II. The Formal Model

Can lustration improve the quality of representation by eliminating the effectiveness of blackmail that uses dark authoritarian secrets? Our model of incomplete information reconstructs the mechanism through which blackmail interferes with politicians’ ability to represent faithfully the policy preferences of constituents. This model allows us to examine the extent to which personnel transitional justice indeed reduces the dependence of politicians on their former leading officers.

Formal modeling has scarcely been used to address questions involving transitional justiceFootnote ¹⁸ and, to the best of our knowledge, it has not been used to predict the effect of personnel transitional justice mechanisms on the long-term quality of democracy. Our model is set following a transition to democracy. It features a former collaborator, who is elected to public office (the politician), and his or her former officer, who demands a policy concession and threatens to reveal compromising information about the politician unless the latter provides that concession (what we understand as blackmail).

We assume that barring blackmail, politicians would implement the policy preferences of their constituents. By making this assumption we are not claiming that politicians always follow voter preferences, but the assumption is instrumental to isolating the effect of lustration that we seek to identify. The quality of representation suffers when the politician is vulnerable to extortion and implements policy preferred by a former secret police officer instead of policy preferred by voters. Lustration should make it more difficult if not impossible to blackmail these politicians.

The key tension in our model derives from the officer knowing whether he or she possesses the evidence he or she threatens to release while the politician does not know. The more stringent the lustration law, the less likely it is that the officer possesses incriminating evidence. Therefore, the severity of lustration is represented by the probability that evidence of the politician’s collaboration remains in the possession of the officer. Restated in terms of the example above, if harsh lustration had been implemented in Poland before Wałęsa served as president, the secret police chief would not have been able to use the files against him.

The model we propose uncovers circumstances under which lustration improves the ability of politicians to represent their constituents. Previewing the results, our model shows that ideological polarization between the blackmailing officer and the politician decreases the quality of representation. The finding that politicians give in to blackmail more frequently when asked to deviate more is intuitive.Footnote ¹⁹ The model also indicates that the quality of representation increases with the severity of lustration and that high-quality representation is easier to achieve as the price of having compromising information revealed decreases. Thus, our model shows that the greater the public’s emphasis on former collaborators being excluded from politics, the more influence authoritarian elites have via this blackmail mechanism.

Preferences

The payoffs are a quadratic function of the euclidean distance between the players’ respective ideal points and implemented policy, as well as the two types of costs characterized above: the cost to the politician of being fired as a result of having compromising information revealed and the cost of bluffing incurred by the officer if he or she makes an empty threat. The utilities, along with the entire game tree, are presented in Figure 1.

Figure 1

Lustration Blackmail Game

Results and Discussion

We solve the model for separating, semiseparating, and pooling equilibria in the supplementary material.Footnote ²³ Proposition 1 summarizes these equilibria and Figure 2 presents them as a function of F, the politician’s cost of having skeletons in the closet revealed. The set of conditions for the pooling equilibria is distinct from the set of conditions defining the separating and semipooling equilibria, with the critical element being the magnitude of F relative to the cost of bluffing, c, the polarization between the officer and the politician and the probability that evidence against the politician exists, π.

Figure 2

Equilibria

Proposition 1. Suppose $$F \lt a - \sqrt {a^2 - c}$$. There is a pure separating pbe in which the officer makes a demand if and only if evidence exists, and in response, the politician makes a counteroffer, $$x^* \equiv \sqrt F$$. The officer’s rejection region in this equilibrium is defined by $$R^* \equiv \left( {0,\sqrt F } \right)$$, so the officer accepts the counteroffer and does not reveal the evidence in his or her possession. The posterior beliefs of P are described by Pr(E|no demand) = 0,Pr(¬E|no demand) = 1,Pr (E|no demand) = 1,Pr (¬E|no demand) = 0.

