The Effect of Referendums on Autocratic Survival: Running Alone and Not Finishing Second

Abstract

Dictatorships that use plebiscites – that is, referendums initiated by the executive – have a longer expected lifespan. A successful plebiscite sends a signal that makes coordination for collective action more difficult and induces the falsification of preferences. It also enhances the status of the dictator within the regime and against potential rivals through the use of agenda power. As a result, plebiscites are followed by decreased mobilization and a reduced risk of palace coups. This, in turn, adds years to the dictator's tenure. We have found evidence to support these propositions by utilizing data from various databases of autocratic regimes that span from 1946 to 2008.

Does the ‘appeal to the people’, a direct voting process on a question formulated by the ruler, help dictators to stabilize and extend their rule? We argue that plebiscites are indeed an effective means to enhance the power of the authoritarian executives.¹ They are effective on two fronts: the intrigues within the regime and the external opposition. Dictators who call for plebiscites suffer less internal violent contestation; and plebiscites are followed by periods of decreased mobilization against their rule. Hence, plebiscites contribute to authoritarian regime survival.

Country studies from all over the world provide scattered evidence that authoritarian referendums are instrumental in the consolidation of non-democratic regimes and authoritarian rulers. For example, there are studies on Azerbaijan (Dominioni 2017), Turkey (Kalaycıoğlu 2012), Venezuela (Kornblith and Jawahar 2005; Penfold 2010), Belarus (Silitski 2005) and Uganda (Bratton and Lambright 2001). Historically, holding plebiscites is a practice that can be found not only among modern dictators and would-be dictators, but also among enemies of constitutional democracy in every authoritarian wave in the past. Matt Qvortrup (2021) examines this issue and argues that democratic breakdown, as highlighted in his work, may not be fundamentally undemocratic. Attention is often drawn to cases in which plebiscites have been part of a democratization process, such as Spain (Sánchez-Cuenca 2014), Uruguay (Gillespie 1985; Rilla 1997) and Chile (Barros 2002). The last is perhaps the most iconic (and misleading) case, for it involved a celebrated miscalculation on the part of the dictator. On 11 October 1988, Fortín Diario, an oppositional newspaper to the Chilean regime, stated with a sardonic headline what had seemed to be true since the midnight after the vote: ‘Pinochet ran alone and finished in second place.’ What happened after the referendum six days earlier would seal Pinochet's regime and begin Chile's transition to democracy. This remarkable historical moment is also the most uncommon for a number of reasons. General evidence shows that what happened is more of a historical outlier than the actual norm. First, because when dictators call for a plebiscite, they almost never lose. Instead, what seems to be the common trait is winning prefabricated plebiscites with overwhelming majorities. And second, because plebiscites usually tend to lengthen the life of the regime, not shorten or terminate it. It is on this second point that we will focus our research.

Over 41% of the regimes categorized as authoritarian by Barbara Geddes et al. (2018) held at least one referendum, and over 54% if we shift from regimes to countries as units of analysis. Although referendums are not uncommon in dictatorships, the large-N comparative research on authoritarian referendums, as opposed to case studies, is relatively scarce. What is missing is a systematic analysis that addresses the question of the consequences of plebiscites for authoritarian rulers, on the stability of their regime, and on their longevity as dictators.

Previous quantitative and analytical research has mainly focused on why plebiscites are used by autocracies, rather than their consequences for the regime (Altman 2010; Penadés and Velasco 2022; Qvortrup et al. 2020). Qvortrup et al. (2020) focus on ‘repressive integrationist referendums’, which refers to plebiscites with an endorsement of over 99%. Their findings show that these referendums tend to happen in ethnically fragmented autocracies and sultanistic regimes, as those regimes need mechanisms of repression and affirmation towards the regime, while other regimes with established control apparatus, such as totalitarian ones, have this need already covered. David Altman's (2010) quantitative analyses of direct democratic mechanisms deal with democracies and non-democratic regimes as part of the same continuum, and he finds those mechanisms positively correlated to higher democratization indices. However, his specific analysis of dictatorships and semi-democracies is qualitative and focuses on why ‘direct democracy’ is used rather than on its consequences for regime stability or duration.

