0.72 0.79 0.78 0.84 1.00 59opp status 0.00 0.00 1.00 0.54 1.00 1.00 59no succ 0.00 0.00 0.00 0.24 0.00 1.00 59

Selection into treatment

First, suppose that more developed countries have both more positive lustrationevents and a more stable democratic party system. There is no within-country tem-poral variation that can be used as evidence of causality (I can use my original datato capture variation of severity across time, but quality of representation is observedonly once). To make up for this, I propose the following plausibility probe: regressseverity in country k on gdp per capita at the time of its transition (tyear), on the yearslapsed since the transition (years since transition), and on the competitiveness of the in-augural elections measured as the margin of victory. I additionally include a dummyindicating whether country k held such inaugural elections at all. This model takesthe form:

severityk =b0 + b1 ∗ log(gdp per capitak,tyear) + b2 ∗ years since transitionk+

b4 ∗margink,eyear + b5 ∗miss elek + ek

In the following step, I use the results from the above model to estimate the sever-ity of lustration in each country k, denoting the predicted outcome variable predictedseverity. Predicted severity captures only economic and political determinants, in addi-tion to how long country k has had an opportunity to engage in lustration. This vari-able is then used to estimate the same models reported in Table 5.1 in Chapter 5, butusing predicted severity instead of the original dependent variable. Table 9.2 indicatesthat the main result is robust to this probe.

Including GDP per capita

Economic development and state capacity are factors that could explain both lustra-tion severity and the quality of representation. To correct for this potential source ofendogeneity in my models, I control for logged GDP per capita. The results are pro-vided in table 9.3.

Using only pre-2008 country-years

Another robustness check involves discarding all post-2008 country years from themeasure of lustration severity, as this is when DALP data was collected. This probeexamines the possibility that programmaticness explains lustration severity and notvice versa. The results are robust to this test. The results are provided in Table 9.4.

128 9 Appendices

Table 9.2: Quality of representation and predicted severity of lustration


(0.034) (0.034) (0.034) (0.034) (0.034)

predicted severity 0.587∗∗∗ 0.579∗∗∗ 0.518∗∗∗ 0.375∗∗∗(0.104) (0.119) (0.118) (0.112)

years since transition 0.016 −0.034 0.096(0.109) (0.107) (0.108)

press freedom 0.420∗∗ 0.357∗(0.191) (0.200)

opposition status −0.007(0.037)

Constant 0.379∗∗∗ 0.171∗∗∗ 0.166∗∗∗ −0.121 −0.046(0.032) (0.046) (0.055) (0.141) (0.152)

Country intercepts Y Y Y Y YObservations 313 307 307 307 307Log Likelihood 177.450 184.076 182.781 184.387 180.693Akaike Inf. Crit. −344.900 −356.151 −351.562 −352.774 −345.387Bayesian Inf. Crit. −326.169 −333.790 −325.474 −322.959 −315.572

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

Using log(years since transition)

I also explore the possibility of a non-linear effect of years since transition. I estimate allof my models using log(years since transition) instead of years since transition, and in-clude an interaction term between log(years since transition) and severity of lustration.The results are provided in Table 9.5.

Using original outcome variable cosalpo 4

In addition, I present the same set of models as in Table 5.1, but instead of usingcosal 3, my programmaticness measure constructed for the analysis in Chapter 5, I useDALP’s cosalpo 4, the measure that also includes polarization. Recall that in Chapter5 I dropped polarization from the measure because party distance was one of the co-variates implied by the model from Chapter 2. These results are in Table 9.6


Table 9.3: Quality of representation (controlling for log(GDP per capita))


(0.034) (0.034) (0.034) (0.034) (0.034)

log(gdp per capita) 0.136∗∗∗ 0.104∗∗∗ 0.109∗∗∗ 0.108∗∗∗ 0.108∗∗∗(0.025) (0.027) (0.027) (0.026) (0.026)

severity 0.161∗∗ 0.126∗ 0.082 0.082(0.065) (0.066) (0.068) (0.069)

years since transition 0.187∗∗ 0.139 0.142(0.094) (0.095) (0.102)

press freedom 0.366∗ 0.366∗(0.190) (0.192)


Constant −0.784∗∗∗ −0.569∗∗ −0.680∗∗∗ −0.928∗∗∗ −0.930∗∗∗(0.216) (0.224) (0.226) (0.255) (0.259)


Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

Conditions of the opposition

A potential criticism of the decision to measure the ideological distance between thePolitician and theblackmailer with the policy distance between a successor and a non-successor party is that the successor party may be a poor approximation of the lo-cation of the blackmailer who worked, after all, for the pre-transition authoritarianparty. Absent controls for the relationship between the pre-transition opposition andthe government, the empirical finding regarding party distance may be biased. One wayto account for this relationship is to use the successor vote share in the election imme-diately following the transition. A second alternative is to use the margin of victory, orcompetitiveness, in that election. Both variables are hence included in the robustnesscheck below. These results are in Table 9.7.

130 9 Appendices

Table 9.4: Quality of representation (countries before 2008)


(0.034) (0.035) (0.035) (0.035) (0.036)

severity (pre-2008) 0.261∗∗∗ 0.258∗∗∗ 0.188∗∗∗ 0.190∗∗∗(0.066) (0.066) (0.070) (0.071)


press freedom 0.573∗∗ 0.570∗∗(0.226) (0.228)


Constant 0.379∗∗∗ 0.309∗∗∗ 0.265∗∗∗ −0.164 −0.169(0.032) (0.038) (0.054) (0.173) (0.176)


Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

Subset of countries with no missing successors

Recall that in Chapter 5, not all countries had successor authoritarian parties. I usedan imputation method to assign a successor to each country, which may be problem-atic for a variety of reasons not picked up by the missingness dummy. Given this, Ialso estimated models from Table 5.1 using only countries that had a non-imputedsuccessor authoritarian party. These results are in Table 9.8.

