Liberty, Security, and Accountability: The Rise and Fall of Illiberal Democracies

Abstract

We study a model of the rise and fall of illiberal democracies. Voters value both liberty and economic security. In times of crisis, voters may prefer to elect an illiberal government that, by violating constitutional constraints, offers greater economic security but less liberty. However, violating these constraints allows the government to manipulate information, in turn reducing electoral accountability. We show how elements of liberal constitutions induce voters to elect illiberal governments that remain in power for inefficiently long—including forever. We derive insights into what makes constitutions stable against the rise of illiberal governments. We extend the model to allow for illiberal governments to overcome checks and balances and become autocracies. We show that stronger checks and balances are a double-edged sword: they slow down autocratization but may make it more likely. We discuss the empirical relevance of our theoretical framework and its connection to real world examples.

1 INTRODUCTION

Around the world, voters are increasingly electing governments that operate beyond the liberal constraints of their constitutions (Foa and Mounk, 2017; Svolik, 2019). The governments of Jarosław Kaczyński in Poland, Viktor Orbán in Hungary, and Recep Tayyip Erdoğan in Turkey have infringed (albeit to different extents) fundamental liberal rights: freedom of speech, civil rights, and the rule of law (Huq and Ginsburg, 2018; Luo and Przeworski, 2023). These governments also routinely abused their constitutional powers to manipulate information. They engaged the state bureaucracy to produce favourable or misleading information, or harassed bureaucrats who produced unfavourable reports. They co-opted media shareholders into the government, or threatened media outlets with lawsuits or the withholding of state funding and advertising.¹ Yet, they all remained formally accountable to the electorate and seldom, if at all, resorted to coercive means of repression.² These illiberal democracies rose to power with significant popular support and some remained sufficiently popular to ensure their own reelection.³

Much attention has been given to the rise of illiberal democracies (recent examples include Rosenbluth and Shapiro, 2018; Berman, 2019; Luo and Przeworski, 2023; Grillo and Prato, 2023; Chiopris et al., 2021; Svolik, 2020). However, illiberal democracies also frequently fall. Why do some illiberal democracies prove more resilient than others? When will voters who support illiberal leaders turn against them? When will voters’ illiberal tendencies ultimately lead to autocratization?

In this paper, we offer a theory of the rise and fall of illiberal democracies. We use this theory to draw insights into what makes some liberal constitutions stable against the lure of illiberalism. Our theory builds off two fundamental premises. Our first premise is that voters in democracies value both liberty and security.⁴ They value the fundamental tenets of the liberal state: individual freedoms and rights, the rule of law, and the protection of underrepresented minorities.⁵ All else equal, they prefer a more just and fair society. But voters also value security: being protected from negative shocks to their economic interests, physical security, and welfare.

Our second premise is that liberal constitutions place formal constraints on executives with the aim of guaranteeing liberty and fostering accountability. For example, the U.S. constitution features anti-majoritarian elements, such as the composition of the Senate or the Supreme Court, that constrain the actions of the administration. Around the world, democratic governments are forbidden from enforcing laws that courts have ruled infringe constitutionally protected rights. They are also often legally forbidden from using their powers to manipulate voters’ information for electoral advantage. For example, governments may not recruit state employees for electoral campaigns,⁶ nor influence bureaucrats or media to censor unfavourable information. However, executives have de facto substantial leeway beyond these formal legal constraints—often limited only by their accountability to the voters.⁷

With these two premises in mind, we think of an illiberal democracy as a democratically elected government that, by operating beyond the liberal limits of the constitution, offers to voters something they may prefer to liberty: security. For example, in response to increasing terrorist attacks, illiberal governments can promise to voters a more incisive reaction, including extrajudicial treatment of suspects, or the suppression of dangerous media, movements, or even bad apples within the state. While limiting liberty, such policies may increase the economic security of the pivotal voter. In fact, illiberal leaders boast about their (presumed) results on economic security, quality public services, and protection against violent and economic threats from within and outside the country. In their rhetoric, their disrespect for formal liberties is exactly what enables them to deliver on these promises.⁸ Therefore, when voters receive information that suggests a greater need for security, they may prefer an illiberal government to one that is committed to operate within the limits of the constitution. Yet, voters should also reason that governments that silence media and bad apples will inevitably also (ab)use their powers to manipulate information in their own favour. Such information manipulation is likely to reduce effective electoral accountability, perhaps explaining the enduring electoral success of some illiberal democracies.⁹

Analysing this dynamic trade-off requires a model with three key ingredients: (i) a long-lived principal (voter) chooses in each period whether to hire an agent (the illiberal government); (ii) the principal’s per-period utility depends on an unknown period-specific binary state and whether she hired the agent or not; (iii) once an agent is hired, and until the principal chooses to dismiss him, the agent enjoys a rent and controls the flow of information the principal can observe about the state in future periods. While our focus is on the rise and fall of illiberal democracies, we believe this model may find useful applications in other contexts. For example, the choice to hire a professional manager exposes a business owner to a similar trade-off. Hiring a professional manager is costly, but the manager can more effectively navigate internal reorganizations or expand the business to new markets. However, once hired, the manager may not disclose to the owner all the relevant information on whether there is any need for reorganizations or new market opportunity. The owner can choose at any time to retake control of the business, reduce management costs, and see all the available information with her own eyes. However, doing so carries the risk of exposing the business to shocks that would have been better addressed by a professional manager.¹⁰

In both our professional manager example and in illiberal democracies, the agent has an incentive to persuade the principal that the next period’s state is such that it is better to retain the agent. However, if the information filtered by the agent is too opaque, the principal may prefer to dismiss the agent and see all the available information with her own eyes. This dynamic trade-off is what drives equilibrium behaviour in our model. We show that the quality of the available information affects both the principal’s value of dismissing the agent and the agent’s ability to manipulate information in his favour.

In our benchmark model, an infinitely lived and forward-looking voter chooses in each period whether to elect a liberal or an illiberal government. Liberal governments are committed to abide by the constraints of the constitution. Therefore, they offer the voter greater liberty, less security, and let the voter observe information from media and whistleblowers as much as guaranteed by the constitution. Illiberal governments do not abide by the constitution, so they offer less liberty, more security, and manipulate the information observed by the voter in their own favour. Depending on the features of the constitution and the observed information, in each period, the voter may prefer either a liberal or an illiberal government, optimally trading off liberty and security.¹¹ However, the voter knows that electing an illiberal government comes with an endogenously determined effective accountability cost, because the voter knows that she may be induced to retain the illiberal government even if, had she access to unadulterated media, she would prefer not to. We model the illiberal government’s problem as one of optimal Bayesian persuasion (Kamenica and Gentzkow, 2011). The voter knows the choice of information manipulation—which we call a censorship policy—but can only indirectly infer the information she would have otherwise observed had she chosen a liberal government.

In equilibrium, the parameters of the model determine which of four different regimes arises. At two extremes, the polity is either a stable liberal democracy with continuously liberal governments or a stable illiberal democracy with continuously illiberal governments. In both these cases, the choice of government is optimal for the voter given the uncensored information provided to her by institutions and media. However, the polity may also be an inefficient stable illiberal democracy. In this regime, the voter eventually resorts to electing an illiberal government that remains in power forever by censoring all relevant information for the voter. Absent censorship, the voter would actually find it optimal to cycle between liberal and illiberal governments. We show that an inefficient stable illiberal democracy arises only if the expected cost of security under full censorship is greater than the sum of the value of liberty and the discounted maximal accountability cost of illiberalism. When this condition fails, illiberal governments cannot ensure their own reelection by simply censoring all information. They need to provide the voter with sufficient information to persuade her to reelect them. In turn, this decreases the accountability cost of illiberalism, increasing the appeal of an illiberal government. This gives rise to a cycling liberal-illiberal democracy in which liberal and illiberal governments alternate, but illiberal governments remain in power for longer than would be optimal for the voter.