Evidence shows that plebiscites tend to come early in the regime's life, not at the end (Penadés and Velasco 2022; see also the Supplementary Material). This is at least prima facie evidence challenging the view that conducting a plebiscite is a step towards liberalization for an authoritarian regime (Altman 2010: 76). Instead, it brings us closer to the idea that plebiscites are events through which the authority of the ruler is forcibly acknowledged, or what Qvortrup et al. (2020) call ‘concessio imperii’.

Alberto Penadés and Sergio Velasco (2022) argue that plebiscites are coercive tools in the hands of dictators, rather than inclusive ones. It is assumed that this institutional instrument is intended to enhance the power of the dictator within the regime, reinforcing the autocrat over potential internal rivals, and to hinder the coordination of opposition outside the regime by sending a signal that almost everyone could be a supporter, or at least willing to act as one. Evidence is found to be compatible with these assumptions. Plebiscites are more likely to occur in institutional conditions where the individual rule of the dictator might be potentially more unstable (such as military dictators or dictators in semi-competitive regimes) and in social and political conditions that are more likely to have an impact on the informational requisites for collective action by the opposition (at early stages of the regime's life and in homogeneous societies).

Penadés and Velasco (2022) study the conditions in which authoritarian referendums occur. Expanding on that work, we delve into the subsequent consequences that emerge from these referendums. Do plebiscites actually work as autocrats might think? Do they make dictatorship more stable and long-lasting? Do plebiscites reduce the dictator's risks of being replaced by a rival? And do they reduce the organized external opposition to the regime? We provide affirmative answers to these questions.

In broad terms, the literature on authoritarian regimes sustains the conclusion that their institutionalization contributes to increasing the chances of regime survival. This is supported by Milan Svolik (2012) and Geddes et al. (2018) in their analysis of the fundamentals of institutional autocratic design as a whole. Jennifer Gandhi and Adam Przeworski (2006, 2007) point to the role of multipartyism and opposition in authoritarian rule's durability; Beatriz Magaloni (2006) highlights the role of political parties, spoils and elections; and Carles Boix and Milan Svolik (2013) find that the institution of a legislature has a positive value for the dictator's tenure. However, we believe that these arguments cannot be directly extended to plebiscites. Authoritarian representative institutions are usually supposed to facilitate inclusion, redistribution and efficient bargains with power rivals based on the information provided by elections. We believe that the role of plebiscites is different. In our understanding of plebiscites, they stop the opposition and internal rivals from acting against the regime, at least until new information is gained about the regime's support. They hinder the coordination of opponents and facilitate the coordination of supporters.

According to Svolik (2012), autocrats must solve a dilemma: maintaining an internal balance within the ruling clique while keeping the general population under control. We propose that a plebiscite would help autocratic regimes' survival since the use of plebiscites could ease the dilemma on both sides of the equation: plebiscites might deter the manifestation of opposition, both from external agents and from rivals within the ruling clique. The public and typically massive display of the leader's apparent popularity generates uncertainty about the true distribution of support for the opposition (Penadés and Velasco 2022). At the same time, we hypothesize that it also makes the prospects of an internal rebellion more uncertain and possibly lower. A successful referendum demonstrates the power to control the state apparatus and may raise policy or constitutional issues that weaken a potential coalition against the ruler. A skilful plebiscitarian dictator will choose the timing and wording of the plebiscite question so that potential rivals within the regime see their expectations of success decrease vastly, if not vanish altogether.

Empirically, we operationalize the manifestation of external opposition as mass mobilization and internal opposition as coups d’état. We argue that regimes that use plebiscites tend to survive longer by deterring mass mobilization and coups. In summary, we argue that plebiscitarian methods help autocrats to survive longer and maintain their power.

