This article examines whether there is an increase in repression in the election year in electoral autocracies. First, we build a simple theoretical model of an electoral autocracy. We assume that autocratic rulers want to maximise the expected rents from office. As a higher vote share in the election is translated into a higher probability of remaining in power, they use repression to exclude those opposing the ruler from the electoral body. Then, in the empirical section, we use a data set of autocracies to examine the existence of an electoral cycle in repression. We use a dynamic inverse probability weighting regression adjustment model, which models the dynamics of elections on the respect of human rights and considers that elections are non-random events. All results, and several robustness tests, indicate a strong presence of an electoral cycle in repression consistent with our theoretical model and priors. Additionally, we find that when it comes to autocracies, this cyclical increase in repression is more pronounced than the pre-electoral increase in government spending.
On 7 November 2021, Nicaragua held general elections. However, several states, including the US, as well as the European Union (EU), and independent observers have heavily criticised these elections as a ‘sham’. The main reason was that in the months preceding the elections, the government had escalated the political violence by jailing most opposition members to secure re-election. This example brings the central prediction of electoral cycle literature to mind (see, for example, Alesina et al, 1993; Schuknecht, 1996; Drazen and Eslava, 2010): politicians will manipulate the electorate to secure an electoral victory. Moreover, elections are not rare in non-democracies (Magaloni, 2008; Levitsky and Way, 2010; Gehlbach et al, 2016; for a recent review, see also Mandon and Cazals, 2019).1 Therefore, we should expect that autocrats will also try to manipulate the electorate at the time of elections. However, autocrats face fewer constraints on the use of their power. Hence, manipulating the economy through spending and taxation may not be their only instrument. As an alternative, they might use repression to silence any opposition as well. This will directly reduce the votes for their opponents and increase their vote share (see, for example, Robinson and Torvik, 2009). This article examines if such an electoral cycle in repression exists.
To motivate our analysis, we first build a theoretical model of an electoral autocracy. In this simple model, we follow the existing literature on autocratic elections and assume that the autocrat seeks to maximise the total number of votes cast in their favour. For that, they also have a powerful instrument at their disposal: costly, in terms of forgone rents, repression. Thus, silencing the opposition excludes those politically aligned against them from the electoral body. We show, then, that in election years, the ruler employs repression to secure a higher share of votes. This latter idea is also found in several other contributions (see, for example, Bratton, 2008; Robinson and Torvik, 2009; Collier and Vicente, 2012).
In the empirical section, we build a data set of autocracies to examine whether the electoral cycles in repression exist. Our primary measure of repression is the respect for human rights latent variable, as constructed by Fariss et al (2020).2 Data for elections are taken from the National Elections across Democracy and Autocracy (NELDA) data set (Hyde and Marinov, 2012). To determine which countries are autocracies and thus where the expected effect of pre-electoral repression might be present, we use the data of Bjørnskov and Rode (2020). To estimate the cyclical impact of elections on repression, we use a dynamic model in the spirit of Angrist et al (2018), Jordà and Taylor (2016) and Kuvshinov and Zimmermann (2019). We opt for this method because it deals with an inherent endogeneity problem. Elections, especially in autocracies, are not random events. Hence, the empirical method creates pseudo-randomisation across observations to deal with this issue. This is achieved by weighting less the instances where elections are expected with a higher probability and vice versa. In this way, we can estimate the causal effect of elections on repression, both in the pre- and post-election periods.
This empirical method has several advantages. First, it allows us to derive the dynamic effect of elections on our variable of interest, directly testing for the hypothesised cyclical effect. Second, it is a two-stage model, where none of the stages relies on any form of exclusion restrictions. Hence, we can estimate the causal effect of elections on repression by including variables that can be considered endogenous (Kuvshinov and Zimmermann, 2019).