In our model, the voter elects illiberal governments during crises that increase her expected cost of security: the likelihood of a shock that can only be neutralized via illiberal means. However, because the illiberal government manipulates information, the voter may be persuaded to reelect the government even if there is no further crisis. We show how the probability that an illiberal government rises from a liberal democracy depends on the amount of constraints on the executive, the voter’s value of liberty, and the transparency of institutions and media. An immediate implication is that excessive constraints on the executive, which render liberal governments ineffective, are more likely to induce voters to elect an illiberal government. This result resonates with the literature suggesting that constitutional reforms that strengthen the executive’s power are needed to safeguard liberal democracies (e.g. Howell and Moe, 2020). We add to this that technological changes that increase transparency may exacerbate this perverse effect of excessive constraints on the government: when information leaks become more frequent, liberal stability demands that constitutions place fewer constraints on the executive.

We also characterize the optimal amount of censorship of illiberal governments and show how it depends on the voter’s values of liberty and security, the constraints on the executive, and the transparency of liberal institutions and media. The latter affects censorship through a persuasion effect and an accountability effect that go in opposite directions and we discuss how the stringency of the constraints on liberal governments determines which of the two effects dominates. The same factors also affect the likelihood that an illiberal democracy falls, returning the country to liberalism, when the illiberal government’s censorship fails to conceal to the voter that there is in fact no crisis. Again, the effect of more transparent liberal institutions depends on other features of the constitution, and can even be non-monotonic.

In Section 7, we study the welfare implications of the design of stable liberal constitutions and show that while more transparency is always favoured by the median voter, it may harm every voter with a higher relative value of liberty—therefore potentially reducing aggregate welfare.

In Section 8, we extend our model to allow us to study the effects of checks and balances that prevent illiberal democracies from consolidating their power and becoming autocracies. We observe that illiberal leaders may exacerbate the voters’ need for their economic security, for example by taking more aggressive diplomatic stances or refusing negotiations with separatist groups.¹² We therefore allow the illiberal government to manipulate the voter’s value of security once in power. Similarly, illiberal governments may sometimes be able to build the ability to overthrow the democratic constitution and dispense of electoral accountability altogether. In both cases, the government becomes a de facto autocracy that remains in power forever. Both processes are more likely to be set in motion when checks and balances preventing total executive takeover are weaker. We show that in the long run this gives rise to only two possible regimes: stable liberal democracies and autocracies. When the probability that the government can build an autocracy is greater, non-stable illiberal democracies become autocracies faster. However, because the voter anticipates this risk, more constitutions induce stable liberal democracies. Our results suggest that stronger checks and balances aimed at preventing total executive takeover, while slowing down this process, may actually induce voters to elect an illiberal government with greater probability—ultimately choosing the path to autocracy.¹³ Finally, we discuss some realistic long-term implications of censorship on the probability of the rise and fall of illiberal democracies.

As we discuss in Section 9, our theoretical framework helps make sense of individual and aggregate phenomena in the real world. We show that within consolidated democracies, individuals’ fears of physical and economic threats correlate with less support for liberalism: more worried voters believe that the protection of civil rights is less important and that it would be better to have a strong executive who does not need to bother with parliament. We argue that the data are consistent with the view we propose in this paper that such individual preferences for illiberalism respond to the amount of constraints on the executive and may have consequences for the action of governments. We also discuss how our results, both theoretical and empirical, relate to the empirical literature on the causes of the rise of illiberal governments in Western countries in the past decade. We conclude (Section 10) by reviewing the key policy implications of our model.

2 RELATED LITERATURE

Scholars have long wondered why and when illiberal leaders and governments are elected. The traditional view of democracy, from Montesquieu to Fearon, maintains that attempts to consolidate power from within a democratic constitution would be blocked by the voters.¹⁴ Besley and Persson (2019) argue that there is a natural complementarity between democratic values and institutions, creating persistence. This view has recently been challenged by the rise of populist, illiberal candidates even in consolidated democracies (Luo and Przeworski, 2023; Graham and Svolik, 2020). Foa and Mounk (2016) and Mounk (2018) argue that protracted dissatisfaction with the economic performance of liberal democracies has led voters to prefer more effective illiberal governments. Yet voters also value individual liberties, civil rights, and the rule of law (Graham and Svolik, 2020; Svolik, 2020). As a result, prevalent theories on the rise of illiberal democracies rely on factors that may make liberty less salient than economic performance, resolving the voters’ trade-off in favour of more effective, yet less liberal governments. Such factors include political polarization (Chiopris et al., 2021; Svolik, 2020) and the weakening of traditional mass parties (Rosenbluth and Shapiro, 2018; Berman and Snegovaya, 2019).¹⁵ Our model captures this fundamental trade-off between liberty and security, but by studying the choice of a forward-looking, infinitely lived voter,¹⁶ we are also able to capture another fundamental and dynamic aspect of illiberal democracies: the loss of electoral accountability.¹⁷ This allows us to understand how various constitutional and technological features interact in determining the rise, and—in contrast with previous literature—the fall of illiberal democracies. In particular, our model allows us to understand the role played by the transparency of the state bureaucracy and the media, informing a growing debate on the role of government transparency in the rise of illiberal politicians (Mudde and Rovira Kaltwasser, 2017; Mounk, 2018; Sgueo, 2018).

We model the ruling of the illiberal government as a process of information manipulation à la Kamenica and Gentzkow (2011). We share this feature with Gehlbach and Sonin (2014), Luo and Rozenas (2023), and Kolotilin et al. (2022) (see also Shadmehr and Bernhardt, 2015; Li et al., 2022). In a series of papers, Guriev and Treisman (2019, 2020b) document that many 21st century autocracies and illiberal democracies (what they call “informational autocracies”) maintain power through subtle control of information, rather than through violent repression of the opposition. Importantly, they also convincingly argue that contemporary autocrats and illiberal leaders focus on economic performance rather than repression when addressing the general public.

Our theory of the rise and fall of illiberal democracies partially overlaps with theories that aim to explain the demand for populism. Guiso et al. (2019, 2020) document how economic insecurity fuelled the demand for protection from globalization and the rise of populist leaders in Europe.¹⁸ Guriev and Papaioannou (2022) review the literature on the rise of populism focusing on the role of both economic and cultural causes. Although illiberalism and populism sometimes overlap, “only a minority of strongmen are populists and only a minority of populists is a strongman [$…$] the authoritarian characteristic of the strongman is not inherent to populism” (Mudde and Rovira Kaltwasser, 2017) (see also Guriev and Papaioannou, 2022).

In our model, illiberal governments offer protection to the median voter that a liberal government is unable to provide. While we do not mean to imply that illiberal governments are superior, we notice that this rhetoric is neither new nor confined to propaganda. Since at least Samuelson (1961), there has been a longstanding theory within economics and political science that some autocratic governments produce higher economic growth. Recently, Collier (2010) showed that, in some instances, a more liberal democracy can constrain economic growth. Furthermore, leaders facing less constraints on their power have been shown to have greater influence on a country’s economic performance and policies (Jones and Olken, 2005). Cheibub et al. (2020) document how more solid democracies were significantly slower in limiting individual freedoms to stop the spread of the COVID-19 pandemic (see also Stasavage, 2020a, 2020b). The offer of protection against the threat of socialism or elite capture was also at the basis of the rhetoric of fascist regimes in the 1930s (Acemoglu et al., 2022).

Our model studies a problem of dynamic delegation in which the illiberal government tries to persuade the voter to reelect it. This problem connects to the growing literature on dynamic Bayesian persuasion (Ely et al., 2015; Ely, 2017; Renault et al., 2017; Bizzotto et al., 2021; Ely and Szydlowski, 2020). In this literature (and using the principal-agent language from above), an agent controls the flow of information observed by a principal, and tries to persuade the principal to take an action either in each period or at a given deadline. Che et al. (2020) study a model in which, as in our framework, the agent can only commit to this period’s information structure, and the principal can choose when to take her action. Somewhat closer to us, Orlov et al. (2020) study a model in which the agent wishes the principal to take a single irreversible action as late as possible. However, in contrast with our model, the incentive for the principal to take a decision earlier is the real cost of waiting and the agent’s optimal choice depends on a persistent state. As a result, these model are not suitable to capture the trade-offs that arise when the principal can “fire” the agent to see (a more informative signal of) the truth for herself. That is, in our model, and in contrast to previous literature, the agent suffers from the fact that the principal dislikes being persuaded and the principal’s incentive to fire the agent increases when the agent is expected to engage in more persuasion in future periods. In addition, our model allows us to separate two concepts: the underlying state unobserved by the principal and the signal that the principal could observe were she to fire the agent. Disentangling these two concepts allows us to capture how transparency (the precision of uncensored information) affects the agent’s trade-off through a (common to the literature) persuasion effect and a (novel) accountability effect, which operate in opposite directions.