We formulate three testable empirical propositions that we evaluate using several statistical tools. We base our analysis on a time-series cross-sectional dataset resulting from the merger of the V-Dem database (Coppedge et al. 2020) with Geddes et al. (2018) and a few other additions. To evaluate the longevity of autocratic regimes using plebiscites, we transformed this database into a survival-shaped form. Additionally, we used the original time-series cross-sectional data and probit models to analyse the occurrence of mass mobilization events and coups d’état.

Our results show that regimes using plebiscites tend to survive for longer periods of time. We also find that plebiscites are correlated with a temporary reduction in the likelihood of mass mobilization and coups d’état, indicating a sedative effect on social unrest. However, this impact is ephemeral. Furthermore, we find that the combined effect of plebiscites on regime survival is significant. In summary, we demonstrate that plebiscites help autocratic regimes survive for longer periods of time and provide two mechanisms by which plebiscites may facilitate their survival.

The idea of appealing to the people at large to make a decision by vote can be said to have pre-democratic roots, in the modern sense of the word. It was first introduced in early modern politics for the ratification of constitutions, such as in revolutionary France, after an attempt by anti-federalists in the US, and for the incorporation of new territories into an emerging state, again in France. Since then, the story has two branches, and the autocratic path seemed stronger than the democratic one for some time. It became a favourite instrument of Napoleon and Napoleon III, and was also used in the first half of the 20th century by the likes of Mussolini and Hitler (Penadés 2018; Qvortrup 2022). Yet we confine our study to the period after World War II, for which there is enough data to test our hypothesis.

This article is organized into four sections. In the first section, we define the main terms of our research and state our hypotheses and expected results. In the second section, we address the empirical side of our research, including the data and models used for inference. In the third section, we present our results and discuss their implications for our hypotheses. In the fourth and final section, we provide a summary of our results and draw our conclusions.

Autocratic institutions, plebiscites and survival

Plebiscites are popular voting processes controlled by the executive in the double sense of controlling the agenda (the specific question or policy position that is subject to vote) and the timing of the decision-making. They are a subset of what Simon Hug and George Tsebelis (2002) call veto-player referendums. Authoritarian referendums can be assimilated to this category (Penadés and Velasco 2022). In dictatorships, suffrage, campaigning, information, actual voting and so on are also ostensibly controlled in the manner that authoritarian regimes control any other voting procedure. This is not what makes plebiscites different from most authoritarian elections. Elections can be tightly controlled and even cancelled, but they are expected to be regular and to offer alternative policies or candidates; plebiscites are not expected to do so. Even presidential elections, which may resemble plebiscitarian votes in the margins of victory and participation, need some regularity and some contestation. Authoritarian plebiscites are discretionary, need not be open to even controlled contestation, and are associated on average with very large participation and huge margins of victory (Altman 2010; Penadés and Velasco 2022).

We contend that plebiscites contribute to the durability of authoritarian regimes, although for different reasons than authoritarian elections and semi-pluralistic representative institutions. Henrik Knutsen et al. (2017) found that while the actual holding of an election process increases the risks for autocratic regime survival in the short run, the institution of elections has a positive survival value. Elections benefit the dictator despite the fact that they increase instability each time they are convened. Plebiscites, on the contrary, increase short-term stability, although their effect may weaken in the long run. We hypothesize that the institution of plebiscites increases the probability of survival for authoritarian regimes because they contribute to silencing the opposition, rather than giving an opportunity for opponents to organize or exercise voice. That is, plebiscites contribute to regime survival because of the short-run consequences of their being held, rather than, as in the case of elections, despite them.

Hypothesis 1: Plebiscitarian dictatorships survive longer than non-plebiscitarian ones.

We suggest two mechanisms for this regularity: plebiscites discourage internal coups as well as the external mobilization of the opposition. We regard plebiscites as part of a survival strategy primarily based on making it harder to coordinate potential opposition against the dictator, by regime insiders as well as by outsiders.

We emphasize the coordination-hindering aspect of plebiscites. In this regard, they are different from both more inclusive institutions, such as controlled elections, and from pure coercion. Opponents are neither incorporated (co-opted, bribed, appeased) nor physically removed (as incarceration or exile would). We suggest that plebiscites aim to silence the opposition by preventing the coordination among dissenters without necessarily changing their preferences. Plebiscites only make their collective action less likely. They may change expectations, induce fear and divide the opposition without necessarily resorting to either prizes or physical coercion.