According to our results, there is an increase in repression in electoral years. However, this increase exists only in autocracies. There is no evidence of a similar cycle in democratic countries. These results stand in sharp contrast to previous studies (for example, Bhasin and Gandhi, 2013), which find that repression falls just before the elections. Furthermore, this increase is only temporary and repression returns to its previous level in the year after the elections.
Interestingly, when we examine the presence of electoral spending cycles, we fail to find such a relationship in autocracies, even though there is some evidence for their occurrence in democracies. Finally, our empirical analysis indicates the validity of the proposed theoretical channel. We find that pre-electoral repression is correlated with lower voter turnout rates. Moreover, several robustness tests verify all our results.
The idea of electoral cycles in repression has been examined in several instances. For example, Davenport (1997) explores the relationship between national elections and repression. In contrast to the present study, he finds that elections negatively affect repression in non-democracies. Similarly, Richards and Gelleny’s (2007) results indicate an increase in the respect for human rights after the year of national elections. Bhasin and Gandhi (2013) show that autocratic regimes moderate violence against citizens before elections; yet, at the same time, they increase violence against the opposition. However, both these effects are reversed after the election.
These studies use events-based variables extracted from news sources to measure repression. In contrast, here, as a measure of repression, we use the latent variable of respect for human rights, constructed by Fariss (2014) and Fariss et al (2020). This variable has the advantage that it is constructed assuming that state repression is unobservable. In autocracies, this might be a very reasonable assumption. After all, a powerful autocrat might be able to control the media, biasing the number of reported events of state violence (Fariss, 2019). Furthermore, their incentives to withhold information are more pronounced before elections, that is, when there is an increase in interest in the events within the country. Thus, the observed decline in pre-electoral violence might be due to media manipulation on the eve of elections. In contrast, a latent variable is not plagued by this shortcoming. Assuming that the underlying repression variable is unobservable, it uses a Bayesian model and several events-based and standards-based data sources to derive state repression’s ‘real’ measure.3
The present article is related to the literature that examines the effect of the electoral cycle on different political systems. Veiga et al (2019) find that electoral cycles are more likely to appear in specific institutional environments. Electoral cycles are more pronounced in new democracies with majoritarian political systems, predetermined elections and weak executive constraints. Similarly, Veiga et al (2017) show that media freedom is a factor that plays a crucial role in the occurrence of electoral cycles. Eibl and Lynge-Mangueira (2017) show that political business cycles are conditioned on executive constraints. The results presented here complement these studies. Our findings suggest that the political regime matters in other manners as well. It not only affects the magnitude of the cycle; instead, the ruler has extra instruments to manipulate the electorate, depending on the regime type. More related to the present article, Bhasin and Gandhi (2013) show that autocrats increase the number of events related to violence against electoral opponents before elections. At the same time, they reduce the related events directed at voters.
The rest of the article is structured as follows. We develop a simple theoretical model in the next section. Then, the third section presents the data used in the empirical analysis. The fourth section then explains the empirical results and the robustness analysis. Finally, the fifth section provides some concluding remarks.
The model borrows from the political budget cycle literature (for a general review of the literature, see, for example, Persson and Tabellini, 2002; for a model of autocratic elections, see, for example, Robinson and Torvik, 2009). We assume a two-period model of an autocratic polity populated by a large number of individuals normalised to unity. At the end of the first period, that is, after t = 1, elections are held. The outcome of these elections affects the probability of the ruler staying in power for the second period, that is, t = 2, the post-election period. There are two groups of individuals. The first one votes only on ideological grounds and is assumed to always vote against the dictator. This group is denoted as opposition and is a share, n, of the total population. The other group cares about its private income and has an individual-specific ideological bias against the use of repression. We denote this latter group as voters.