3 THE MODEL

3.1 Summary

We study a model with a forward-looking and infinitely lived voter. In each period, the voter chooses whether to elect a liberal or an illiberal government. A liberal government is a non-strategic actor that operates according to a constitution that constrains its action. An illiberal government does not abide by the constitution. We capture the stringency of the constitutional constraints on the executive with the parameter $π∈(0,1)$⁠.

The liberal constraints on the action of the executive have three effects.¹⁹ First, more stringent constraints guarantee greater liberties for the voter if she elects a liberal government. Therefore, more stringent constraints increase the value of liberty, $L=L(π)>0$⁠, of electing a liberal government, where $L$ is strictly increasing and continuous. Second, more stringent constraints limit a liberal government’s ability to adequately respond to negative shocks. Therefore, more stringent constraints increase the chances of a shock that only an illiberal government would be able to adequately respond to. If such a shock occurs and the voter has elected a liberal government, then the voter will suffer a cost.²⁰ After a normalization, electing a liberal government increases the expected cost of security for the voter by $πS$⁠, where $S>0$ is the voter’s value of security. The parameter $π$ is the probability of a shock that only an illiberal government would be able to adequately respond to.²¹ Third, more stringent constraints on the executive protect the flow of information from bureaucrats and mass media to the voter from governmental interference. Therefore, more stringent constraints increase the transparency of the information observed by the voter under a liberal government $q=Q(π)$⁠, where $Q$ is increasing and continuous.

Since it does not respect the constraints imposed by the constitution, an illiberal government does not guarantee the liberties that the voter would have under a liberal government and can freely set policies that intimidate bureaucrats from whistle-blowing and censor the media, so that the information observed by the voter is a garbled version of the one she would otherwise observe under a liberal government. The voter can observe the policies and intimidation chosen by the illiberal government but can only indirectly infer what information she may have received if she had chosen a liberal government instead. We call this information manipulation a censorship policy, but our model is more general in that it encompasses any mapping of the underlying signal of the state to a distribution of messages observed by the voter.

The objective of an illiberal government is to remain in power as long as possible. If the voter does not reelect an incumbent illiberal government, we assume that the fallen illiberal government ceases to exist; a new illiberal government is generated if and when the voter chooses to elect an illiberal government again.

Within our framework, a constitution is summarized by the constraints on the executive $π$ which, by limiting the power of the government, guarantee an amount of liberties to individuals and minorities $L=L(π)$ and of transparency and independence of the bureaucracy, the judiciary, and the media $q=Q(π)$⁠. The functions $L$ and $Q$ are the technological possibility frontier on the design of constitutions. In the analysis of Sections 4–6, we describe comparative statics on $π$⁠, $L$⁠, and $q$ separately. I.e. we study the effects of relaxing the technological constraint. We then return to the connection between the three elements in Section 7, where we derive normative conclusions about the design of stable liberal democracies.

3.2 Formal setup

A forward-looking voter lives for infinitely many periods $t∈{1,2,…}$⁠. In each period $t$⁠, the voter chooses the period-$t$ government $gt∈{i,ℓ}$⁠. The voter’s payoff in period $t$⁠, $v(gt,θt)$⁠, depends on the period-$t$ government, $gt$⁠, and a period-specific i.i.d. state, $θt∈{0,1}$⁠:

and $π$ is the probability that $θt=1$⁠, i.e. of the occurrence of a shock that the voter cannot be protected from unless the government is illiberal.

The government in period $t−1$⁠, $gt−1∈{i,ℓ}$ with $g0=ℓ$⁠, determines the information available to the voter at the beginning of period $t$⁠. In particular, for a given stringency of the constraints $π$⁠, there exists a signal $s(θt)∈{0,1}$ where $Pr(s(θt)=θt)=q∈(1/2,1)$⁠. The government in period $t−1$ chooses a censorship policy from the signal $s(θt)$ into a message $mt∈{0,1}$ observed by the voter. More precisely, the government chooses the censorship policy $ct:{0,1}→[0,1]$⁠, where $ct(s)$ is the probability that the voter observes a period-$t$ message $mt=1$ when $s(θt)=s$⁠.²² Without loss of generality, we focus on censorship policies $ct$ such that $ct(1)≥ct(0)$⁠. If $gt−1=i$⁠, an illiberal government strategically chooses the censorship policy. If $gt−1=ℓ$⁠, then the liberal government is constrained by the constitution and cannot interfere with the level of transparency guaranteed by it. Formally, the liberal government chooses a censorship policy $ct=cℓ$ that fully reveals the signal $s(θt)$⁠, i.e.  $cℓ(s)=s$ for all $s∈{0,1}$⁠.

In each period $t$⁠, the timing of the events is as follows. First, $θt$ is realized, and the voter observes $ct$ and $mt$⁠. Second, the voter chooses whether to elect $gt=i$ or $gt=ℓ$⁠. Finally, the period-$t$ government, $gt$⁠, chooses the censorship policy, $ct+1$⁠.

An illiberal government elected for the first time in period $t$ receives a rent $R>0$ for period $t$ and each subsequent period until the first period $t′>t:gt′=ℓ$⁠, at which point a liberal government is elected and the previous illiberal government is replaced by a new illiberal government. The voter and the illiberal government maximize their present discounted value of their payoffs and discount future periods with factor $δ∈(0,1)$⁠.²³

3.3 Welfare

We note here that social welfare in our model is not fully captured by $v(gt,θt)$ as this only represents the utility of the pivotal voter, who may trade-off the liberty of a minority for a greater security of her own economic interests. In Section 7, we extend the model to allow for a continuum of voters with differing values of liberty and security. In this case, the pivotal voter is the one with the median relative value of liberty, $L/S$⁠. Therefore, the single voter in our benchmark model should be interpreted as representing this median voter. In Section 7, we will return to this distinction and discuss how different members of a polity are differently affected by changes in the elements in its constitution.

3.4 Equilibrium concept

We characterize the perfect Bayesian equilibria (Fudenberg and Tirole, 1991) of our model in which the voter and the illiberal governments play pure Markovian strategies. Following the Bayesian persuasion literature (Kamenica and Gentzkow, 2011), we focus on equilibria in which the voter chooses an illiberal government whenever she is indifferent. Because $θt$ is i.i.d., at time $t$⁠, the voter’s payoff relevant history is fully captured by $ct$⁠. Similarly, in period $t$⁠, the illiberal government’s payoff relevant history is fully captured by $gt$⁠. However, the illiberal government only takes action if $gt=i$⁠. Therefore, for any period $t$⁠, the illiberal government’s set of optimal censorship policies is time independent. For simplicity, we focus on equilibria where the illiberal government’s choice of censorship policy is also time independent.²⁴

A Markovian pure strategy for the voter is a mapping $g$ from the message observed by the voter, $mt$⁠, and the censorship policy, $ct$⁠, into the period-$t$ government, $gt$⁠. A Markovian pure strategy for the period-$t$ illiberal government is a choice $c$ of censorship policy, $ct+1$⁠. The voter’s belief that $θt=1$⁠, denoted by $μt$⁠, is a mapping from the message observed by the voter, $mt$⁠, and the censorship policy, $ct$⁠, into a probability. A Markovian assessment is therefore a triple $σ=(g,c,{μt}t=1∞)$⁠.

Let

and

denote the expected continuation payoffs from electing a liberal and illiberal government, respectively. A strategy $g*$ for the voter is sequentially rational, given $σ$⁠, if in each period $t$⁠, the voter’s choice of government maximizes her expected payoff, i.e.

A strategy $c*$ for the illiberal government is sequentially rational, given $σ$⁠, if in each period $t$ in which $gt=i$⁠, the illiberal government’s choice of censorship maximizes (the discounted value of) their expected time in power

I.e. for all censorship policies $c′$⁠,

Finally, in equilibrium, in each period $t$ the voter’s belief $μt$ is derived using Bayes’ rule,²⁵  i.e.

and

Equilibrium

 

In the following sections, we characterize the essentially unique equilibrium of the model.²⁶ All proofs appear in Supplementary Appendix A.