Distance normalized by country

As a final robustness check, I replicate the results from Table 5.1 using a modifiedversion of the party distance variable. I normalize this variable so that the maximumdistance of any party in country k to the authoritarian successor party in country k is1, and the minimum distance of any party in country k to the authoritarian successorparty in country k is 0.


Table 9.5: Quality of representation (years since transition logged)

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6party distance −0.103∗∗∗ −0.107∗∗∗ −0.108∗∗∗ −0.110∗∗∗ −0.111∗∗∗ −0.112∗∗∗

(0.034) (0.034) (0.034) (0.034) (0.034) (0.034)

severity 0.278∗∗∗ 0.260∗∗∗ 0.214∗∗∗ 0.213∗∗∗ 0.109(0.062) (0.065) (0.070) (0.071) (0.152)

log(years since transition) 0.031 0.009 0.007 0.021(0.036) (0.038) (0.039) (0.043)

press freedom 0.410∗ 0.412∗ 0.452∗∗(0.222) (0.223) (0.230)

opposition status −0.006 −0.006(0.037) (0.037)

severity*log(years since transition) −0.152(0.198)

Constant 0.379∗∗∗ 0.278∗∗∗ 0.313∗∗∗ −0.016 −0.014 −0.032(0.032) (0.036) (0.054) (0.186) (0.188) (0.190)

Country intercepts Y Y Y Y YMissing successor Y Y Y Y YObservations 313 313 313 307 307 307Log Likelihood 177.450 184.327 182.293 178.196 175.841 175.436Akaike Inf. Crit. −344.900 −356.654 −350.586 −340.393 −333.682 −330.871Bayesian Inf. Crit. −326.169 −334.177 −324.362 −310.578 −300.140 −293.603

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

Table 9.9 presents the HLM regressions of programmaticness on lustration sever-ity and distance to successor authoritarian parties when all countries in the PersonnelTransitional Justice Dataset are included with imputed values for countries withoutsuccessors.

132 9 Appendices

Table 9.6: Quality of representation (measured as DALP’s cosalpo 4)

Model 1 Model 2 Model 3 Model 4severity 0.200∗∗∗ 0.186∗∗∗ 0.140∗∗∗ 0.144∗∗∗

(0.045) (0.045) (0.049) (0.049)

years since transition 0.125∗ 0.098 0.118(0.071) (0.072) (0.078)

press freedom 0.319∗∗ 0.313∗∗(0.155) (0.155)


Constant 0.182∗∗∗ 0.121∗∗∗ −0.101 −0.118(0.020) (0.040) (0.113) (0.116)

Country intercepts Y Y Y YObservations 313 313 307 307Log Likelihood 286.689 286.497 280.789 278.365Akaike Inf. Crit. −565.377 −562.995 −549.578 −542.730Bayesian Inf. Crit. −550.393 −544.264 −527.216 −516.642

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01


Table 9.7: Quality of representation (vote share of successor and margin ofvictory)

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6party distance −0.103∗∗∗ −0.107∗∗∗ −0.109∗∗∗ −0.111∗∗∗ −0.115∗∗∗ −0.111∗∗∗

(0.034) (0.034) (0.034) (0.034) (0.034) (0.034)

severity 0.278∗∗∗ 0.255∗∗∗ 0.208∗∗∗ 0.215∗∗∗ 0.206∗∗∗(0.062) (0.064) (0.069) (0.068) (0.069)

years since transition 0.150 0.108 0.093 0.104(0.103) (0.107) (0.105) (0.108)

press freedom 0.370∗ 0.348 0.367∗(0.215) (0.212) (0.216)

successor vote share −0.003∗(0.002)

margin of victory 0.001(0.002)

Constant 0.379∗∗∗ 0.278∗∗∗ 0.217∗∗∗ −0.039 0.046 −0.050(0.032) (0.036) (0.055) (0.158) (0.163) (0.160)

Country intercepts Y Y Y Y Y YMissing successor Y Y Y Y Y YObservations 313 313 313 307 307 307Log Likelihood 177.450 184.327 184.036 179.721 175.681 174.469Akaike Inf. Crit. −344.900 −356.654 −354.073 −343.442 −333.362 −330.939Bayesian Inf. Crit. −326.169 −334.177 −327.849 −313.627 −299.820 −297.397

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

134 9 Appendices

Table 9.8: Quality of representation, only countries with successor party

Model 1 Model 2 Model 3 Model 4 Model 5party distance −0.086∗ −0.090∗∗ −0.092∗∗ −0.094∗∗ −0.095∗∗

(0.044) (0.044) (0.044) (0.044) (0.044)

severity 0.200∗∗ 0.163∗ 0.160∗ 0.159∗(0.080) (0.090) (0.091) (0.093)


press freedom 0.254 0.263(0.312) (0.325)


Constant 0.374∗∗∗ 0.302∗∗∗ 0.247∗∗∗ 0.077 0.080(0.032) (0.042) (0.074) (0.222) (0.226)

Country intercepts Y Y Y Y YObservations 153 153 153 153 153Log Likelihood 83.746 85.065 84.581 84.667 82.794Akaike Inf. Crit. −159.492 −160.130 −157.162 −155.333 −149.588Bayesian Inf. Crit. −147.370 −144.977 −138.979 −134.120 −125.345