Suppose $$F \gt {{a - \sqrt {a^2 - c} } \over \pi }$$. There is a pure pooling pbe in which the officer always makes a demand, and the politician responds with a counteroffer, $$x'' \equiv \sqrt {\pi F}$$. The officer’s rejection region in this equilibrium is defined by $$R'' \equiv \left( {0,\sqrt {\pi F} } \right)$$, thus, the officer accepts the counteroffer and does not reveal the evidence in his or her possession. The posterior beliefs of P are described by Pr(E|demand) = π, Pr(¬E|demand) = 1 − π.

Suppose a $$a - \sqrt {a^2 - c} \le F \le {{a - \sqrt {a^2 - c} } \over \pi }$$. There is a semiseparating pbe in which the officer always places a demand when evidence exists, but if evidence does not exist, the officer places a demand with probability λ* and refrains from placing a demand with probability 1 − λ*. In response to the demand, the politician makes a counteroffer, $$x' \equiv a - \sqrt {a^2 - c}$$, and since the officer’s rejection region is defined by $$R' \equiv \left( {0a - \sqrt {a^2 - c} } \right)$$, the officer accepts the counteroffer and does not reveal the evidence in his or her possession. The posterior beliefs of P are described by Pr(E|no demand) = 0, Pr(¬E|no demand) = λ*, Pr(E|demand) = 1, Pr(¬E|demand) = 1 − λ*, where $$\lambda^* \equiv {{\pi \left( {F - \left( {a - \sqrt {a^2 - c} } \right)} \right)} \over {\left( {1 - \pi } \right)c}}$$.

Informally, first, Proposition 1 states that for very low costs of revealing skeletons in the closet relative to polarization and the costs of bluffing, there is a separating equilibrium where the officer only makes a demand if evidence exists and the politician makes a counteroffer, $$x^* = \sqrt F$$. Second, it states that if the costs of revealing skeletons are very high relative to the costs of bluffing, there is a pooling equilibrium in which the officer always makes a demand, irrespective of whether evidence exists, and the politician makes a counteroffer, $$x'' \equiv \sqrt {\pi F}$$. Third, for moderate costs of revealing skeletons in the closet, there is a semipooling equilibrium where the officer sometimes bluffs (makes a demand even when he or she has no evidence) and the politician responds with a counteroffer, $$x' = a - \sqrt {a^2 - c}$$. In all equilibria, the counteroffer is accepted.

For the given set of parameters, all the equilibria are unique, provided two plausible refinements. The first is that neither type plays a weakly dominated strategy.Footnote ²⁴ The second is that rejection regions that would not be credible for the officer to play cannot be part of an equilibrium profile.Footnote ²⁵

The first implication from this set of propositions is that the quality of representation decreases with a, the distance between the officer and the politician. Intuitively, the greater the preference divergence between the two, the more the politician must concede to keep the compromising information from coming out (Figure 3).Footnote ²⁶ The second is that the quality of representation is decreasing in π. This is intuitive because a higher probability that evidence does not exist induces the politician to make a counteroffer closer to his or her ideal point. The third implication is that the quality of representation is highest under the pure separating equilibrium and lowest under the pure pooling equilibrium, indicating that it decreases with the cost of firing. Similarly, a decrease in c, the cost of bluffing, leads to more misrepresentation (Figure 4).

Figure 3

Misrepresentation as a Function of Cost of Revealing Skeletons, F

Figure 4

Misrepresentation as a Function of Cost of Bluffing, c ^a

aThe figure presents misrepresentation in response to the officer’s demand, not total misrepresentation. This means that if the misrepresentation was weighted by the frequency of occurrence, it would be lower under the separating equilibrium (where demands are only made π of the time) than under the pooling equilibrium (where demands are made all of the time).