4 THE OPTIMAL VOTING AND CENSORSHIP STRATEGIES

We begin by characterizing the voter’s sequentially optimal strategy. In each period $t$⁠, the voter observes a message, $mt$⁠, about $θt$ and forms a belief about whether $θt=1$⁠. She then decides whether to elect an illiberal or a liberal government. In period $t$⁠, electing a liberal government gives more liberty, $L$⁠, but has an expected cost of security, $μt(mt,ct)S$⁠; electing an illiberal government always provides a payoff of zero. Therefore, absent dynamic considerations, the voter elects an illiberal government whenever the expected cost of security is greater than the value of liberty. However, electing an illiberal government comes at an additional dynamic cost for the voter. As an illiberal government can engage in censorship, electing an illiberal government today induces the voter to take a less-informed choice of government tomorrow. This means that the expected continuation payoff of electing an illiberal government, $V(gt=i∣σ*)$⁠, is smaller than the expected continuation payoff of electing a liberal government, $V(gt=ℓ∣σ*)$⁠. We define the (equilibrium) accountability cost of illiberalism as

We will show that the accountability cost of illiberalism is decreasing in the informativeness of the censorship policy that the voter expects the illiberal government to choose. In particular, the maximum accountability cost of illiberalism is²⁷

This maximal cost may be achieved if an illiberal government chooses a full censorship policy $cF:={c:c(s)=c¯∈[0,1]∀s∈{0,1}}$ and is able to ensure its own reelection.

Lemma 1 says that in equilibrium the voter elects an illiberal government if and only if the expected cost of security, $μt(mt,ct)S$⁠, is greater than the sum of the value of liberty, $L$⁠, and the discounted accountability cost of illiberalism, $δA(π,q,L,S,δ∣σ*)$⁠.

The voter’s optimal strategy

 

We now turn to the illiberal government’s problem. Lemma 2 establishes that the illiberal government’s problem can be reduced to the problem of choosing the censorship policy that maximizes their probability of being reelected in the following period.

The illiberal government’s problem

 

In what follows, we characterize the illiberal government’s optimal strategy. Lemma 3 says that, in equilibrium, illiberal governments are either indifferent between all censorship policies, or choose one of two types of censorship policies.

Optimal censorship

 

To understand Lemma 3, it is easier to focus on the censorship policies that always send signal $mt=1$ whenever $s(θt)=1$⁠, i.e.  $c(1)=1$⁠. Point 1 of Lemma 3 considers the case when the voter reelects the illiberal government absent any information about the state (i.e. the equilibrium expected cost of security is greater than the value of liberty plus the discounted maximum accountability cost). In this case, it is optimal for the illiberal government to choose $c(0)=1$—the full censorship policy.

Otherwise, the illiberal government engages in partial censorship: it chooses $c(0)<1$⁠. The precise probability is the one that induces a voter’s belief $μt*(1,c)$ such that, upon observing message $mt=1$⁠, the voter is indifferent between electing a liberal or an illiberal government for the next period. The intuition behind the optimal partial censorship is akin to that in the Bayesian persuasion literature.²⁸ Intuitively, greater censorship of unfavourable signals (a greater $c(0)$⁠) increases the chances of reelection for the illiberal government up to the point at which the voter would correctly infer that, even upon observing message $mt=1$⁠, a liberal government is preferable.

However, there is a key difference between the role of censorship in our model and the one it plays in the Bayesian persuasion literature. In our model, greater censorship affects the voter’s optimal choice through two distinct channels. First, as mentioned above, more censorship directly increases the probability that the voter believes that today she would prefer illiberal policies to protect her from negative shocks. However, a second effect arises when the voter expects more censorship from illiberal governments. More (expected) censorship dynamically increases the accountability cost of illiberalism, increasing the voter’s value of a liberal system in which she can make a better-informed choice of government for the future. The trade-off between these two effects drives the results in the following sections.

5 REGIME STABILITY AND CYCLES

We now show that, depending on the constitution parameters $π$⁠, $L$⁠, and $q$⁠, the equilibrium behaviour of our model gives rise to four different regimes.

Efficient stable liberal democracy

In this regime, the voter always elects a liberal government, and this choice is optimal given the uncensored information offered by the constitution, $s(θt)$⁠. In equilibrium, an illiberal government chooses any censorship mapping $c*$⁠; the voter elects an illiberal government if and only if $μt*(mt,ct)S≥L$⁠; and $gt=ℓ$ for all $t≥1$⁠.

5 Efficient stable illiberal democracy

In this regime, the voter always elects an illiberal government. Yet, her choice of government is optimal given the uncensored information offered by the constitution, $s(θt)$⁠. In equilibrium, an illiberal government chooses any censorship mapping $c*$⁠; the voter elects an illiberal government if and only if $μt*(mt,ct)S≥L$⁠; and $gt=i$ for all $t≥1$⁠.

5 Inefficient stable illiberal democracy

In this regime, the voter eventually (but not necessarily at $t=1$⁠) elects an illiberal government that remains in power forever. Crucially, absent censorship, the voter would find it optimal to cycle between liberal and illiberal governments. However, once she elects an illiberal government, the illiberal government chooses a full censorship policy and ensures its own reelection forever. To completely characterize the equilibrium, under a liberal government, the voter elects an illiberal government if and only if $mt=1$⁠, which occurs with probability $πq+(1−π)(1−q)$⁠. Once an illiberal government has been elected, it chooses the full censorship policy, $cF$⁠, or any other censorship policy $c*$ satisfying

and the voter reelects the illiberal government with probability one.

5 Cycling liberal-illiberal democracy

In this regime, the voter cycles between liberal and illiberal governments. In equilibrium, in any period $t$⁠, the voter elects the illiberal government if and only if $mt=1$⁠, and the illiberal government engages in partial censorship, $c*$⁠. Therefore, if $gt−1=ℓ$⁠, the voter elects the illiberal government with probability $πq+(1−π)(1−q)$⁠, and if $gt−1=i$⁠, the illiberal government falls with probability $(1−c*(0))(π(1−q)+(1−π)q)∈(0,1)$⁠. However, this cycling liberal-illiberal regime is still inefficient: absent censorship, the voter would optimally revert back to a liberal government with strictly higher probability.

Proposition 1 describes when each of these regimes arises in equilibrium.

Regime stability and cycles

 

Proposition 1 says that if the value of liberty, $L$⁠, is sufficiently large, while the cost of security, $S$⁠, the constraints on the executive, $π$⁠, as well as the transparency of the institutions, $q$⁠, are sufficiently small, then a stable liberal democracy arises. Intuitively, under this constitution, the liberal government offers valuable liberty and sufficient security to the voter, so that the voter always prefers to elect a liberal government.²⁹ Conversely, if the value of liberty, $L$⁠, is sufficiently small, while the cost of security, $S$⁠, and the constraints on the executive, $π$⁠, are sufficiently large, but the transparency of the institutions, $q$⁠, is sufficiently small, then a stable illiberal democracy arises. Intuitively, under this constitution, the liberal government does not offer enough liberty nor security, and so the voter always prefers to elect an illiberal government.³⁰

In between these two extreme cases lie the two intermediate regimes. First, for some parameters the liberal government offers valuable liberty and security but, whenever the voter observes the period-$t$ message $mt=1$⁠, her expected cost of security is so high that she prefers to elect an illiberal government. She correctly anticipates that the decision to elect an illiberal government is forever: because of censorship she will continue to elect an illiberal government even though, absent censorship, she would optimally revert back to a liberal government with probability $π(1−q)+(1−π)q$ in each period. In fact, the voter correctly attributes to an illiberal government the maximal accountability cost, $A¯$⁠. Therefore, an inefficient stable illiberal democracy arises only if the expected cost of security under full censorship is greater than the sum of the value of liberty and the discounted maximal accountability cost:

When this condition is not satisfied, an illiberal government cannot ensure its own reelection. However, if anything, this increases the voter’s propensity to elect an illiberal government in the first place because illiberalism now comes with a smaller accountability cost. This gives rise to cycles of liberal and illiberal governments. Under this constitution, whenever the voter observes the period-$t$ message $mt=1$⁠, her expected cost of security is so high that she prefers to elect an illiberal government. However, because the voter values liberty sufficiently, the illiberal government is unable to ensure its reelection and the full censorship policy is suboptimal. Instead, the illiberal government optimally chooses a partial censorship policy, $c*$⁠, that induces the voter to reelect the illiberal government only if she observes the message $mt=1$⁠. Therefore, the illiberal government falls with positive probability equal to $(1−c*(0))(π(1−q)+(1−π)q)$⁠.