Take the insiders first. A plebiscite may contribute to reinforcing the dictator's powers within the regime in at least three ways. The display of control of the state apparatus linked to a large and suspicious victory, especially one of that kind, sends a costly signal to potential conspirators inside the regime (Simpser 2013). For example, Pinochet's 1978 so-called ‘national consultation’ did not raise any policy or constitutional issue; the demand for support to his rule against a vague external threat was aimed in part to establish his pre-eminence over the Chilean military Junta (Barros 2002: 111). But second, the autocrat may also use the agenda power to divide a potential opposition coalition. For example, Franco in 1945, rather than just demanding legitimacy for his perpetual rule, proposed that the Spanish state should be turned into a monarchy and designated a king as his eventual successor, a position that must have seemed more tolerable for royalists among the military. Third, the plebiscite may explicitly overrule other institutional actors, such as constitutional courts or legislatures. Transitions to electoral authoritarianism in the 21st century make frequent use of them to entrench presidential powers (an early example was Venezuela in 1997) but they also take place in classical dictatorships (e.g. Iran's 1953 referendum to dissolve the parliament).

Plebiscites are not always successful in containing opposition, as was the case with Iran in 1953, which experienced a US-backed coup afterward. However, we expect that they will tend to discourage or at least delay conspiracies against the ruler. We do not expect the effect to be a long-term one, as we do not assume that plebiscites co-opt or target benefits to potential rivals. The same actors may reorganize later.

Nam Kyu Kim and Jun Koga Sudduth (2021) have argued that authoritarian elections and parties only have an effect on one type of internal coup, which they refer to as ‘reshuffling coups’, but not on ‘regime-changing’ coups. Their argument is based on the idea that the mechanisms operating through semi-pluralistic institutions are not effective in preventing regime-changing coups. However, our mechanism does not depend on the ability of institutions to produce efficient bargains or provide potential opponents with a stake in the regime. Rather, it hinges on the reduced ability of the opposition to organize for any purpose after a plebiscite. Hence, we do not need to make that differentiation.

Hypothesis 2: The risk of a coup d’état against the autocrat is reduced after a plebiscite.

Consider now the organization of protests or other forms of contestation that might be available to outsiders of the regime. We suggest that plebiscites make collective action among opponents of the regime more difficult. Coordination is always uncertain if there is not some common knowledge about preferences by potential participants; even if the potential for opposition is substantial, collective action may be blocked in the absence of that knowledge (Chwe 2001, 1999). A successful plebiscite, especially one that obtains huge support, even if it is not credible, prevents the formation of common knowledge about the true distribution of attitudes towards the regime. It sends a signal that nearly any citizen could be a supporter or at least be willing to pass as a supporter, and this can be enough to make communication of true political beliefs a risky action. Social conformity may lead non-believers not just to comply but also to punish other non-believers for not complying (Centola et al. 2005; Willer et al. 2009).

A regime that is successful at making people feign support and falsify their beliefs can be nearly as intimidating as a regime with full popular support. The Chilean opposition in the aforementioned 1978 ‘national consultation’ was aware of that: the clandestine opposition propaganda used messages such as ‘Do not be afraid to say what you think’ and ‘Do not say YES when you mean NO’ (‘Informe “La Consulta Nacional del 4 de Enero de 1978”’ 2018). As argued by Timur Kuran (1997), the falsification of preferences is an important element of authoritarian stability. In the absence of genuine consensus, a plebiscite may be a theatrical, nationwide display of such falsification. If we are right, and it contributes to the silencing of protests, this should reinforce the effect that the plebiscite has on potential internal conspirators, who not only observe the display of control of the state by the dictator but will also anticipate peace in the streets. The status quo will appear as the only game in town.

Hypothesis 3: Collective action against the autocratic regime is reduced after a plebiscite.