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01


Table 9.9: Quality of representation (party distance normalized by country)

d party normalized by country

Model 1 Model 2 Model 3 Model 4 Model 5party distance (normalized) −0.019 −0.019 −0.020 −0.020 −0.020

(0.023) (0.023) (0.023) (0.023) (0.023)

severity 0.199∗∗ 0.165∗ 0.162∗ 0.161∗(0.080) (0.090) (0.090) (0.092)


press freedom 0.189 0.192(0.311) (0.322)


Constant 0.354∗∗∗ 0.279∗∗∗ 0.227∗∗∗ 0.101 0.103(0.030) (0.041) (0.073) (0.219) (0.224)

Observations 152 152 152 152 152Log Likelihood 80.913 82.248 81.708 81.642 79.747Akaike Inf. Crit. −153.825 −154.496 −151.417 −149.283 −143.494Bayesian Inf. Crit. −141.730 −139.377 −133.273 −128.116 −119.303

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

136 9 Appendices

Table 9.10: Explaining quality of representation (country intercepts notshown)

Model 1 Model 2 Model 3 Model 4 Model 5distance party −0.071∗∗ −0.084∗∗ −0.081∗∗ −0.081∗∗ −0.082∗∗

(0.036) (0.035) (0.036) (0.036) (0.036)

severity 0.257∗∗∗ 0.195∗∗∗ 0.286 0.200∗∗∗(0.061) (0.066) (0.230) (0.068)

n year 0.096 0.120 0.116(0.097) (0.114) (0.109)

press freedom 0.350∗ 0.369∗ 0.340(0.212) (0.218) (0.215)

n year ∗ severity −0.176(0.427)

opp status 0.019(0.044)

Constant 0.378∗∗∗ 0.306∗∗∗ 0.002 −0.024 −0.012(0.027) (0.030) (0.156) (0.169) (0.161)

Observations 307 307 307 307 307Log Likelihood 174.472 180.504 180.680 180.830 178.570Akaike Inf. Crit. −340.944 −351.008 −347.359 −345.661 −341.140Bayesian Inf. Crit. −326.037 −332.374 −321.272 −315.846 −311.325

Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01

9.4 GIS appendix 137

9.4 GIS appendix

The following appendix presents a GIS coded version of our dataset with all threemeasures applied to each of the transitional justice mechanisms. The first map presentsGIS coded data on the severity of lustration in countries making up our universe ofcase. The second map presents the severity of leadership purges for the same group ofcountries. The third map presents the severity of thorough purges and the fourth mappresents the severity of truth commissions. This order is repeated for polarization andurgency/delay.

Fig. 9.2: World Map of Lustration Severity (Darker represents higher severity)

138 9 Appendices

Fig. 9.3: World Map of Leadership Purge Severity (Darker represents higherseverity)

Fig. 9.4: World Map of Thorough Purge Severity (Darker represents higherseverity)


Fig. 9.5: World Map of Truth Commission Severity (Darker represents higherseverity)

Fig. 9.6: World Map of Lustration Polarization (Darker represents higher po-larization)

140 9 Appendices

Fig. 9.7: World Map of Leadership Purge Polarization (Darker representshigher polarization)

Fig. 9.8: World Map of Thorough Purge Polarization (Darker representshigher polarization)


Fig. 9.9: World Map of Truth Commission Polarization (Darker representshigher polarization)

Fig. 9.10: World Map of Lustration Urgency (Darker represents higher delay)

142 9 Appendices

Fig. 9.11: World Map of Leadership Purge Urgency (Darker represents higherdelay)

Fig. 9.12: World Map of Thorough Purge Urgency (Darker represents higherdelay)


Fig. 9.13: World Map of Truth Commission Urgency (Darker representshigher delay)

References Cited

(1991).Albertus, M. & Menaldo, V. (2014), ‘Gaming democracy: elite dominance during

transition and the prospects for redistribution’, British Journal of Political Science44(3), 575–603.

Albertus, M. & Menaldo, V. (2018), Authoritarianism and the Elite Origins of Democracy,Cambridge University Press.

Alivizatos, N. C. & Diamandouros, P. N. (1997), Politics and the Judiciary in the GreekTransition to democracy.

Ang, M. (2019), Corrupting Accountability:Crime, Orphaned Bureaucracies, and the Failureof the Democratic State, Unpublished Manuscript.

Ang, M. & Nalepa, M. (2019), ‘Can transitional justice improve the quality of repre-sentation in new democracies?’, World Politics 71(4), 631–666.

Archive, f. t. (n.d.), ‘War’s end in el salvador. but peace needs time to grow’.Auerbach, A. M. & Thachil, T. (2018), ‘How clients select brokers: Competition and

choice in india’s slums’, American Political Science Review 112(4), 775–791.Baik, C. J. et al. (2011), The Park Chung Hee Era, Harvard University Press.Bakiner, O. (2016), Truth Commissions: Memory, Power, and Legitimacy, University of

Pennsylvania Press.Barahona de Brito, A., Gonzalez-Enrıquez, C. & Aguilar, P., eds (2001), The Politics of

Memory: Transitional Justice in Democratizing Societies, Oxford University Press.Bates, G., Cinar, I. & Nalepa, M. (2020), ‘Accountability by numbers: A new global

transitional justice dataset (1946–2016)’, Perspectives on Politics 18(1), 161–184.Bendor, J. & Meirowitz, A. (2004), ‘Spatial models of delegation’, American Political

Science Review 98(2), 293–310.Beres (n.d.), Genera Kiszczak mowi–prawie wszystko.Bernhard, M. & Kubik, J. (2014), Twenty Years After Communism: The Politics of Memory

and Commemoration, Oxford University Press.Binningsbø, H. M., Loyle, C. E., Gates, S. & Elster, J. (2012), ‘Armed conflict and post-

conflict justice, 1946–2006: A dataset’, Journal of Peace Research 49(5), 731–740.Blaydes, L. (2010), Elections and distributive politics in Mubarak’s Egypt, Cambridge Uni-

versity Press.Blaydes, L. (2018), State of Repression: Iraq under Saddam Hussein, Princeton University

Press.