For any given value of security $S$⁠, Proposition 1 maps constitutions (parameters $π$⁠, $L$⁠, and $q$⁠) into regimes. It is useful to normalize the value of liberty as the voter’s relative value of liberty, $L¯:=L/S$⁠. Corollary 1 says that, as the voter’s relative value of liberty increases, the regime transitions from an efficient stable illiberal democracy to an inefficient stable illiberal democracy to a cycling liberal-illiberal democracy, and then finally to an efficient stable liberal democracy.

A typology of regimes

 

The closed form of the cut-offs is given in Proposition 1 in Supplementary Appendix A. Panels (a) and (b) in Figure 1 illustrate this regime typology in the $(π,L¯)$ and $(q,L¯)$ space, respectively. In each figure, the plotted curves from bottom to top correspond to $κ_(π,q)$⁠, $κ(π,q,δ)$⁠, and $κ¯(π,q)$⁠, respectively. As Panel (b) shows, for a given $π$ an increase in $q$ can change the regime from a stable one to an unstable one. Intuitively, more precise information for the voter implies that the voter is more likely to change which government she prefers upon observing a new signal. This key intuition drives many of the results in Section 6, which focuses on the stability of constitutions by studying how the different parameters affect the rise and fall of illiberal democracies.

5.1 Exploring the boundaries of the model

We now briefly pause our analysis of the rise and fall of illiberal democracies to return to the general problem we have identified in a principal’s choice to hire an agent to manage a project. Doing so allows us to clarify the implications of some of our assumptions.

We have thus far studied one possible version of the problem. In our benchmark model, the agent has the ability to commit to a censorship policy, but only for one period. In the context of our main application, we believe it is natural to think that the illiberal government sets, at least for some time, policies that intimidate media and bureaucrats, but cannot commit to further relaxations or restrictions in a more distant future. However, in other contexts it may be more realistic to assume that the agent can commit to a multi-period policy of information design. In Supplementary Appendix C, we study a variation of the model in which the government commits to long-term censorship policies and we show that the key insights and the characterization of the regimes are very similar to those in Section 5.

The Bayesian persuasion assumption that the agent commits to an information design rather than choosing what message to send is analytically convenient. In our context, this assumption also has important elements of realism, as in practice governments establish policies and intimidate media and whistleblowers with the objective of limiting information leaks; however, they are limited in how much they can conceal information once it has been leaked. In other contexts, this assumption may be less realistic, for example if the principal only receives information directly from the agent. For completeness, we study in Supplementary Appendix D a non-commitment version of our model. The results mimic very closely those we present in Section 5. However, in contrast with our benchmark model, when in equilibrium the voter cycles between liberal and illiberal governments, she does so inefficiently because she reelects the illiberal government too infrequently (instead of too frequently). Intuitively, when the agent’s communication is cheap talk, the agent is unable to convey any decision-relevant information to the principal. Therefore, in equilibrium, the principal expects the agent to—in effect—censor all information. Thus, she either always retain the agent because it is optimal absent any period-specific information, or she fires him after just one period. This is inefficient: if the principal could access all the available information, she would retain the agent (on average) whenever $s(θt)=1$⁠.

As we discussed, it is essential to our trade-off of interest that the voter (principal) may prefer to return to a liberal government (fire the agent) so to observe more precise information. For completeness, we study in Supplementary Appendix E a version of the model in which the ability to manipulate information is independent of the ability to provide security and both liberal and illiberal governments can censor strategically. This partially mutes our key trade-off. Therefore, in equilibrium we either have stable liberal democracies or stable illiberal democracies. Said otherwise, since the principal has no way to see things with her own eyes, once she decides to hire an agent, there is no value for her in changing her mind again.³¹

Our benchmark model assumes that the underlying state $θt$ is independently distributed across periods. Essential to our trade-off of interest is that the state is not fully persistent. Otherwise, our model would be a stopping game of learning for the principal. Realistically, shocks to countries and firms alike are partially persistent. In this case, our main trade-off of interest still drives behaviour; however, its effects are compounded by other dynamics. For example, under a liberal government, upon observing the realization of a negative shock at time $t$⁠, the voter’s belief regarding the probability of a negative shock at time $t+1$ would increase, therefore increasing her likelihood to elect an illiberal government for any message observed at time $t$⁠. Furthermore, the illiberal government’s ability to censor would also depend on how much time it has been in power and the whole history of messages observed by the voter under its regime.

6 THE RISE AND FALL OF ILLIBERAL DEMOCRACIES

This section studies how different parameters of our model affect the dynamics of evolution of democracies. In particular, we ask what favours the rise of illiberal democracies, how different constitutions affect the type of censorship illiberal governments engage in, and, finally, what induces illiberal democracies to fall sooner, if at all.

6.1 The rise

We begin by studying what causes the rise of illiberal democracies from a liberal government. An immediate implication of Proposition 1 is that, unless the constitution induces an efficient stable liberal or efficient stable illiberal democracy, an illiberal government is elected whenever $s(θt)=1$⁠. When this happens, the probability that the period-specific shock is one that the voter cannot be protected from by a liberal government (and therefore the expected cost of security) is high. This can be thought of as a crisis that increases the voter’s perception that their welfare is under threat from terrorists, foreign powers, recessions, or a pandemic. The probability that such a crisis arises, as well as whether the voter is willing to elect an illiberal government during a crisis, depends on the different elements of the constitution.

Proposition 2 shows how the different elements of a constitution $π$⁠, $L$⁠, and $q$ affect the probability that an illiberal government is elected following a liberal one. First, by Corollary 1, the probability that an illiberal government is elected equals $1$ when the relative value of liberty is sufficiently small, and zero when it is sufficiently large. Otherwise, in inefficient stable illiberal democracies and in cycling democracies, an increase in the relative value of liberty induces voters to prefer a liberal government more, as the value of liberalism is greater. So the probability that an illiberal government follows a liberal one (weakly) decreases in the relative value of liberty. Second, more constraints on the executive (a greater $π$⁠) increases the expected cost of security of choosing a liberal government for all levels of transparency, $q$⁠. Therefore, more constraints on the executive induce the voter to prefer an illiberal government more. So the probability that an illiberal government follows a liberal one (weakly) increases with the amount of constraints placed on the executive by the constitution. Finally, the transparency of the liberal constitution also plays a role in the voter’s choice. Proposition 2 says that whether greater transparency increases or decreases the probability of an illiberal government depends on the other parameters of the constitution as well as on the amount of transparency itself. These non-monotonicities are caused by the fact that, as shown in Panel (b) of Figure 1, an increase in $q$ can cause a switch from a stable to an unstable regime. For example, Proposition 2 says that when $π<L¯$⁠, increasing $q$ has no effect on the probability of a rise of an illiberal democracy for sufficiently small $q$⁠, because the regime is a stable liberal democracy (the probability of a transition is zero). However, after $q$ passes a threshold, the regime becomes a cycling liberal-illiberal democracy, and in the case in which $π<1/2$⁠, the probability of a liberal-illiberal transition is positive and maximal precisely at this threshold and then decreases for larger $q$⁠, so that the total effect of $q$ on the probability of a rise of an illiberal democracy is single-peaked.

The rise of an illiberal democracy

 

Panels (a) and (b) in Figure 2 illustrate the relationship between the illiberal government’s election probability and the transparency of liberal institutions, $q$⁠, under two parametric assumptions about $π$⁠. This figure captures all the cases discussed in Proposition 2.

Intuitively, Point 1 of Proposition 2 refers to a constitution in which there are few constraints on the executive and the voter values liberty very much. Then, absent any information about the period-specific shock, the voter would always elect a liberal government (efficient stable liberal regime). The voter would resort to electing an illiberal government only if she were to observe sufficiently precise information about the current period’s shock. Therefore, if the signal $s$ is sufficiently precise (large $q$⁠), the regime transitions to a cycling democracy and she elects an illiberal government whenever she observes $mt=1$⁠. Because $π<1/2$⁠, a more precise signal sends message $mt=1$ with lower probability, and therefore the probability of electing an illiberal government is strictly decreasing in $q$⁠.