Income is the same across individuals and constant in time, denoted by y. The government raises tax revenues by imposing a fixed (exogenously given) tax rate in each time period. This tax rate finances a general public good, gt, and the rents rt of the ruler. As we model an autocratic polity, it is natural to assume that an inherent feature of an authoritarian regime is the ability of the ruler to ignore unfavourable results or avoid being deposed by the ruling coalition despite winning the elections (see, for example, Bueno de Mesquita, 2010; Fearon, 2011; Egorov and Sonin, 2014). Thus, we further assume that winning the elections is not the only factor affecting the ruler’s probability of remaining in office. However, this probability is a function of the total share of votes cast in favour of the ruler. Hence, in our model, remaining in office is determined by the power projected by the ruler in the elections (Magaloni, 2008) or the regime’s legitimacy bolstered by the electoral outcome (Joseph, 1997; Williamson, 2021).4
Individuals with will choose to abstain, whereas individuals with will show up and vote against the ruler.
The first two terms in Equation 2 represent the private utility gain, as derived by after-tax income and the public good provided by the government. Similarly, the third term suggests that state repression might put voters off. For example, a highly repressive regime might lose its legitimacy even in the eyes of its supporters. Alternatively, individuals might want to ‘punish’ the highly repressive regime by not voting for the ruler. Thus, we assume that Rt creates a welfare loss even for those willing to support the incumbent. Further, we conceive the voters as holding different views regarding the autocratic ruler, translating into an extra disutility cost from repression. Thus, we assume that , that is, the disutility of repression, is individual specific, and it is uniformly distributed in the range.5 An individual who highly values the ruler has a very low value of and derives low disutility from the regime’s repressive activities. Similarly, denotes the density of the uniform distributions, with higher values representing higher density and more homogeneous supporter preferences. Thus, the term represents the ideological component in the voters’ utility, which is individual specific, and also corresponds to the degree that ideology matters in the individual’s voting behaviour. The preceding formulation allows us to model an inherent trade-off of using repression for the ruler. It prevents the opposition from voting, but this comes at the cost of losing support and legitimacy from the rest of the voters.6
Here, is the probability of remaining in power in the second period. This function reflects the fact that the election result does not fully predict whether the ruler will stay in power. The ruling coalition in several autocratic regimes can easily ignore unfavourable electoral results and keep the ruler in office, and, of course, the opposite holds as well. Elections may be a destabilising force in an autocracy (see Howard and Roessler, 2006; see also the discussion in Kaya and Bernhard, 2013). Therefore, even electoral wins with a wide margin may result in regime breakdown. In essence, then, can be considered to capture various institutional characteristics of the autocratic regime.
Regarding , we make two assumptions. First, we assume that it is increasing in . An electoral victory increases the ruler’s power and allows them to rule for t = 2. A low first derivative implies that the electoral outcome does not affect the probability of remaining in power. In contrast, if the polls primarily affect the probability of remaining in office, the electoral outcome will matter a lot. Second, we assume a negative second derivative. At least after the third wave of democratisation, elections in authoritarian regimes were introduced to respond to pressures from the international community and to legitimise the ruler’s position (Han, 2014). However, it was typical, at least initially, for authoritarian regimes to receive extreme votes for the incumbent. Yet, ‘farce’ elections (Egorov and Sonin, 2021) do not confer more legitimacy to autocrats. To some extent, there is a trade-off between a high vote margin to project strength (Magaloni, 2008) and a credible vote margin that would favour partnerships with foreign partners (logrolling, trade and so on).7 With the preceding consideration in mind, we assume that the probability of remaining in office is an increasing function of the vote share and that the second derivative of this function is negative. In this manner, a higher vote share in favour of the leader increases the probability of remaining in power. However, if this share of votes is excessive, the associated probability of remaining in office does not change much.8 These two assumptions guarantee an internal maximum for the ruler’s maximisation problem.
Individuals with will vote against the ruler, whereas individuals with will vote in favour.