These results are what we would expect from a model of blackmail. But the calculation of the separating, pooling, and semiseparating equilibria provide additional insights that allow us to evaluate the forward-looking argument for lustration. First, note that lustration works as it ought to only under the pure separating equilibrium where it prevents blackmail entirely: only officers with evidence will succeed at making effective demands. Under the pure separating equilibrium, bluffing never occurs and the amount of misrepresentation tracks the amount of evidence left in the officer’s possession so it is directly responsive to the severity of lustration. But both the semiseparating equilibrium and especially the pooling equilibrium allow blackmail to take place even after lustration has been implemented. The conditions for the pooling equilibrium vis-à-vis the semipooling equilibrium and the semipooling equilibrium vis-à-vis the separating equilibrium are easier to satisfy as the cost of bluffing decreases because for higher values of c, it is easier for F to exceed $$a - \sqrt {a^2 - c}$$ and $${{a - \sqrt {a^2 - c} } \over \pi }$$, respectively. Figures 2 and 5 capture these intuitions graphically by calculating (at the bottom of Figure 2), for all three equilibria, the expected amount of misrepresentation.

Figure 5

Expected Misrepresentation as a Function of F

As explained above, in the separating equilibrium blackmail is restricted to cases beyond the reach of lustration because evidence is left in the hands of former secret police officers. The more severe the lustration policy, the less blackmail: π of the time, the misrepresentation is $$\sqrt F$$. The expected misrepresentation is then $$\pi \sqrt F$$. But in the pooling equilibrium, bluffing occurs every time evidence does not exist, regardless of how much evidence transitional justice has left in the hands of former officers. The frequency of blackmail is completely orthogonal to the actual existence of evidence of collaboration. Under this equilibrium, the extent to which lustration has successfully eliminated evidence of collaboration has no bearing on the effectiveness of blackmail. The expected misrepresentation here is $$\sqrt {\pi F}$$.

Additionally, in the semipooling equilibrium, the officer always makes a demand when the evidence exists and makes it with probability λ* when evidence does not exist. The key feature of the semipooling equilibrium is that when it is played, the officer extracts policy concessions (undermining representation) even when evidence does not exist, with probability λ*. Misrepresentation does not exactly track transitional justice, but at least in 1 – λ* of the cases, when the transitional justice process has swept evidence, bluffing does not take place. We can calculate the expected departure from perfect representation in this equilibrium. It is given by $$\pi \left( {a - \sqrt {a^2 - c} } \right) + \left( {1 - \pi } \right){{\pi \left( {F - \left( {a - \sqrt {a^2 - c} } \right)^2 } \right)} \over {\left( {1 - \pi } \right)\left( {a - \sqrt {a^2 - c} } \right)^2 }}$$, which simplifies to $$\pi \left( {a - \sqrt {a^2 - c} } \right) + {F \over {\left( {a - \sqrt {a^2 - c} } \right)}} - 1$$. Figure 5 summarizes this argument graphically for π = 0.65, a = 0.55, and c = 0.3.

In this section, we identify conditions under which lustration works as it ought to according to the forward-looking argument—preventing blackmail that uses secret police files. We also identify conditions for the pooling and semiseparating equilibria where lustration is compatible with bluffing and thus does not prevent departures from democratic representation. Under all three equilibria, the total expected amount of misrepresentation decreases with lustration. Even when the mechanism through which lustration affects the quality of representation is somewhat different from that which proponents of the forward-looking argument maintain, lustration still decreases departures from the politician’s ideal point toward the blackmailing officer. Indeed, a somewhat counterintuitive result is that in terms of F, the cost of having skeletons in the closet revealed, the lowest level of policy concessions occurs in the pooling equilibrium. Recall that this is the instance where lustration works nothing like the proponents of the forward-looking argument expect it to, as blackmail occurs regardless of whether evidence exists. This result is intriguing. The intuition behind it is that because the politician knows that the chance that evidence against him or her exists is merely π, the politician is not willing to put up with the kind of high concessions an officer with evidence would demand (for example, the demand in the separating equilibrium where there is 100 percent certainty that insubordination would result in revealing compromising information). The fascinating thing about this equilibrium is what keeps the officer from extracting more than $$\sqrt {{\rm{\pi F}}}$$: the politician is willing to risk exposure rather than put up with an excessively high concession. After accounting for the frequency with which these concessions occur, it becomes apparent that the separating equilibria, in expectation, lead to the highest quality representation.

The next section derives this observation and other empirical implications from our model and illustrates these expectations with original data from the Global Transitional Justice Dataset.