Point 2 (Point 3) refers to a situation in which there are few (many) constraints on the executive and the voter’s value of liberty is smaller (greater) than the expected cost of security under full censorship, $πS$⁠. In this case, the voter always prefers to elect an illiberal (liberal) government when her information is sufficiently imprecise. However, the voter would rather elect the illiberal government when the current period’s shock requires illiberal means of protection (⁠$θt=1$⁠) and otherwise to elect the liberal government. Therefore, if the signal $s$ is sufficiently precise, the regime transitions to one in which the voter elects an illiberal government if and only if she observes $mt=1$⁠. Because $π<1/2$ (⁠$π>1/2$⁠), a more precise signal sends message $mt=1$ with lower (higher) probability, and therefore the probability of electing an illiberal government is strictly decreasing (increasing) in $q$⁠.

Point 4 is a mirror image of Point 1: the voter would elect a liberal government only if she were to observe sufficiently precise information that the current period’s shock does not require illiberal means of protection. For more informative signals, because $π>1/2$⁠, the probability of electing an illiberal government is strictly increasing in $q$⁠.

6.2 The reign: optimal censorship

We now turn to studying how different elements affect an illiberal government’s censorship policy. Recall from Lemmas 2 and 3 that, in stable illiberal democracies, the illiberal government is indifferent among many censorship policies, including the full censorship policy, $cF$⁠. In this section, we focus on the equilibrium where the illiberal government chooses the full censorship policy whenever it is optimal. Without loss of generality, we identify the full censorship policy, $cF$⁠, with the censorship policy that always sends $mt=1$⁠.

Proposition 3 says that greater constraints on the executive (greater $π$⁠) increases the expected cost of security for all censorship policies, and therefore allows an illiberal government to engage in more censorship (greater $c*(0)$⁠). In contrast, a greater relative value of liberty (greater $L¯$⁠) forces an illiberal government to engage in less censorship. However, the effect of the transparency of liberal institutions, $q$⁠, on the optimal amount of censorship depends on the other parameters of the constitution.

The reign of an illiberal democracy

 

As the transparency of liberal institutions increases (greater $q$⁠) there are two effects: a persuasion effect and an accountability effect. Intuitively, more transparent institutions increase an illiberal government’s ability to persuade the voter to reelect the government. More precisely, when an illiberal government engages in the optimal partial censorship, upon observing message $mt=1$⁠, the voter is indifferent between electing a liberal or an illiberal government for the next period. For a given censorship policy, a more transparent liberal institution (greater $q$⁠) provides the voter with greater incentive to elect an illiberal government upon observing message $mt=1$⁠. Therefore, this persuasion effect induces an illiberal government to increase the amount of censorship when the liberal institutions are more transparent. This effect is common in the Bayesian persuasion literature (Kamenica and Gentzkow, 2011; Kolotilin et al., 2019).

In our model, more transparent institutions also affect an illiberal government’s censorship through an accountability effect. Recall from Lemma 1 that the voter’s choice to reelect the illiberal government depends whether the expected cost of security exceeds the sum of the value of liberty and the accountability cost of illiberalism. Naturally, when liberal institutions are more transparent, the accountability cost of illiberalism is greater because returning to a liberal government allows the voter to make even more informed decisions in the future. To keep the voter indifferent, an illiberal government must provide the voter with greater information to compensate them for the greater accountability cost. Therefore, this accountability effect induces an illiberal government to decrease the amount of censorship when the liberal institutions are more transparent.

Proposition 3 says that the persuasion effect dominates when the constitution has few constraints on the executive and the voter values liberty so much that, absent any information about the period-specific shock, the voter would always elect a liberal government. Otherwise, the accountability effect dominates.

6.3 The fall

We now turn to studying what causes illiberal governments to fall if they do at all. Because of censorship, the voter is unable to cause the fall of an illiberal government as soon as there is no crisis, i.e. as soon as $s(θt)=0$⁠. Instead, the only thing that may trigger a fall of an illiberal democracy is the illiberal government’s failure to conceal that there is no crisis: $mt=0$⁠. Therefore, in cycling democracies and in inefficient stable illiberal democracies, illiberal governments remain in power for longer than the voter would optimally choose in the absence of censorship.

We now show what factors increase the probability that an illiberal government stays in power. Proposition 4 says that an illiberal government’s reelection probability is increasing in the constraints on the executive, $π$⁠; decreasing in the relative value of liberty, $L¯$⁠; but not necessarily monotonic in the transparency of liberal institutions, $q$⁠.

The fall of an illiberal democracy

 

To understand how the transparency of liberal institutions affects the ability of an illiberal government to be reelected, it is useful to write the explicit formula for the derivative of the illiberal government’s reelection probability with respect to $q$⁠. Recall that, in equilibrium, the voter reelects an illiberal government whenever she observes $mt=1$⁠; therefore, this formula is

This formula highlights that there are two effects of transparency on the illiberal government’s reelection. First, there is a mechanical component whereby, for a given censorship policy, the transparency of the liberal institutions affects the probability that the voter observes message $mt=1$ because the institutions determine the likelihood of an underlying signal $s(θt)=1$⁠. The direction of the mechanical effect depends on the constraints on the executive, $π$⁠. When $π<1/2$ (⁠$π>1/2$⁠), a more transparent institution produces an underlying signal $s(θt)=1$ with lower (higher) probability and, all else equal, the illiberal government’s reelection probability decreases (increases) with $q$⁠. Second, there is a strategic component whereby the transparency of the liberal institutions affects the optimal censorship policy of an illiberal government. By Proposition 3, when $π<L¯$ (⁠$π>L¯$⁠) the persuasion (accountability) effect determines the sign of the strategic component, and the illiberal government’s reelection probability increases (decreases) with $q$⁠.

Point 1 (Point 2) of Proposition 4 refers to a constitution where the persuasion (accountability) effect determines the sign of the strategic component and the mechanical component operates in the same direction; therefore, the illiberal government’s reelection probability monotonically increases (decreases) with $q$⁠. In the remaining cases, the two components have opposite signs. As illustrated by Figure 3, for low values of $q$⁠, the strategic component dominates but, for high values, the marginal effect of $q$ on the optimal censorship policy vanishes and, therefore, the mechanical component dominates.

7 CONSTITUTIONAL DESIGN AND LIBERAL STABILITY

We now turn to the problem of the optimal design of a constitution. As we noted, welfare in our model is not necessarily captured by the utility of the pivotal voter. However, we can study what constitution the pivotal voter would choose if she had to pick one within the possibility frontier given by the functions $L$ and $Q$⁠. We focus on the long-run payoff of the pivotal voter.³² By choosing a constitution, the voter also selects which of the four regimes in Proposition 1 will be sustained. Proposition 5 says that, unless the technological constraints only afford the voter constitutions that sustain stable illiberal democracies,³³ she always prefers constitutions that sustain either a stable liberal democracy or a cycling democracy.³⁴

The voter’s optimal constitution

 

Beyond the pivotal voter’s welfare, the focus of our paper is on which liberal democratic constitutions are stable. One immediate implication of our analysis is that excessive constraints on the executive (high $π$⁠) necessarily lead to illiberalism. Intuitively, excessive constraints render liberal governments ineffective and induce voters to elect illiberal governments that promise greater economic security, efficient public services, and protection from domestic and international violence. This result resonates with the literature that roots the rise of illiberal governments in the ineffectiveness of liberal democratic institutions (Berman, 2019; Matthijs and Blyth, 2018; see also Rodrik, 2000). Howell and Moe (2020) suggest that constitutional reforms that give greater power to the U.S. executive (and the president, in particular) are needed to safeguard the future of American liberal democracy. However, reducing constraints on the executive is not without costs: more constraints on the executive are needed to guarantee greater individual freedoms. This is captured by the constitutional possibility constraint $L=L(π)$⁠. Therefore, our framework allows us to derive implications for the design of the optimal stable liberal constitution: a constitution that maximizes liberty, $L(π)$⁠, without causing instability, i.e. subject to

The optimal stable liberal constitution

 

Intuitively, Proposition 6 says that the optimal stable liberal constitution needs to limit how much liberty it affords to the voter. In fact, more liberty has both a direct and an indirect detrimental effect on liberal stability. More liberty directly requires more constraints on the executive, thus increasing the expected cost of security of electing a liberal government, increasing the voter’s propensity to elect an illiberal government. Moreover, because more liberty requires greater constraints on the executive, it also indirectly implies that media and state bureaucracy are more transparent (because $Q$ is increasing). This increases the precision of the voter’s information, increasing her propensity to elect an illiberal government when she observes information that suggests that the liberal government may be unable to adequately respond to this period’s shocks.