The ruler has a fixed share of taxes that they expropriate as rents, use to finance the public good and use to exert costly repression on the opposition. An increase in has two effects. First, it reduces the share of votes of the opposition by excluding those individuals who have a high disutility from repression from the electorate. At the same time, however, it also makes the marginal voters less likely to vote for the ruler; as for these individuals, the ideology cost from repression increases. Similarly, higher first-period rents directly increase the utility of the ruler but, at the same time, come at the cost of forgone votes as g1 falls.
Testable Hypothesis 1: Repression is expected to rise in the period leading to elections. After the elections, the level of repression will return to its previous level. Hence, elections only have a positive short-run effect on repression, and the long-run effect is zero.
Testable Hypothesis 2: Increased state repression is associated with lower voter turnout in the upcoming elections.
Even though providing valuable insights, the model outlined earlier considers repression to be the sole strategy of electoral manipulation available to the autocrat. Following Cheesman and Klaas (2018), autocrats also have other instruments besides repression. Other types of strategies employed are distorting the size of districts before the elections, vote buying, election hacking, adding fake votes or, if everything fails, duping the international community into legitimising poor-quality polls. Of course, all these are complementary, and their relative importance will depend on the characteristics of each autocratic regime. Obviously, the distortion of the size of distracts and vote buying are also strategies that can only work if the regime employs them early on, well before elections occur. At the same time, the other three techniques are used if everything else fails and are only evident in the electoral results (see, for example, Deckert et al, 2011). However, autocracies, especially military dictatorships, have a comparative advantage in the use of violence (Wintrobe, 1998). As violence has a multifaceted impact (Cheeseman and Klaas, 2018), repression is a valuable tool in the ruler’s hands despite the considerable cost it bears. As such, state violence is expected to be widely used and to erupt in the period before the elections.
Another point worth noting is that the increase in pre-electoral repression is consistent with several other channels that emphasise the role of the opposition or other important actors in an autocratic polity, for example, the army, the police or the autocrat’s ruling coalition. In the present setting, the opposition was relatively passive. For example, when elections are crucial in determining the regime’s survival, autocrats may increase repression in the pre-election years, as this is a time when tensions around power may be exacerbated. In this framework, repression limits the potential for rebellion and protest among the population. However, as the population assumes that elections affect a leader’s survival, we expect this to increase voter turnout to overthrow the ruler. Therefore, we should expect that Testable Hypothesis 2 will not be verified in this case. An alternative explanation might also involve a reverse chronology: as the proximity of elections increases unrest among the population, leaders respond to the observed turmoil by increasing repression. For this reason, in the empirical section, we estimate a model that places a higher weight on unexpected elections, thus effectively controlling for this latter issue.
Data and empirical methodology
Our sample consists of 167 countries over the 1975–2020 period.11 As testable Hypothesis 1 refers to autocratic regimes, we exclude from the sample election events and observations where a country has been democratic. These observations are only included in the robustness analysis when examining the associated effects on democracies. To distinguish between democracies and autocracies, we use the Bjørnskov and Rode’s (2020) dichotomous index of democracy, which updates and extends the variable initially constructed by Cheibub et al (2010). The underlying definition of democracy is minimalistic and depends on whether elections were conducted, whether these were free and fair, and whether there was a peaceful turnover of legislative and executive offices following those elections.