III. Empirical Analysis

The game theoretic model presented above represents the dynamic between individual politicians and former officers of authoritarian regimes. The main implication drawn from the model is that lustration prevents blackmail and thus curbs departures from policies that are desirable to the electorate. Put differently, we argue that in a perfect representative democracy, the politician’s ideal point would correspond to policies that the voters want implemented. Because the outcome of interest is the distance between the politician’s ideal point and the policy that is ultimately implemented, the model predicts individual departures from such perfect representation. Of course, obtaining the necessary evidence to document these departures would require careful analysis of the preferences of individual politicians and former authoritarian elites, a task that would be hard at best and almost impossible for a large number of countries. Hence, we opt to test the model’s implication using a combination of party-level data on ideological positions and quality of representation and country-level data on lustration policies.

We operationalize our outcome of interest, quality of representation, using a measure that captures both the cohesion within a party and the salience of the policies that the parties represent. The ideological cohesion of a party relies on its members sharing the same ideological platform and supporting policies that advance that platform.Footnote ²⁷ If one of the members were blackmailed into making policy concessions to a former agent, the departure from the shared platform would decrease party cohesion. Hence, we can rely on data aggregated at the party level to pick up irregularities in individual behavior.

To measure quality of representation at the party level, we build on the measure CoSalPo constructed by Herbert Kitschelt and colleagues.Footnote ²⁸ The original measure draws on a large-scale survey of experts in eighty-eight countries and condenses information on three key components of 506 political parties: cohesion, salience, and polarization.Footnote 29 We use a similar measure but leave out questions pertaining to polarization because it is conceptually close to the distance between the ideal point of the politician and the officer, a key parameter of interest in our model. The result is a score between 0 and 1 that captures the cohesion and salience of each party in the countries of interest. The new proposed measure (CoSal_3), we argue, accurately captures the quality of representation that we refer to in the formal model.Footnote ³⁰

The model’s first implication suggests that the probability of the existence of secret evidence against the politician (parameter π in the model) decreases the quality of representation because it enables blackmail. Throughout this article, we contend that lustration publicizes evidence of covert collaboration so the probability of evidence remaining secret and susceptible to being used as blackmail decreases. Therefore, we operationalize the probability of secret evidence of collaboration (π) with country data on lustration policies.

Although most cross-national empirical analyses of transitional justice rely on dichotomous variables to indicate lustration,Footnote ³¹ we argue that indicator variables fail to distinguish between countries with limited lustration policies (for example, screening only politicians running for elected top-level positions) and countries with comprehensive lustration policies that screen candidates for all elected positions. Yet it is more likely that secret information would be left undisturbed and ready for use as blackmail under a limited lustration policy than under a comprehensive one. An adequate measure should reflect the variation on lustration policies across countries.

To capture this variation, we rely on the Global Transitional Justice Dataset.Footnote ³² To construct it, we started by building chronologies of personnel transitional justice events. For each country, we recorded each nudgeFootnote ³³ in the direction of promoting or restricting lustration, and coded such events as progressive or regressive, respectively. Progressive events introduce new legislation or broaden the existing one, whereas regressive events, such as the constitutional court voting down or vetoing lustration provisions, attempt to constrain lustration. Submitting a lustration proposal to the floor of the legislature, passing or upholding lustration legislation, or overturning a presidential veto against such legislation are all examples of progressive events, whereas amendments decreasing the set of offices subject to lustration are coded as regressive events.Footnote ³⁴ The events we analyze are only those that deal with uncovering secret collaboration with the authoritarian regime. Although this is a narrower definition of lustration than that used by other scholars,Footnote ³⁵ it is an excellent empirical match for our model, which entails revealing information that no one except the officer knows.