Point 2 in Proposition 6 also provides some insights into the effects of technological changes that shift the constitutional possibility frontier. For example, technological changes that affect the quality of information observed by voters and increase $Q(π)$ for all $π$⁠. The right-hand side of (11) is increasing in the transparency of liberal institutions, $q=Q(π)$⁠. Therefore, the optimal constraints on the executive, $π*$⁠, will crucially depend on $Q$⁠. Proposition 7 says that the arrival of technologies that increase the transparency of liberal democratic institutions must be accompanied by reductions in the constraints on the action of the executive. Otherwise, voters will demand stronger illiberal governments.³⁵

The paradox of liberal transparency

 

The possibly detrimental effects of greater transparency are most evident when considering the welfare implications of different constitutions. To make ideas precise, suppose there is a continuum of voters, $j$⁠, indexed by their relative value of liberty, $L¯j$⁠, and decisions are made via majority rule. Then the pivotal voter is the voter with median relative value of liberty. Therefore, the single voter in our model corresponds to this median voter. Any increase in transparency, including a marginal increase that moves the constitution from a stable liberal democracy to a cycling liberal-illiberal democracy, benefits the median voter. However, it can be shown that this same marginal increase in transparency harms every voter with a higher relative value of liberty and, therefore, may reduce aggregate welfare.

8 PATHS TO AUTOCRACY AND DEMOCRATIC STABILITY

A major concern expressed by political leaders, scholars, and commentators is that illiberal democracies may constitute a first step towards autocracy (see, e.g. Lührmann and Lindberg, 2019; Rhodes, 2020). In reality, once an illiberal government rises to power, its ability to operate beyond constitutional constraints may not be restricted to information manipulation. We see two possible paths through which illiberal governments may build an autocracy. A first path is through the manipulation of the liberty-security trade-off. A second path is through the direct repression of electoral accountability.

8 Manipulating security

We extend our model to allow the illiberal government to manipulate the voter’s value of security. In particular, in each period that the illiberal government is in power, with probability $ζ$⁠, it can choose to increase the value of security, $S$⁠, to $S′$ such that

For example, the illiberal government can engage in aggressive diplomacy with a foreign power or exacerbate internal conflicts with separatists. From the point of view of the voter, this increases the expected security cost of electing a liberal government which cannot employ illiberal measures to protect the voter from foreign invasions or domestic terrorism. Condition 12 says that the increased value of security is large enough to induce the voter to forever prefer an illiberal government, thus inducing a de facto autocracy (see Proposition 1).

8 Repressing accountability

We extend our model to allow the illiberal government to directly repress electoral accountability, thus seizing power and establish an autocracy. Formally, in each period that the illiberal government is in power, with probability $γ$⁠, it can choose to overthrow the constitution and seize power. If the government chooses to overthrow the constitution, from that period onward, there is no further election and the illiberal government is in power forever.

In equilibrium, illiberal governments obviously always choose to seize a chance to manipulate security or repress accountability. Therefore, both extensions imply that whenever the voter elects an illiberal government, there is a positive probability that this will induce an autocracy. In the long run, this means that a constitution either induces a stable liberal democracy, or it will result in an autocracy.³⁶ The greater the probability that an illiberal government can build an autocracy (greater $ζ$ or $γ$⁠), the sooner any non-stable liberal democracy becomes an autocracy. However, Proposition 8 says that a greater probability that an illiberal government can build an autocracy allows for a greater set of constitutions to induce a stable liberal democracy.

Long-run democratic stability

 

Intuitively, when electing an illiberal government comes with a greater risk of a total loss of accountability, the voter is more wary of the potential consequences of such a decision. Therefore, she elects an illiberal government only if the expected cost of security is very large. Thus, the set of constitutions that allow for a stable liberal democracy to emerge is greater.

We can interpret the parameters $ζ$ and $γ$ as measuring the weakness of checks and balances preventing total executive takeover. Thus, our result says that stronger checks and balances, while reducing the speed at which illiberal democracies may turn into autocracies, may actually induce democracies to turn illiberal with greater probability, and in the long run lead to autocracy.

Importantly, two types of long-run regimes may arise in equilibrium. First, some stable liberal democracies may be inefficient. For some constitutions, absent the risk that illiberal governments can build an autocracy, the voter would prefer to cycle between liberal and illiberal governments. However, because illiberal democracies may turn into autocracies, the voter refrains from electing an illiberal government and instead prefers to endure the full cost of security of liberalism. This happens when the liberal constraints on the executive are intermediate.

Second, some long-run autocracies are also inefficient for the median voter. For some constitutions, the voter prefers to cycle between liberal and illiberal governments. However, she chooses illiberalism conscious that this may lead with positive probability to a long-run autocracy, which on average offers too much security and too little liberty.

8 Informational legacy of illiberalism

Another realistic dynamic effect of illiberalism is that information manipulation may have long-lasting effects on the quality of some institutions of control. For example, a censored bureaucracy may find it hard to rebuild capabilities and experience after an illiberal spell. Similarly, intimidated media may not recover immediately after the fall of an illiberal government. In our model, this amounts to a permanent fall in $q$ following an illiberal government. Notice that this means that the accountability cost of illiberalism is no longer constant: it is large before the voter ever elects an illiberal government, but it is small (in some cases even equal to zero) once the illiberal government rises to power. As a consequence, the voter is more wary of electing an illiberal government, therefore reducing the probability of a rise of an illiberal democracy, but she also is less likely to return to liberalism once she has elected an illiberal government, reducing the chances of a fall of an illiberal democracy.

9 EMPIRICAL RELEVANCE

The focus of our theory is at the voter’s level. In particular, we offer a framework to think about variation in support for liberalism even in consolidated democracies, which in turn may lead to the rise of an illiberal democracy. Our key idea is that a greater belief that individual welfare is under threat should come with lower support for liberal aspects of democracy but not necessarily with an outright rejection of democracy. We now argue that this premise is consistent with real-world data.

As we noted earlier, Graham and Svolik (2020) and Svolik (2019) show experimentally that voters are willing to trade-off liberty with security. In the context of the recent pandemic, Alsan et al. (2020) show that voters who are more exposed to the risk of the pandemic are more willing to impose limitations to individual freedoms to mitigate the spread of the virus. In a recent field experiment in Turkey, Baysan (2022) shows that conservative voters who are exposed to information about the recent increase in terrorist activity are more likely to vote in favour of lowering the accountability of the president.

We use data from Wave 6 of the World Values Survey (2010–14), focusing our attention to respondents in fully consolidated democracies, i.e. countries scoring a 10 out of 10 in the Polity5 democracy score in 2010. We study how different fear factors correlate with support for liberal elements of democratic constitutions. We use two important variables capturing fear: the fear of terrorist attacks and the fear of losing employment (or not finding employment).³⁷ We measure (positive or negative) support for liberal aspects of democratic constitutions via questions regarding support for the protection of civil rights (positive), and for “having a strong leader who does not have to bother with parliament and elections” (negative).

Figure 4 shows that fears of terrorist attacks and losing employment strongly correlate with lower support for the protection of civil rights in democracy and greater support for a strong leader. However, the bottom panels of Figure 4 show that fears of terrorism or of losing employment do not have a similarly clear impact on support for democracy in general. All these results replicate in linear regressions with country fixed effects; all relations have the expected direction and are highly statistically significant, bar for the relation between fear of terrorism and support for democracy. The results also extend to including all democracies (Polity5 score of 6–10) but the effect of fear on civil rights protections is lost, perhaps because the importance of civil rights protection is perceived as more fundamental among citizens of democracies with very marginal scores.³⁸

Albeit far from establishing any causal relation, the data support the view that individual fears are connected with lower support of key liberal aspects of democratic constitutions. Even in Western consolidated democracies, economic insecurity maps into much stronger preferences for governments that do not need to bother with the procedural aspects of democratic decision-making. These results naturally map into the mechanics of our model. Upon observing information suggesting that individual welfare is under greater threat, a voter’s expected cost of security rises, so that she may more likely prefer an illiberal government: one that provides less liberties, such as the protection of civil rights, but provides security through the effective action of a strong and less constrained leadership.