Again, the effects of plebiscites at inhibiting communication and coordination are not expected to be permanent, unless they are supplemented by other means. Plebiscites do not remove people from their positions; they just incite them to stay silent. In the absence of very effective coercion, communication can be expected to resume through hidden or discreet channels (Scott 1990).

In summary, we propose that plebiscites are an effective means to hinder the organization of the opposition by creating higher uncertainty around both internal conspiracies and collective action, thereby making coordination against the ruler more difficult to achieve. In a successful plebiscite, the authoritarian executive is buttressed against internal rivals and drives the public mood towards silence. Although these effects may diminish over time, they are still effective at prolonging the dictator's tenure.

Data and methods

In the following sections we will subject the aforementioned assumptions to empirical trial. To do so, we need to operationalize these concepts into data to test hypotheses. The first step in this process is selecting the dependent variables.²

The first dependent variable is regime survivability (also known as ‘regime tenure’ or ‘regime duration’ by other authors), which is the number of years a given regime lasts. This variable is extracted from Geddes et al.'s (2018) database on autocratic regimes. Therefore, we are using their definition of a regime, in which political structures (and leader selection procedure) take precedence over who is the head of state. Hence, a regime may be centred on a single dictator or feature multiple dictators under the same regime. For instance, Egypt is classified as a single regime under both Nasser and Sadat, while Iran's monarchy and subsequent Islamic Republic are considered two separate regimes.

The two dependent variables for the other two hypotheses are the occurrence of coups d’état and of mass mobilization. What we expect from a coup is that members of the regime aim for office circumventing the paths of established succession, as we are testing internal opposition. Jonathan Powell and Clayton Thyne's (2011) definition and database on coups serve this purpose.³ They define coups as events in which the chief executive of the state is targeted and the perpetrator is any section of the elite of the state apparatus – not exclusively the armed forces. This definition rules out foreign impositions or popular uprisings and focuses on our area of interest, regime internal subversion.

Our final dependent variable is mass mobilization. We obtained data on mass mobilization from Sebastian Lührmann and Anna Hellmeier's study (in Coppedge et al. 2020). They define mobilizations as events such as ‘demonstrations, strikes, protests, riots, and sit-ins’, which are most commonly organized by outsiders to the regime and are common protest instruments of external opposition.⁴ Originally, mass mobilization data were ranked on an ordinal scale that increases with the intensity of the mobilization event in terms of participant abundance and geographical scope. This variable is subsequently transformed to a dichotomous form by introducing a threshold in the third level, which marks the first step of large-scale mass mobilizations, to enable simpler and more comprehensive probit models and marginal effects interpretations.

Let us turn to the independent variables. The most relevant variable – which is the object of our study – is the occurrence of plebiscites. Altman (2010) classifies direct democracy into a threefold scheme, distinguishing between a vote that is convoked by the government (plebiscites), by citizens (referendums) or automatically triggered. As mentioned previously, there is no place for referendums in an autocratic context per Altman's definition; therefore, we use the terms interchangeably. Regarding automatically triggered votes, it is quite clear that the vast majority of them are triggered by the executive, and not automated by some legal mechanism or political compromise. Even in the most extreme cases of ‘compulsory autocratic referendums’, evidence suggests that this compulsory nature was not genuine. Mark Walker (2003) discusses this, arguing that in 1988 Pinochet was given several exit options from the so-called compulsory plebiscite, such as holding parliamentary elections or stepping aside in favour of other potential leaders. However, quoting Orlando Saenz's words, ‘Pinochet lived in a world of make-believe’ in which he would win the vote.

Data on the use of plebiscites are obtained from the V-Dem database (Altman in Coppedge et al. 2020). This variable is transformed for the survival models in order to build a variable that accounts for the occurrence of at least one plebiscite in a given regime. Moreover, in the probit models, we employ both the raw variable and a yearly timer generated from the use of a plebiscite. This procedure is inspired by Arturas Rozenas's (2010) statistical treatment of elections in the Soviet Union. Since most plebiscites cannot be foreseen in terms of predictable timing and there is no clear schedule for when they will take place, the only measure available is a yearly counter since the last plebiscite occurred. Additionally, in regimes in which no plebiscite takes place, the time has a null value. This variable is used in a polynomial form to account for the non-linear shaping of the effect. The degree of the polynomial was increased until statistical significance was lost.