146 References Cited

Boix, C. & Svolik, M. W. (2013), ‘The foundations of limited authoritarian government:Institutions, commitment, and power-sharing in dictatorships’, The Journal of Politics75(2), 300–316.

Borchert, J. (2006), Die Zusammenarbeit des Ministeriums fur Staatssicherheit (MfS) mitdem sowjetischen KGB in den 70er und 80er Jahren: ein Kapitel aus der Geschichte derSED-Herrschaft, LIT Verlag Munster.

Bouguerra, B. (2014), Reforming Tunisia’s Troubled Security Sector, Atlantic Council,Rafik Hariri Center for the Middle East.

Brehm, J. O. & Gates, S. (1999), Working, shirking, and sabotage: Bureaucratic response toa democratic public, University of Michigan Press.

Brun, D. A. & Diamond, L. (2014), Clientelism, social policy, and the quality of democracy,JHU Press.

Callander, S. et al. (2008), ‘A theory of policy expertise’, Quarterly Journal of PoliticalScience 3(2), 123–140.

Capoccia, G. (2015), Lustration and support for democracy: Evidence from post-wargermany, in ‘22nd International Conference of Europeanists’, Ces.

Carey, J. M. & Shugart, M. S. (1995), ‘Incentives to cultivate a personal vote: A rankordering of electoral formulas’, Electoral studies 14(4), 417–439.

Carroll, R. & Nalepa, M. (2014), Can parties represent after communism? the dis-sipation of the regime divide cleavage in post-communist party systems, in ‘21stInternational Conference of Europeanists’, CES.

Carroll, R. & Nalepa, M. (2020), ‘The personal vote and party cohesion: Modelingthe effects of electoral rules on intraparty politics’, Journal of Theoretical Politics 32(1), 36–69.

Cepl, V. (1992), ‘Ritual sacrifices’, E. Eur. Const. Rev. 1, 24.Chapman, A. R. (2009), ‘Truth finding in the transitional justice process’, Assessing the

impact of transitional justice: challenges for empirical research pp. 91–114.Cheibub, J. A., Gandhi, J. & Vreeland, J. R. (2010), ‘Democracy and dictatorship revis-

ited’, Public Choice 143(1-2), 67–101.Childs, D. & Popplewell, R. (2016), The Stasi: The East German Intelligence and Security

Service, Springer.Chodakiewicz, M. J. (2009), ‘Agent bolek’, Intelligencer: Journal of US Intelligence Studies

17(2), 108–110.Closa Montero, C. (2010), ‘Study on how the memory of crimes committed by total-

itarian regimes in europe is dealt with in the member states’, Madrid: Institute forPublic Goods and Policy .

Cohen, S. (1995), ‘State crimes of previous regimes: Knowledge, accountability, andthe policing of the past’, Law & Social Inquiry 20(1), 7–50.

Congreso de los Diputados, Comision de Cultura (2013), ‘161/002111’, Boletın Oficialde Las Cortes Generales D(360), 27–28.

Coppedge, M. (2012), Democratization and research methods, Cambridge UniversityPress.

Coppedge, M., Gerring, J., Lindberg, S. I., Skaaning, S.-E., Teorell, J., Ciobanu, V. I. &Saxer, L. (2017), ‘V-dem country coding units v7’.

Coppedge, M., Gerring, J., Lindberg, S., Skaaning, S. & Teorell, J. (2018), ‘v-dem code-book v8. varieties of democracy (vdem)’.

Dancy, G. & Montal, F. (2017), ‘Unintended positive complementarity: Why interna-tional criminal court investigations may increase domestic human rights prosecu-tions’, American Journal of International Law 111(3), 689–723.

References Cited 147

Dancy, G. & Wiebelhaus-Brahm, E. (2018), ‘The impact of criminal prosecutions dur-ing intrastate conflict’, Journal of Peace Research 55(1), 47–61.

Darden, K. A. (2001), ‘Blackmail as a tool of state domination: Ukraine under kuchma’,E. Eur. Const. Rev. 10, 67.

David, R. (2006), ‘From prague to baghdad: Lustration systems and their political ef-fects’, Government and Opposition 41(3), 347–372.

David, R. (2011), Lustration and transitional justice: Personnel systems in the Czech Repub-lic, Hungary, and Poland, University of Pennsylvania Press.

De Brito, A. B., Enrıquez, C. G. & Aguilar, P. (2001), The Politics of Memory and Democ-ratization: Transitional Justice in Democratizing Societies: Transitional Justice in Democ-ratizing Societies, Oxford University Press.

De Greiff, P. & Mayer-Rieckh, A. (2007), Justice as prevention: vetting public employ-ees in transitional societies, SSRC.

Debs, A. (2016), ‘Living by the sword and dying by the sword? leadership transitionsin and out of dictatorships’, International Studies Quarterly 60(1), 73–84.

Dragu, T. (2014), ‘On repression and its effectiveness’, Journal of Theoretical Politicsp. 0951629817710563.

Dragu, T. & Lupu, Y. (2018), ‘Collective action and constraints on repression at theendgame’, Comparative Political Studies 51(8), 1042–1073.

Dragu, T. & Polborn, M. (2013), ‘The administrative foundation of the rule of law’, TheJournal of Politics 75(4), 1038–1050.