Our theory offers a mechanism through which these individual tendencies towards illiberalism become consequential: if the median voter prefers an illiberal government, she will elect one, leading the country to cycles of illiberalism, a stable illiberal democracy, or even to an autocracy. Survey data may not necessarily represent consequential preferences—perhaps only reflecting the respondents’ frustration with the complications of democracy. However, even within our sample of consolidated democracies, country-average support for a strong leadership negatively correlate with the v-Dem measure of liberal democracy in 2010 (p-value: 0.023), suggesting that survey data, at least about the support for strong leaders, capture a substantial phenomenon that affects both voters’ preferences and the action of governments.³⁹ Therefore, we use this variable as a measure of voters’ support for illiberalism.

A distinctive feature of our model is that too many constraints on the executive aimed at guaranteeing liberty may induce worried voters to prefer an illiberal government. More specifically, our theory predicts that fears of terrorism, war, or economic insecurity should translate into preferences for illiberalism more in countries that have more stringent constraints on the executive. Following Besley and Persson (2016) and Besley and Mueller (2018), we measure executive constraints using a dummy variable which is equal to one when a country in 2010 receives the highest score (7) on this basis on the Polity5 data. In order to observe variance in this measure, we study the sample of all democracies (a score of 6 or more in the Polity5 democracy measure). We report in Table 1 the results of regressing our preferred measure of support for illiberalism (respondents’ support for a strong leader) on our measures of fear, the executive constraints dummy, and the interaction term between fear and executive constraints. Consistently with our theory, the effect of fear on preferences for illiberalism are larger—and in fact exist only—in countries that have stronger constraints on the executive.

Our key premise that insecurity—and especially economic insecurity—drives voters to support illiberal governments resonates with empirical data. Mounk (2018) argues that the stagnation of median household income since the 1980s is one of the three main causes of rising support for illiberal democracies. Rothwell and Diego-Rosell (2016) and Kolko (2016) document that voters’ support for Trump in the 2016 election did not reflect poor economic conditions; rather it reflected economic vulnerability to shocks in automation, immigration, globalization, and offshoring—in our words, voters who feared for their economic security. Much attention has been given to the recent rise of populist parties in Europe. Some of these populist parties (but not all, as pointed out by Mudde and Rovira Kaltwasser, 2017) share the key features of our illiberal governments: they promise economic security to the median voter by disrespecting constitutional (and supranational) constraints on the executive. Guiso et al. (2019, 2020) document that the rise of populist parties in Europe resulted from mainstream parties and status quo institutions failing to manage the shocks faced by their economies. In line with our framework that too many constraints on the executive may exacerbate voter insecurity and induce demand for illiberal governments, Guiso et al. (2019) find that “the effects of globalization and the financial crisis on voting for populist parties in a European country crucially depends on whether or not such a country belongs to the Eurozone.” Belonging to the Eurozone, in fact, increases the institutional constraints on the fiscal action of governments, resulting in frustrated voters who turn to politicians that offer an alternative strategy to increase their economic security.

Our model also offers a framework to make empirical predictions related to the interaction between economic insecurity, information, and culture. In particular, economic insecurity is more (less) likely to spur demand for illiberalism when information about potential crises is more abundant if these crises are sufficiently recurrent (rare). Furthermore, economic insecurity is more likely to spur demand for illiberalism among voters who have a relatively low value of liberty, say because the cautionary tales of the hardship of dictatorship are in a distant past or because the voters inherit a local tradition of authoritarianism (a process of reactivation recently explored within the German context by Cantoni et al., 2019).

10 CONCLUSION

We proposed a theory of the rise and fall of illiberal democracies. We showed that, during times of crisis, voters may choose to elect illiberal governments that, by operating beyond the constraints of a liberal constitution, offer greater security and less liberty but may also manipulate information. We showed that the manipulation (or censoring) of information reduces effective electoral accountability. Therefore, the trade-off that we highlight for the voters is one between liberty, security, and accountability.

By focusing on the ability of illiberal governments to manipulate information, we drew insights into what makes a liberal constitution stable. First, the constitution must give sufficient powers to the government to address emergencies. In this sense, an ideal liberal constitution that perfectly protects individual liberties and rights is ineffective: it inevitably leads to illiberal governments that systematically infringe those same rights. Importantly, the effect of too many and too stringent constraints on the executive may be exacerbated by excessive transparency. In other words, in a world in which information leaks are unavoidable, a stable liberal democracy requires greater powers in the hands of the executive.

We also derived the conditions under which voters may resort to illiberal governments even when they expect such governments to be able to completely overthrow the checks and balances of democracy and dispense with elections altogether. The key insight from this analysis is that cycles of illiberalism are short-term symptoms of a long-run tendency to authoritarianism. Our results suggest that stronger checks and balances, while delaying the process of executive takeover, may nonetheless be counterproductive as they induce voters to be excessively optimistic about experimenting with short-term illiberal spells. If illiberal democracies are a step towards autocracy, then the only safety mechanism against them is to avoid them altogether by increasing the effectiveness of the government and possibly balancing executive constraints with transparency.

Our model also suggests a reason for why illiberal governments may be more popular in countries in which the constraints on the executive originate from supranational institutions. For example, voters in Poland, Greece, or Hungary may perceive that their governments are unduly constrained by decisions taken by the European Union or the International Monetary Fund. However, supranational institutions also guarantee stronger checks and balances. In our model, this combination of constrained sovereignty (high $π$⁠) and strong checks and balances (low $γ$ and $ζ$⁠) is the perfect storm for the rise of illiberalism.

Our model focuses on the popular demand for illiberalism in that it assumes that viable illiberal candidates always exist. Whether such candidates exist may also depend on supply driven factors such as the political and party system. Berman and Snegovaya (2019) and Rosenbluth and Shapiro (2018) argue the increase in demand for illiberal candidates is explained by recent trends in the design and ideology of political parties and that parties have a role in stemming the supply of illiberal candidates (see also Benedetto et al., 2020). Grillo and Prato (2023) show that even leaders who have no preferences for illiberal policies may partially adopt them to gain electoral advantages when voters have reference-dependent preferences. These supply driven factors may exacerbate the vulnerability of liberal constitutions to changes in voters’ demand for illiberalism.

Our results about the role played by transparency in liberal democracies may offer insights into how to interpret the relation between the expansion of internet, 3G networks, and social media and the rise of populism in Europe (Campante et al., 2018; Guriev et al., 2021). In our framework voters always hold consistent beliefs about the probability that their welfare is under threat. In reality voters’ beliefs may overreact to information, overestimate small probabilities (e.g. the probability of dying in a terrorist attack), and in general be on average inconsistent with reality (Alesina et al., 2018, 2020). Therefore, our results are likely to underestimate the probability that the median voter may resort to inefficient illiberal regimes, increasing the potential cost of constitutions that are more exposed to the menace of illiberalism.

Acknowledgments

We thank the editor and three anonymous referees for their constructive advice. We are also grateful to Dan Bernhardt, Steve Callander, Juan Carlos Carbajal, Pauline Grosjean, Richard Holden, Anton Kolotilin, Federico Masera, Massimo Morelli, Carlo Prato, Adam Szeidl, Galina Zudenkova, and participants at various seminars and conferences. Chris Burnitt and Jonathan Nathan provided useful research assistance. Gratton is the recipient of an Australian Research Council Future Fellowship (FT210100176). During part of this project, Lee was supported by a Fellowship by Examination (JRF) at Magdalen College, Oxford; he is grateful for his time there.

Supplementary Data

Supplementary data are available at Review of Economic Studies online.

Data availability statement

The data underlying this article are available in the Zenodo repository, at https://doi.org/10.5281/zenodo.7302150.

References

Footnotes

Author notes