Figure 6 summarizes these data and shows that there is empirical variation in the lustration policies implemented by each country (fifty-nine).Footnote ³⁶ Each panel in the figure depicts lustration events in a given country after transition. The horizontal axis indicates the calendar year and the vertical axis indicates the frequency of events in that year: black, upward lines mark progressive events, and gray, downward lines mark regressive events. Taken together, the panels show that in many countries, lustration was not implemented at a single point in time and was allowed to persist. Rather, among proposed lustration bills, some were passed via the legislature, whereas others failed or were eventually overturned (by an executive, an upper chamber, or a constitutional court) or survived one term and were struck down by a subsequent government. Figure 6 also suggests that lustration is popular beyond postcommunist Europe, contrary to what the concentration of work on lustration in postcommunist Europe may imply.

Figure 6

Summary of Regressive and Progressive Lustration Events

To maximize the richness of the lustration trajectories presented here, we propose a score that captures the severity of lustration. This score is defined as the total number of progressive lustration events divided by the total number of lustration events plus one,Footnote ³⁷ taking the value of 0 when country k did not move lustration forward, and approaching 1 when the number of progressive events is closer to the total number of events.Footnote ³⁸ We add 1 to the denominator to ensure that this score is defined for countries that have had no lustration events. Importantly, the severity variable is observed at the country level, whereas the dependent variable, quality of representation, is measured at the party level.

We believe that the events-based severity score is an improvement over existing measurement, although it is far from ideal. First, our measure of severity is time invariant, which means that it cannot explain variation in the quality of democracy over time.Footnote ³⁹ Second, our measure could improve if it accounted for how many past and current positions were subject to vetting, how severe the sanctions for collaboration were, or how many individuals were exonerated given the number of people investigated. Such detailed information, although clearly superior, would require meticulous data on policy implementation. Although collecting it might be feasible for a handful of cases, it is prohibitively costly for the entire sample that we present here. Given our resources, we opted to focus on the theoretical drivers of politicians’ behavior. In our model, this behavior is driven by the politician’s beliefs about how likely it is that blackmail will occur. The higher the proportion of progressive lustration events, as reflected in our measure of severity, the more likely it is that politicians will not be blackmailed with secret police files. With this in mind, we state our first hypothesis:

A second empirical implication of the model is that the quality of representation decreases as the cost of having compromising information revealed increases (parameter F in the model). We operationalize it as the electoral costs associated with revealing skeletons in the closet. We argue that in democracies recovering from authoritarian rule, voters care about keeping former secret collaborators out of office, but that the saliency of this issue diminishes over timeFootnote ⁴⁰ when holding lustration constant. This claim is supported by the literature on historical memoryFootnote ⁴¹ and is also associated with the fact that more voters in a young democracy will have experienced (and remember) life under autocracyFootnote ⁴² than voters in a seasoned democracy. Hence, we operationalize F as the years elapsed since transition,Footnote ⁴³ and propose that the longer the democratic experience of a country, the less attentive voters will be to the fact that their politicians collaborated with the authoritarian regime. When voters pay less attention to collaboration, publicizing such skeletons becomes less costly and politicians will give in to blackmail less frequently, regardless of the lustration policies implemented in that country. Formally:

It is important to note that this operationalization is somewhat coarse because the cost of revealing collaboration is not only a function of the how voters react to such allegations, but also of other institutional features. For example, revelations about collaboration could be costlier in electoral systems with the personal vote, such as single member districts or open list proportional representation, than in systems in which voters cannot directly specify which candidate is elected into office.Footnote ⁴⁴ Similarly, voters in some countries could be more interested in elections than those in other countries, in which case revelations would be mediated by an overall interest in electoral participation or turnout. Nevertheless, we believe that the political cost of revealing skeletons in a politician’s closet should be higher, on average, in recently transitioned countries than in countries that transitioned a long time ago.