In order to correctly estimate the effect of plebiscites on regime survival, we need to control for other potential factors affecting regime survival. A possible concern for empirically demonstrating the dynamics of our hypotheses is that the different kind of voting procedures (parliamentary, presidential, local etc.) may have similar effects and incentives, which could impact our independent variable. For instance, an autocrat may face a challenge to his position by a potential rival and decide to hold a presidential election with his name on the ballot in order to establish a clear dominance.

Therefore, it is necessary to isolate the effects of different voting procedures for an accurate account of the effect of plebiscites. Regarding the implementation of these controls in survival models, we use two variables that account for the use of presidential elections and for the use of any other type of election. Further elaboration on the implementation of these controls in probit models will be provided below.

There are other relevant control variables that affect overall regime tenure, such as economic performance. For this purpose, we include controls of natural resources (obtained from Haber and Menaldo 2011), GDP per capita and economic growth (both from the Maddison Project Database 2018 (Bolt et al. 2018)). Regarding the political structure of the regime, we have included controls for the type of regime: military, party, personalist or monarchical, with the last being the reference category, as these are the longest-lived regimes. Data on type of regime are obtained from Geddes et al. (2014).

We will now turn to statistical modelling. We assess whether plebiscites promote regime survival. This proposition implies using survival models, in which the dependent variable is the duration of the regime. We used various specifications for the model, in addition to the non-parametrical analysis. The basic model is the Cox Proportional Hazard model (Cox 1972). It was found that the proportional hazard assumption was not satisfactorily met (see Figure A1 in the Supplementary Material). In order to solve this shortcoming, various parametric models were evaluated, and the log-normal specification was ultimately chosen as it had the better Akaike statistic (see Figure A2 in the Supplementary Material),⁵ though all other options produced converging results for our hypothesis.

Besides this, we added two supplementary survival models. The first model used the full sample of regimes ranging from 1946 to 2008, which includes both ongoing regimes and those that have ended. The second model, however, uses only the regimes that ended before 2008 in order to rule out regimes with ongoing activity. Both models show equivalent results.

The second and third propositions – regarding coups and mass mobilization – were examined using panel probit models. Models for both hypotheses have random effects panel specifications to address the issue of unobserved heterogeneity, common in time-series cross-section data. We could not use fixed effects since its implementation would imply losing every time-invariant variable. Consequently, we opted for a random effects specification (Beck and Katz 2007).⁶ The first of these probit models concerns the occurrence of coups.

Similar to the survival models, there was an issue regarding the potential similarities among the different kinds of voting procedures and their impact on our three hypotheses. To address this issue, we introduced the control variables ‘time since last election’ (TElec) and ‘time since last presidential election’ (TPres). We have constructed these variables using the same method as we did with the plebiscite variable, following the approach described by Rozenas (2010). They are implemented in three separate models. This is due to some regimes not using these political instruments, resulting in a loss of observations. Although the abundance of observations contributes to the robustness of the model, not accounting for these issues seems a significant limitation. We have therefore chosen to include the models with and without these controls. This procedure is also applied to the models regarding mass mobilization that conclude the results section.

Results and discussion

In this section, we will test the propositions from the previous theoretical sections using the empirical methods explained earlier. Firstly, we will examine our hypothesis that plebiscitarian regimes have longer survival rates than non-plebiscitary regimes. After testing the main hypothesis, we will examine the models for two potential causal mechanisms, namely the reduction of both coup occurrences and social turmoil within autocratic regimes due to the use of plebiscites.

However, prior to proceeding with the modelling, it is relevant to examine a descriptive figure of interest. Specifically, this is the Kaplan Meier survival estimate, a non-parametric analysis that we find particularly illustrative in face of subsequent results (Figure 1).