The dependent variable that measures the degree of repression in each country is the latent human rights protection scores of Fariss et al (2020). It is constructed by combining several items from various sources that measure human rights, composing a final multidimensional measure. These are standards-based variables (for example, the widely employed CIRI variables, the Political Terror Scale, the Hathaway Torture data, and the Ill-Treatment and Torture data set) and events-based ones, which include documentary evidence from multiple sources to identify the occurrence of certain repression events (for more details, see Fariss et al, 2020). The latent variable model is estimated to capture the non-observable characteristics of state repression by incorporating information from these multiple data sources. The non-observability of state repression is an important issue; after all, information about human rights violations does not flow freely, especially in autocratic regimes, and, most importantly, restrictions on the freedom of the press can be expected to be higher in the pre-electoral period. Hence, using only events-based scores for human rights violations may bias the results, especially when considering the case of pre-electoral violence. Finally, the measure considers the dynamic structure of the state’s repressive behaviour, thus controlling for the systematic change over time in how human rights monitoring agencies collect and interpret behaviour about state repression (Fariss, 2014). For the present study, we use the mean value of the continuous latent variable for each country and year, recoded to fall in the –100 to 100 scale, with lower values indicating lower respect for human rights and higher repression. We denote this variable as , and this is our main dependent variable.12
To examine the robustness of our results, we also use the Political Terror Scale index of state repression (Wood and Gibney, 2010: 373) as an alternative dependent variable. This is a discrete variable built from the Amnesty International annual reports on human rights practices across countries. It codes the type of violence carried out by the state, the frequency of the violence and the portion of the population targeted. In contrast to the baseline measure, it depends on the subjective coding of the researchers, as well as the information provided in the reports of Amnesty International. The final score of the variable takes values on a 1–5 scale, with 1 denoting countries under a secure rule of law and 5 denoting countries where terror has been extended to the whole population.
To determine the year that elections were held in each country, we use the NELDA (version 6) data set (Hyde and Marinov, 2012). This data set provides extensive information on all election events from 1945 to the end of 2020. Only elections for a national executive figure or a national legislative body where there was a direct vote by the ‘people’ are included in the data set. Thus, local elections and elections by a committee or institution are not included. In contrast, it consists of all elections, even when they are not competitive and/or free. Our main variable of interest, denoted as Autocratic Elections, is a dummy variable that takes the value of 1 in the year that elections occur.
Typically, to test the effect of elections on a variable, the political cycles literature employs a specification where the impact of an election dummy on the variable of interest is estimated. This, however, implicitly assumes that elections occur randomly. For autocracies, though, this is seldom the case. Nygård (2020) shows that autocratic regimes that strategically choose the timing of elections are more stable. Authoritarian rulers can delay the timing or stage ‘snap’ elections when the opposition’s collective action problems are more severe. Of course, past repression will make it harder for the challenger to organise. Therefore, regimes will have an advantage in calling elections if repression was high in the past. At the same time, as the timing of regular elections approaches, it is more likely that the regime will increase the level of repression to prevent the opposition from organising.
To correct for this endogeneity problem and, at the same time, estimate a dynamic model, we use the inverse probability weighting scheme, as in Angrist et al (2018), Jordà and Taylor (2016) and Kuvshinov and Zimmermann (2019). More specifically, we estimate the change in the repression in the year that elections occur and the change in repression in off-election years by weighting more instances where elections were less likely to occur. Similarly, cases where elections are more likely to occur, as determined by the degree of repression in the past and the time since the previous elections, receive a lower weight.
Here, is the predicted probability of elections in an autocratic country i at time t, is the cumulative distribution function of the standard normal distribution, is the one-period lagged , is the vector of control variables used in both stages (as in Jordà and Taylor  and Kuvshinov and Zimmermann ), is a variable that takes a value of 1 if the country held elections four or five years before, are time fixed effects, and finally is a vector of estimated coefficients.
Here, in election years and ) otherwise.13 Therefore, is the Average Treatment Effect (ATE) of the treatment , that is, elections in t = 0, under the inverse propensity score weighted regression adjustment (IPWRA) estimator. Similarly, the restriction that , gives the ATE under the inverse propensity score (IPW) estimator (Angrist and Kuersteiner, 2011; Jordà and Taylor, 2016; Angrist et al, 2018; Kuvshinov and Zimmermann, 2019). Equation 13 is estimated for each horizon h = 0, 1, 2, 3; thus, we compute the change in repression for each time period h.14 We follow Jordà and Taylor (2016) and use the cluster-robust method to compute the estimated coefficients’ standard errors.