The third implication of our theory is that the ideological distance between politicians and officers negatively impacts the quality of democracy. Like the dependent variable, quality of representation, our operationalization of distance relies on the Democratic Accountability and Linkages Project (dalp) survey and is also measured at the party level. Specifically, we take advantage of the fact that dalp reports average ideological placements of political parties in countries from our sample. Furthermore, since some of these parties are successor authoritarian parties, we use their placement to proxy the ideological placement of officers. Therefore, for each nonsuccessor party p in country k, we define party distance as the absolute difference in left-right placement between party p and the placement of the country’s successor authoritarian party in country k. We state our third hypothesis:

It is important to note that some countries in our sample do not have a party that is a successor to the authoritarian regime. For these countries, we replaced the missing distance with the average distance between successor and authoritarian party across all remaining countries and we included an indicator variable to account for these cases.Footnote ⁴⁵

We operationalize parameter c, the cost of bluffing, with the freedom of the press index as reported by Reporters Without Borders.Footnote ⁴⁶ Our logic is that when an officer of the authoritarian regime tries to bluff a politician into thinking that he or she is in possession of evidence against the politician, such a bluff can be easily exposed in an environment with free media. Put differently, claims regarding the collaboration between a politician and the former secret police will likely be reported, and if the press is free, claims eventually will be investigated and denounced if found to be false. A wide literature discusses how thorough journalists are when it comes to revealing evidence of former secret police conduct.Footnote ⁴⁷ Although the scrutiny offered by journalists is not perfect, the higher the degree of media independence, the more likely an officer’s bluff will be called.

Our model predicts that an increase in the costs of bluffing increases the chances of an equilibrium with no bluffing, that is, an equilibrium in which only politicians with skeletons in their closets are blackmailed. Therefore, holding the severity of lustration constant, an increase in media freedom should increase the quality of representation because it decreases (or eliminates altogether) the amount of bluffing. Put differently, our model implies that severe lustration directly improves the quality of representation by decreasing the pool of politicians who could be subject to blackmail or bluffing (Hypothesis 1), but even for two countries with equal lustration policies, the country with greater media freedom will have an added positive effect on the quality of representation by eliminating bluffing. Thus, we hypothesize:

Our hypotheses can be translated to the following linear specification:Footnote ⁴⁸

$$\eqalign{ quality \, of \, representation_{p,k} = & {\text{ }}\beta_k + \beta_a^* party \, distance_{p,k} \cr & + {\text{ }}\beta \cr _s^* severity_k \cr & + \beta_n^* years \, since \, transition_k \cr & + \beta \cr _c^* press \, freedom_k \cr & + \beta_x^* \sum X_k \cr & + e_{p,k} + u_k , \cr} $$

where quality of representation is operationalized using the measure of programmaticness (CoSal_3) for party p in country k. In addition to the main variables of interest described above, the model incorporates a series of country-level control variables (ΣX_k), as well as a series of country-specific intercepts (β_k).

As implied by the equation above, the outcome of interest (quality of representation) and one regressor (party distance) are party-level attributes, while the remaining variables are measured at the country-level. To account for this structure, we use a multilevel regression. Given that parties are nested within countries, estimating this model using an ordinary least squares regression would risk biased and inefficient results because characteristics of parties within the same country are not independent of one another.Footnote ⁴⁹ Multilevel regression can be thought of as a compromise between assuming that all parties are independent of each other and assuming that all variation across parties can be explained by the country in which they exist. A multilevel approach will structure the variation in the dependent variable according to party-level and country-level characteristics. Specifically, we estimate the model above with a series of country-level coefficients (β_k) that share an underlying distribution with a global mean (μ_βk) and standard deviation (σ_βk).Footnote ⁵⁰ This equation, we argue, is the best way to model the outcome of interest given the data available.

Before we present and discuss the results, we address the concern that countries where lustration is successfully implemented also have parties that are better at representing voters. If this were the case, there would be a risk of overestimating the relationship between lustration severity and quality of representation, but such overestimation would be captured by the country-level intercepts. Put differently, if countries that implemented transitional justice were indeed the same ones whose parties offer higher quality representation, then all parties in that country should be affected. This effect, in turn, would be captured by the country-specific estimates.

The results of our estimations are shown in Table 1. For ease of interpretation, all variables have been linearly transformed to range between 0 and 1. All models presented here include country-level intercepts (estimations for β_k) and an indicator variable controlling for missing successor parties, both of which are omitted from the table due to space considerations. The table reports the average change in quality of representation of a nonsuccessor party that is associated with an increase of one unit in the covariate of interest.