Dragu, T. & Przeworski, A. (2019), ‘Preventive repression: Two types of moral hazard’,American Political Science Review 113(1), 77–87.

Dudek, A. & Gryz, R. (2003), Komunisci i Kosciol w Polsce: 1945-1989, Znak.Ellis, M. S. (1996), ‘Purging the past: The current state of lustration laws in the former

communist bloc’, Law and Contemporary Problems 59(4), 181–196.Elster, J. (1998), ‘Coming to terms with the past. a framework for the study of justice

in the transition to democracy’, European Journal of Sociology/Archives Europeennes deSociologie 39(1), 7–48.

Elster, J. (2004a), Closing the books: Transitional justice in historical perspective, CambridgeUniversity Press.

Elster, J. (2004b), Closing the books: Transitional justice in historical perspective, CambridgeUniversity Press.

Elster, J. (2005), ‘Preference formation in transitional justice’, Katznelson, Ira/Weingast,Barry R pp. 247–278.

Encarnacion, O. G. (2008), ‘Reconciliation after democratization: coping with the pastin spain’, Political Science Quarterly 123(3), 435–460.

Encarnacion, O. G. (2014), Democracy without justice in Spain: the politics of forgetting,University of Pennsylvania Press.

Epstein, D. & O’Halloran, S. (1999), Delegating powers: A transaction cost politics approachto policy making under separate powers, Cambridge University Press.

Escriba-Folch, A. & Wright, J. (2015), ‘Human rights prosecutions and autocratic sur-vival’, International Organization 69(02), 343–373.

Europe: A Rude Awakening; Spain’s Past (2007).URL: https://search-proquest-com.proxy.uchicago.edu/docview/224004849?accountid=14657

Europe: Painful memories; Spain’s civil war (2006).URL: https://search-proquest-com.proxy.uchicago.edu/docview/224004455?accountid=14657

Felli, L. & Hortala-Vallve, R. (2015), ‘Collusion, blackmail and whistle-blowing’, Quar-terly Journal of Political Science 11, 279–312.


Gandhi, J. & Lust-Okar, E. (2009), ‘Elections under authoritarianism’, Annual review ofpolitical science 12, 403–422.

Gandhi, J. & Przeworski, A. (2007), ‘Authoritarian institutions and the survival of au-tocrats’, Comparative Political Studies 40(11), 1279–1301.

Gasiorowski, M. J. (1995), ‘Economic crisis and political regime change: An event his-tory analysis’, American political science review 89(4), 882–897.

Geddes, B., Wright, J. & Frantz, E. (2014), ‘Autocratic breakdown and regime transi-tions: A new data set’, Perspectives on Politics 12(2), 313–331.

Gehlbach, S. (2013), Formal Models of Domestic Politics, Cambridge University Press.Gibson, J. L. (2006), ‘Overcoming apartheid: Can truth reconcile a divided nation?’,

The Annals of the American Academy of Political and Social Science 603(1), 82–110.Ginsburg, T. & Simpser, A. (2013), Constitutions in authoritarian regimes, Cambridge

University Press.Goertz, G. (2006), Social science concepts: A user’s guide, Princeton University Press.Gostkiewicz, M. (2013), ‘Grubej kreski nigdy nie bylo. co naprawde powiedzial ma-

zowiecki’.Grzymala-Busse, A. (2001), ‘Coalition formation and the regime divide in new democ-

racies: East central europe’, Comparative Politics pp. 85–104.Grzymala-Busse, A. (2017), ‘Hoist on their own petards? the reinvention and collapse

of authoritarian successor parties’, Party Politics p. 1354068817740336.Grzymala-Busse, A. M. (2002), Redeeming the communist past: The regeneration of com-

munist parties in East Central Europe, Cambridge University Press.Haggard, S. & Kaufman, R. R. (2016), Dictators and democrats: masses, elites, and regime

change, Princeton University Press.Halmai, G., Scheppele, K. L. & McAdams, A. J. (1997), Transitional Justice and the Rule

of Law in New Democracies, University of Notre Dame Press Notre Dame, IN.Hansen, L. J. & Sigman, R. (2020), ‘Leviathan’s latent dimensions’, American journal of

Political Science .Hayner, P. B. (2001), Unspeakable truths: Confronting state terror and atrocity, Psychology

Press.Henry, C. (2007), ‘Tunisia’s ‘sweet little’ regime’, Worst of the worst: dealing with repres-

sive and rogue nations pp. 300–323.Hicken, A. & Martınez Kuhonta, E. (2011), ‘Shadows from the past: Party system in-

stitutionalization in asia’, Comparative Political Studies 44(5), 572–597.Hoekstra, Q. (n.d.), ‘Helping the contras: The effectiveness of us support for foreign

rebels during the nicaraguan contra war (1979–1990)’, Studies in Conflict & Terrorism.

Holmes, S. (1994), ‘The end of decommunization’, E. Eur. Const. Rev. 3, 33.Horne, C. (2017a), ‘Vetting, purges, and lustration: Measurement choices and empir-

ical implications’.Horne, C. M. (2017b), Building Trust and Democracy in Transition: Assessing the Impact

of Transitional Justice in Post-Communist Countries, Oxford University Press.Horne, C. M. & Levi, M. (2004), Does lustration promote trustworthy governance? an

exploration of the experience of central and eastern europe, in ‘Building a trustwor-thy state in post-socialist transition’, Springer, pp. 52–74.

Huber, J. D. & McCarty, N. (2004), ‘Bureaucratic capacity, delegation, and politicalreform’, American Political Science Review 98(3), 481–494.