Figure 1. Kaplan–Meier Survival Estimates

We can observe that in the initial years of an autocratic regime, the difference in survival is indistinguishable. However, from that moment onwards until around 20 years of existence, the use of plebiscites demonstrates a clear positive effect on regimes' longevity. In the Supplementary Material (Figure A4) we show how this effect is magnified when clustering regimes according to their usage of plebiscites. Of course, non-parametric analysis cannot account for the multiple factors that may influence the longevity of an autocratic regime. For that purpose, we employ the survival models of Table 1.

Table 1. Survival Models for Autocratic Regimes

Notes: Standard errors are in parentheses. Reference category in regime variables is ‘monarchy’. The difference in sign between Cox and log-normal is attributed to the estimation procedure, interpretation of the estimation is therefore opposite. Ln(sigma) omitted in table. *** p < 0.01, ** p < 0.05, * p < 0.1.

In Table 1 we present the results of our regime survival models. They are organized in two groups of columns, the first containing two Cox survival models and the second containing two log-normal specification models. Before analysing the table, it is important to note that Cox models estimate the effect on hazard rates, while log-normal models estimate the effect on survival itself. Although the signs may be reversed, the underlying meanings and implications remain the same. The first column shows that holding at least one plebiscite accounts for a reduction of over 27% in the hazard rate,⁷ while in the log-normal model we estimate that holding at least one plebiscite is correlated with an increase of over 46% in longevity. Regardless of the strength of the effect, it consistently supports our hypothesis across the different models. Political control variables reveal some interesting effects. While elections as a whole seem to have a similar impact as plebiscites – as expected – presidential elections have the opposite effect. Presidential elections seem to shorten regime survival. According to Tyson Roberts (2015), autocratic systems with parliament-based political institutions have incentives to co-opt and cooperate, while presidential systems without such incentives would have a lower survival rate. This effect is also reflected in our model. Additionally, it highlights the distinction between presidential elections and plebiscites in both empirical and theoretical terms. Besides this, we can also find that personalist and military regimes have the highest risk rates, suggesting that less institutionalized forms of autocratic rule carry a greater risk. In Figure 2, we find the survival and hazard functions for the first log-normal model.

Figure 2. Survival and Hazard Functions

Two notable insights can be drawn from the hazard function in Figure 2. The first is the typical hazard pattern exhibited by the autocratic regimes studied. After a low hazard estimate in the initial years, it rapidly increases to a peak in the first ten years of the regime, followed by a steady decline thereafter. The second finding is that the largest disparity in risk rates is observed in the interval between the first ten to twenty years of the regime, which marks the point of maximum risk. This suggests that the impact of plebiscites on risk is most pronounced during these particularly risky years.

We will now proceed with the modelization for the second hypothesis. To recap, we expect plebiscites to reduce the occurrence of coups d’état, while also expecting this effect to be ephemeral. The probit models for this phenomenon are presented in Table 2.

Table 2. Effect of Plebiscites on Coups d’état

Notes: TPleb = time since last plebiscite; TElec = time since last election; TPres = time since last presidential election. Robust clustered standard errors are in parentheses. Reference category in regime specification is ‘monarchy’. Ln(sigma) omitted in table. *** p < 0.01, ** p < 0.05, * p < 0.1.

In Table 2 we can observe the three versions of the models regarding coups. In the first column, we find the model without controls for ‘time since election’ and ‘time since presidential election’. In the second, we control for ‘time since election’, and in the third we include both controls. Therefore, the latter model is the sparsest in observations but is more reliable in terms of controlling for the intertwined effects of other types of elections. This created a trade-off between observations and reliability in this area, which is reflected in Table 2. In any case, our results regarding plebiscites are consistent with any model specification.

Results show that while the ‘plebiscite’ variable does not have a significant effect on coups, the time elapsed since the last plebiscite was held does. This means that although the impact at the regime-year unit level may not be significant, we account for the impact in a structured way over time. Furthermore, we can identify a significant impact that also exhibits a specific shape. We can better appreciate this shape in Figure 3, obtained from the third model.