TABLE 1

Quality of Representation and Severity of Lustration

*p <0.1, **p <0.05, ***p <0.01

The resulting estimations are generally consistent with the implications of our theoretical model. Most important, they show that higher scores of lustration severity are on average associated with more programmatic parties. The effect is statistically significant at the 0.05 percent confidence level and robust to the inclusion of country-specific intercepts and other relevant covariates. In addition, its average effect is quite substantial: the difference between a country that did not attempt lustration (severity score of 0) and a country with the highest lustration (severity score of 0.82) is 0.17 points, according to the point estimate of model 4. Since the variable quality of representation ranges from 0 to 1, this average effect accounts for one-fifth of the possible range of our outcome of interest.

The models reported are static, meaning that they take a snapshot of a country, and thus cannot produce evidence of changing quality of representation over time. Nevertheless, they have many redeeming features. First, dalp extends to over eighty democracies and exhibits almost precise overlap with the set of countries included in our database. No competing data sets (ches [Chapel Hill expert surveys] and the Manifesto Project) allow us to measure quality of representation for even half the countries in the Global Transitional Justice Dataset.

We find a negative and statistically significant effect for the ideological distance between each party and its successor (party distance). This finding is consistent with Hypothesis 3, which states that as the distance between each party and the successor party increases, the quality of representation should decrease. This result is robust to several specifications, most notably using only countries with successor parties (see the supplementary material).Footnote ⁵¹

But this result could be driven by preexisting dynamics between the opposition and authoritarian parties. Suppose, for instance, that parties that existed during the authoritarian period share some ideological characteristics that made their members more likely to be blackmailed. We control for this possible confounder using a measure of how active the opposition was in representative state institutions during the authoritarian era. José Cheibub, Jennifer Gandhi, and James VreelandFootnote ⁵² report data on the status of the opposition in authoritarian legislatures.Footnote ⁵³ In our sample, only two authoritarian regimes had no legislature at all, so we transformed the variable opposition status into an indicator taking the value of 1 when 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 our results.

Models 3–5 include the variable years since transition, which captures time since transition to democracy as a proxy for the cost of having skeletons revealed (F). We hypothesize (Hypothesis 2) that over time, politicians pay a lower cost for having been exposed as collaborators of the former authoritarian regime. As a result, they should be less prone to blackmail and the quality of representation should increase. But the coefficient associated with years since transition is not significant, which means that we find no evidence to support Hypothesis 2. We attribute this lack of support to the coarse measure of the costs of firing used in our analysis.

Models 4 and 5 also include the variable press freedom, which is how we operationalize the cost of bluffing. We find that the variable is positively related to the quality of representation, although this association is significant at the 90 percent level. We also find that it does not change the estimated effects of the rest of the theoretically relevant regressors, which we take as evidence supporting our previous findings. Interestingly, the constant term is no longer statistically significant when press freedom is included, and we interpret this as a type of omitted variable bias; in models 1–3, the average effect of a free press and time since transition is encapsulated by this constant.

Figure 7 shows our main relationship of interest, the one between quality of representation and severity of lustration. The figure shows the severity score on the horizontal axis, and the average country effects estimated in model 1 on the vertical axis. Model 1 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-level quality of representation. Figure 7 suggests a positive relationship between these two variables. The pattern seems to be much clearer for the observations that fall in the right half of the panel, that is, for the countries that implemented more severe lustration policies. This latter point supports our decision to operationalize lustration as a process rather than as a dichotomous measure.

Figure 7

Severity of Lustration and Average Quality of Representation: Intercepts by Country

In addition, the evidence shows that the severity of lustration is positively related to the quality of representation. This result is promising insofar as it suggests that lustration severity contributes to higher quality representation. But the theoretical model also shows that lustration by itself does not prevent the blackmailing of former dissidents turned politicians. To function properly, lustration policies require an accompanying free press to prevent former authoritarian elites from bluffing.