Huber, J. D. & Shipan, C. R. (2002), Deliberate discretion?: The institutional foundationsof bureaucratic autonomy, Cambridge University Press.


Hubert, R. & Little, A. T. (2020), ‘Kompromat’.Huntington, S. P. (1993), The third wave: Democratization in the late twentieth century,

Vol. 4, University of Oklahoma press.Jensen, N. M., Malesky, E. & Weymouth, S. (2014), ‘Unbundling the relationship be-

tween authoritarian legislatures and political risk’, British Journal of Political Science44(3), 655–684.

Jiang, J. & Yang, D. L. (2016), ‘Lying or believing? measuring preference falsificationfrom a political purge in china’, Comparative Political Studies 49(5), 600–634.

Kaminski, M. M. & Nalepa, M. (2006), ‘Judging transitional justice: A new criterion forevaluating truth revelation procedures’, Journal of Conflict Resolution 50(3), 383–408.

Kaminski, M. M. & Nalepa, M. (2014), ‘A model of strategic preemption: why do post-communists hurt themselves?’, Decyzje (21), 31–66.

Kaminski, M. M., Nalepa, M. & O’neill, B. (2006), ‘Normative and strategic aspects oftransitional justice’, Journal of Conflict Resolution 50(3), 295–302.

Keefer, P. (2007), ‘Clientelism, credibility, and the policy choices of young democra-cies’, American journal of political science 51(4), 804–821.

Keesing’s Record of World Events (1974), ‘Announcement of general election date. –resumption of party politics. – other internal developments.’.

Keesing’s Record of World Events (1998), ‘South africa’.King, G., Pan, J. & Roberts, M. E. (2013), ‘How censorship in china allows govern-

ment criticism but silences collective expression’, American Political Science Review107(2), 326–343.

Kitschelt, H. (1999), Post-communist party systems: competition, representation, and inter-party cooperation, Cambridge University Press.

Kitschelt, H. & Freeze, K. (2010), Programmatic party system structuration: De-veloping and comparing cross-national and cross-party measures with a newglobal data set, in ‘APSA Annual Meeting, Washington, DC’, Vol. Available athttps://goo.gl/xhxeyj.

Kitschelt, H. & Wilkinson, S. I. (2007), Patrons, clients and policies: Patterns of democraticaccountability and political competition, Cambridge University Press.

Koehler, J. O. (1999), STASI: The untold story of the East German secret police, Basic Books.Krauss, C. (2000), ‘New argentine president orders purge of ’dirty war’ remnants’.Kritz, N. J. (1995), Transitional justice: how emerging democracies reckon with former

regimes, Vol. 1, US Institute of Peace Press.Laplante, L. J. & Phenicie, K. (2009), ‘Mediating post-conflict dialogue: the media’s

role in transitional justice processes’, Available at https://goo.gl/tvDDs6.Leskiewicz, R. (2016), ‘Od sluzby bezpieczenstwa do urzedu ochrony panstwa’, Dzieje

Najnowsze 48(1), 165–188.Lessing, B. (2017), ‘Counterproductive punishment: How prison gangs undermine

state authority’, Rationality and Society 29(3), 257–297.Letki, N. (2002), ‘Lustration and democratisation in east-central europe’, Europe-Asia

Studies 54(4), 529–552.Levitsky, S. & Way, L. A. (2010), Competitive authoritarianism: Hybrid regimes after the

Cold War, Cambridge University Press.Linz, J. J., Stepan, A. & Valenzuela, A. (1978), The breakdown of democratic regimes, Vol. 1,

Johns Hopkins University Press Baltimore.Lundy, P. & McGovern, M. (2008), ‘Whose justice? rethinking transitional justice from

the bottom up’, Journal of Law and Society 35(2), 265–292.


Macdonald, G. & Waggoner, L. (2018), ‘Dashed hopes and extremism in tunisia’, Jour-nal of Democracy 29(1), 126–140.

Magaloni, B. (2006), Voting for autocracy: Hegemonic party survival and its demise in Mex-ico, Vol. 296, Cambridge University Press Cambridge.

Mainwaring, S. (1999), Rethinking party systems in the third wave of democratization: thecase of Brazil, Stanford University Press.

Mallinder, L. (2008), Amnesty, human rights and political transitions: bridging the peaceand justice divide, Bloomsbury Publishing.

Malloy, J. M. & Gamarra, E. (1988), Revolution and reaction: Bolivia, 1964-1985, Trans-action Books.

Masri, S. M. (2017), Tunisia: An Arab Anomaly, Columbia University Press.Mattingly, D. C. (2019), The Art of Political Control in China, Cambridge University

Press.Medina, L. F. & Stokes, S. (2007), ‘Monopoly and monitoring: an approach to political

clientelism’, Patrons, clients, and policies pp. 68–83.Meng, A. (2020), Constraining Dictatorship: From Personalized Rule to Institutionalized

Regimes, Cambridge University Press.Miller, J. (1998), ‘Settling accounts with a secret police: The german law on the stasi

records’, Europe-Asia Studies 50(2), 305–330.Montagnes, B. P. & Wolton, S. (2019), ‘Mass purges: top-down accountability in au-

tocracy’, American Political Science Review 113(4), 1045–1059.Muller, J. W. (2002), Memory and Power in Post-war Europe: Studies in the Presence of the

Past, Cambridge University Press.Murphy, C. (2017), The conceptual foundations of transitional justice, Cambridge Univer-

sity Press.Myerson, R. B. (2008), ‘The autocrat’s credibility problem and foundations of the con-

stitutional state’, American Political Science Review 102(1), 125–139.Nalepa, M. (2008), ‘To punish the guilty and protect the innocent comparing truth

revelation procedures’, Journal of Theoretical Politics 20(2), 221–245.Nalepa, M. (2010a), ‘Captured commitments: an analytic narrative of transitions with

transitional justice’, World Politics 62(2), 341–380.Nalepa, M. (2010b), Skeletons in the closet: Transitional justice in post-communist Europe,

Cambridge Studies in Comparative Politics, Cambridge University Press.Nalepa, M. (2012), ‘Tolerating mistakes: How do popular perceptions of procedu-

ral fairness affect demand for transitional justice?’, Journal of Conflict Resolution56(3), 490–515.