Figure 3. Marginal Effect of ‘Years since Last Plebiscite’ Polynomial

Figure 3 shows a negative impact in the first few years, followed by a stable low level for ten years. After this, we find the disappearance of the effect, which supports our theoretical propositions.

Regarding the control variables of Table 2, economic growth and GDP have significant effects, showing a negative correlation between coups and economic performance. In autocratic contexts, it is commonly understood that regime stability is closely tied to economic performance. This understanding suggests that economic performance strengthens an autocrat's position to rule by providing more resources to distribute to their supporters and co-opt potential challengers to their rule. Wealthier and growing regimes are therefore expected to have fewer incentives to depose their dictator than their poorer counterparts. Additionally, strong economic performance can improve the autocrat's popularity with the general public. As a result, it is expected that a strong economic stance deters attempts at a violent overthrow of the government.

In the first two models, we found a positive correlation between coups and military regimes, which may be explained by Anne Meng's (2020) research. She found that coup leaders – usually military officers in various shapes and forms – were less likely to stablish institutional thresholds for their regimes, there being a clear path for succession, which in turn may explain subsequent coups d’état. Tenure and time since election, whichever the form, seem to have no significant impact in the occurrence of coups.

We will now proceed to assess our third hypothesis, the reduction of social mobilization through plebiscitarian methods. Similarly to our proposition regarding coups d’état, we expect to find a temporary decrease in collective action against the regime. The results can be found in Table 3.

Table 3. Effect of Plebiscites on Popular Turmoil

Notes: TPleb = time since last plebiscite; TElec = time since last election; TPres = time since last presidential election. Robust clustered standard errors are in parentheses. Reference category in regime specification is ‘monarchy’. Ln(sigma) omitted in table. *** p < 0.01, ** p < 0.05, * p < 0.1.

The results likewise show that the plebiscite variable is not correlated with social turmoil. However, plebiscites have an effect through the time variable, following the same pattern described in the previous proposition. In this case, the effect is weaker as the effect loses significance in the first and partially in the second model, while it remains significant in the third model. The time effect of plebiscites can be best visualized in Figure 4, which is elaborated with the coefficients of the third model.

Figure 4. Marginal Effect of ‘Years since Last Plebiscite’ Polynomial

In Figure 4, extracted from the third model, we can appreciate a negative impact on the odds of mobilization in the first years since a plebiscite took place. After that, the expected anaesthetic effect fades away, leading to a steady increase in the probability of turmoil.

Returning to Table 3, the economic control variables show non-significant estimators. On the other hand, some of the regime configuration variables show expected results. Specifically, military regimes have a positive correlation with social turmoil in the first two models. This is consistent with the idea of institutionalization as a way of reducing social unrest and gaining political legitimacy, as mentioned in the models related to coups. Regarding the occurrence of elections, we find a significant positive correlation in the second and third models. The time effect of elections indicates that they tend to foster turmoil in the short term, but this effect diminishes in the long run.

Conclusion

The aim of this article is to determine whether the use of referendums extends the survival of autocratic regimes. Although there is anecdotal evidence of this phenomenon in numerous historical vignettes, so far we have lacked generalized evidence to support the claim that referendums generally prolong the life of a dictatorship. In this article, we provide that evidence.

We propose two ways in which plebiscites help autocrats extend their rule. Firstly, by deterring internal opposition from overthrowing the current leaders, thus defusing coups d’état. We argue that referendums reduce the risk of coups since they send a strong signal of control of the state corps to potential rivals. Additionally, the voting mechanism can be used to promote cleavages within rival coalitions, as was the case in Franco's referendum on monarchical succession.

Plebiscites also help to deactivate opposition at the grassroots level outside the regime. These mechanisms disrupt the formation of common knowledge about who supports the regime and, in the face of serious repercussions, encourage the falsification of political preferences, making it difficult for the opposition to mobilize support and grow.

We have found statistical evidence of both phenomena occurring in autocratic regimes during the second half of the 20th century. This article aims to contribute to a better understanding of autocratic plebiscites and their impact on the functioning of an autocratic regime, given that this mechanism is sometimes misunderstood as a mechanism for liberalizing the political system.