Nalepa, M. & Sonin, K. (2019), ‘How does kompromat affect politics? a model of trans-parency regimes’.

Nalepa, M. & Sonin, K. (2020), ‘How does kompromat affect politics? a model of trans-parency regimes’, A Model of Transparency Regimes (March 23, 2020). University ofChicago, Becker Friedman Institute for Economics Working Paper (2020-29).

Nalepa, M., Vanberg, G. & Chiopris, C. (2018), ‘Authoritarian backsliding’, Unpub-lished manuscript .

Nedelsky, N. (2013), ‘From velvet revolution to velvet justice: The case of slovakia’,After Oppression pp. 390–417.

Nugent, E. (2019), ‘The psychology of repression and polarization’, World Politics,Forthcoming .


Olsen, T. D., Payne, L. A. & Reiter, A. G. (2010), ‘The justice balance: When transitionaljustice improves human rights and democracy’, Human Rights Quarterly 32(4), 980–1007.

Osiatynski, W. (1992), ‘Agent walesa’, E. Eur. Const. Rev. 1, 28.Osiatynski, W. (1994), ‘Decommunization and recommunization in poland’, E. Eur.

Const. Rev. 3, 36.Otis, J. (1992), ‘Chamorro backers warn of ’new dictatorship’ in nicaragua’.

URL: http://www.upi.com/Archives/1992/06/05/Chamorro-backers-warn-of-new-dictatorship-in-Nicaragua/5442707716800/

Paine, J. (2019), ‘The loyalty-efficiency tradeoff in authoritarian repression’.Payne, S. G. (2011), The Franco Regime, 1936–1975, University of Wisconsin Pres.Pemstein, D., Marquardt, K. L., Tzelgov, E., Wang, Y.-t. & Miri, F. (2015), ‘The v-dem

measurement model: Latent variable analysis for cross-national and cross-temporalexpert-coded data’.

Perkins, K. (2014), A history of modern Tunisia, Cambridge University Press.Pettai, E.-C. & Pettai, V. (2014), Transitional and retrospective justice in the Baltic States,

Cambridge University Press.Pinto, A. C. (2001), ‘Settling accounts with the past in a troubled transition to democ-

racy: the portuguese case’, The politics of memory: transitional justice in democratizingsocieties pp. 65–91.

Pinto, A. C. (2008), ‘Political purges and state crisis in portugal’s transition to democ-racy, 1975-76’, Journal of Contemporary History 43(2), 305–332.

Pinto, A. C. (2010), ‘The authoritarian past and south european democracies: An in-troduction’, South European Society and Politics 15(3), 339–358.

Pop-Eleches, G. (2007), ‘Historical legacies and post-communist regime change’, Jour-nal of Politics 69(4), 908–926.URL: https://www.journals.uchicago.edu/doi/abs/10.1111/j.1468-2508.2007.00598.x

Pop-Eleches, G. (2010), ‘Throwing out the bums: Protest voting and unorthodox par-ties after communism’, World Politics 62(2), 221–260.

Pop-Eleches, G. & Tucker, J. A. (2011), ‘Communism’s shadow: postcommunist lega-cies, values, and behavior’, Comparative Politics 43(4), 379–408.

Pop-Eleches, G. & Tucker, J. A. (2012), ‘Associated with the past? communist legaciesand civic participation in post-communist countries’, East European Politics & Soci-eties p. 0888325412465087.

Posner, E. A. & Vermeule, A. (2004), ‘Transitional justice as ordinary justice’, HarvardLaw Review 117(3), 761–825.

Powell, J. M. (2014), ‘Trading coups for civil war: The strategic logic of tolerating re-bellion’, African Security Review 23(4), 329–338.

Preysing, D. (2016), Transitional Justice in post-revolutionary Tunisia (2011–2013): Howthe past shapes the future, Springer.

Price, R. M. (1991), The apartheid state in crisis: Political transformation in South Africa,1975-1990, Oxford University Press New York.

Przeworski, A. (1991), Democracy and the market: Political and economic reforms in EasternEurope and Latin America, Cambridge University Press.

Rauch, J. E. & Evans, P. B. (2000), ‘Bureaucratic structure and bureaucratic perfor-mance in less developed countries’, Journal of public economics 75(1), 49–71.

Remmer, K. L. (1990), ‘Democracy and economic crisis: the latin american experience’,World Politics 42(3), 315–335.

Renewed threats to peace process Purge of army officers (1993).


Rev, I. (2005), Retroactive Justice: Prehistory of Post-Communism, Stanford Universitypress.

Riedl, R. B. (2014), Authoritarian origins of democratic party systems in Africa, CambridgeUniversity Press.

Rosenfeld, M. (1997), ‘Constitution-making, identity building, and peaceful transitionto democracy: Theoretical reflections inspired by the spanish example’, Cardozo L.Rev. 19, 1891.

Sang-Hun, C. (2015), ‘Kim young-sam, south korean president who opposed military,dies at 87’.

Schwartz, H. (2000), The struggle for constitutional justice in post-communist Europe, Uni-versity of Chicago Press.

Siegel, R. L. (1998), ‘Transitional justice: A decade of debate and experience.’, HumanRights Quarterly 20(2), 431–454.

Sikkink, K. (2011), The Justice Cascade: How Human Rights Prosecutions Are ChangingWorld Politics (The Norton Series in World Politics), WW Norton & Company.

Sikkink, K. & Walling, C. B. (2007), ‘The impact of human rights trials in latin america’,Journal of peace research 44(4), 427–445.

Simon, H., Smithburg, Donald & Thompson, V. (1950), Public Administration, Knopf.Snyder, J. & Vinjamuri, L. (2004), ‘Trials and errors: Principle and pragmatism in

strategies of international justice’, International security 28(3), 5–44.Southall, R. (2020), ‘Flight and fortitude: the decline of the middle class in zimbabwe’,

Africa 90(3), 529–547.Stan, L. (2006), ‘The vanishing truth? politics and memory in post-communist europe’,

East European Quarterly 40(4), 383–409.Stan, L. (2012), ‘Witch-hunt or moral rebirth? romanian parliamentary debates on lus-

tration’, East European Politics and Societies 26(2), 274–295.Stan, L. (2013), ‘Reckoning with the communist past in romania: A scorecard’, Europe-

Asia Studies 65(1), 127–146.Stan, L. & Nedelsky, N. (2015), Post-communist Transitional Justice: Lessons from Twenty-

five Years of Experience, Cambridge University Press.Stan, L. et al. (2009), Transitional justice in Eastern Europe and the former Soviet Union:

Reckoning with the communist past, Routledge.Stokes, S. C., Dunning, T., Nazareno, M. & Brusco, V. (2013), Brokers, voters, and clien-

telism: The puzzle of distributive politics, Cambridge University Press.Svolik, M. W. (2012), The politics of authoritarian rule, Cambridge University Press.Szczerbiak, A. (2002), ‘Dealing with the communist past or the politics of the present?

lustration in post-communist poland’, Europe-Asia Studies 54(4), 553–572.Tavits, M. (2005), ‘The development of stable party support: Electoral dynamics in

post-communist europe’, American Journal of political science 49(2), 283–298.Teitel, R. G. (2003), ‘Transitional justice genealogy’, Harv. Hum. Rts. J. 16, 69.Thoms, O. N., Ron, J. & Paris, R. (2010), ‘State-level effects of transitional justice: what

do we know?’, International Journal of Transitional Justice 4(3), 329–354.Tillack, H.-M. (n.d.), ‘A tale of gazoviki, money and greed’.Todd, S. C. et al. (2007), A commentary on Lysias, speeches 1-11, Oxford University Press

on Demand.Todd, S. C. et al. (n.d.).Tyson, S. A. (2016), ‘The agency problem underlying the use of repression’.UN Integrated Regional Information Networks (Nairobi) (2011), ‘Comoros; missing

guns delay demobilization process’, Africa News .


United States Institute of Peace (2011a), ‘Truth commission: Ecuador 07’.United States Institute of Peace (2011b), ‘Truth commission: Ecuador 96’.United States Institute of Peace (2011c), ‘Truth commission: Germany 92’.United States Institute of Peace (2011d), ‘Truth commission: Germany 95’.United States Institute of Peace (2011e), ‘Truth commission: Kenya’.United States Institute of Peace (2011f), ‘Truth commission: Liberia’.United States Institute of Peace (2011g), ‘Truth commission: Paraguay’.United States Institute of Peace (2011h), ‘Truth commission: Peru 01’.Van der Merwe, H., Baxter, V. & Chapman, A. R. (2009), Assessing the impact of transi-

tional justice: Challenges for empirical research, US Institute of Peace Press.Vilasi, A. C. (2015), The History of the Stasi, AuthorHouse.Vinjamuri, L. & Snyder, J. (2004), ‘Advocacy and scholarship in the study of interna-

tional war crime tribunals and transitional justice’, Annual Review of Political Science7, 345–362.

Waldmeir, P. (1997), Anatomy of a miracle: The end of apartheid and the birth of the newSouth Africa, WW Norton & Company.

Weber, M. (1968), ‘Economy and society (g. roth & c. wittich, eds.)’, New York: Bed-minster .

Weeks, J. (2014), Dictators at war and peace, Cornell University Press.Wilde, A. (1999), ‘Irruptions of memory: expressive politics in chile’s transition to

democracy’, Journal of Latin American Studies 31(02), 473–500.URL: https://www.cambridge.org/core/journals/journal-of-latin-american-studies/article/irruptions-of-memory-expressive-politics-in-chiles-transition-to-democracy/1AFD03BA8793ACB237A57EAFC610B219

Williams, K., Fowler, B. & Szczerbiak, A. (2005), ‘Explaining lustration in central eu-rope: a ’post-communist politics’ approach’, Democratization 12(1), 22–43.

Wilson, J. Q. (2019), Bureaucracy: What government agencies do and why they do it, BasicBooks.

Wright, J. (2008), ‘Do authoritarian institutions constrain? how legislatures affect eco-nomic growth and investment’, American Journal of Political Science 52(2), 322–343.

Yarhi-Milo, K. (2013), ‘Tying hands behind closed doors: the logic and practice of se-cret reassurance’, Security Studies 22(3), 405–435.

Zvobgo, K. (2019a), ‘Demanding truth: The global transitional justice network and thecreation of truth commissions’, Unpublished manuscript .

Zvobgo, K. (2019b), ‘Designing truth: Facilitating perpetrator testimony at truth com-missions’, Journal of Human Rights 18(1), 92–110.

· contents 1 letting sleeping dogs lie . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....

Documents