The estimated from Equation 13 uncovers the causal effect of the elections in autocracies under the assumption of the selection on observables, that is, after conditioning on observables in Equation 12, all other variation in the treatment variable is due to randomness. The latter, of course, is a highly restrictive assumption; therefore, we take several measures to ensure that it is satisfied. First, we consider the more general model, that is, the IPWRA model, which estimates the true ATE as long as either the first-stage probit regression or the equation that models changes in repression is correctly specified. Second, and most importantly, we include the lag of repression as our main predictor. Moreover, the presence of elections in the past four or five years takes into account the constitutional constraints in having elections and, hence, acts as a strong predictor of future elections. The inclusion of this variable rests on the idea that a country with a previous history of elections has a higher probability of holding elections in the present.15
Furthermore, we include a series of additional control variables in the model. First, the vector contains the log of gross domestic product (GDP) per capita; higher GDP per capita, that is, greater economic wealth, is expected to lead to a stronger and better educated middle class, which demands more access to the decision-making process and, thus, more civil liberties (Lipset 1959; Kaempfer and Lowenberg 1988; Barro 1999). As an additional control variable, we include the share of international trade, that is, exports plus imports, in GDP. Enhancing economic ties with the rest of the world is expected to lead to a pro-democracy movement in less democratic countries (Levitsky and Way, 2010). Thus, it will lead to more civil liberties and lower state repression. Similarly, higher economic growth rates can be associated with better population satisfaction; hence, the regime may not need to resort to repression to secure more votes. Finally, we include the share of the population aged 14 to the total population. A higher percentage of the younger population is associated with more anti-state political violence (see, for example, Austin, 2011; Apolte and Gerling, 2018). Consequently, governments that face a ‘youth bulge’ are forewarned and respond by being more repressive than other states (Nordås and Davenport, 2013).16
The method described earlier has a series of advantages. First, it allows us to estimate the dynamic effect of elections and derive the effect of the electoral cyclical on repression. Second, it does not impose a specific functional form and accommodates possibly non-linear dynamic effects on the outcome variable. Third, we extend the empirical model to allow for the local projection of repression by estimating a two-stage model that relies on the inverse probability weighting and the regression adjustment method (Jordà and Taylor, 2016; Jordà et al, 2016). This latter model has the significant advantage that it requires only one of both stages, ordinary least squares (OLS) or probit, to be correctly specified to derive correct estimates for the treatment effect, that is, elections (see Wooldridge, 2010). Furthermore, neither method relies on exclusion restrictions; thus, all variables can be considered endogenous in our data set (Kuvshinov and Zimmermann, 2019). As a result, our analysis effectively takes into account the endogeneity between elections and repression, and, thus, we derive the causal effect of elections on repression. Finally, as Equation 13 is estimated for different time periods, we can estimate the effect of elections on repression for both the election and the post-election years. Thus, we can derive the full cyclical effect of elections on repression.
The following section also examines the empirical validity of Testable Hypothesis 2 that we derived in the previous section, which effectively tests the underlying theoretical channel. Specifically, it investigates the effect of repression on voter turnout: according to the theoretical model, autocratic leaders strategically employ repression to reduce voter turnout in the upcoming elections. To examine whether such a relationship exists, we use a simple linear model in which the dependent variable is the voter turnout, that is, the share of people that voted to the total number of registered voters in the legislative or executive elections. The variables for voter turnout are taken from the International Institute for Democracy and Electoral Assistance (IDEA) data set. Finally, to examine the robustness of the results, we also use the same variable as taken from Armingeon et al (2021).
Table 1 presents the main results.17 Each line corresponds to a different model, whereas each column (for a given line) corresponds to a separate regression of Equation 13 for the various values of h. First, we estimate a model as specified in the previous section, where the treatment is the occurrence of elections. The first column presents the change in repression in the year of the elections from the corresponding value in the pre-election year. In a similar vein, Column 2 shows the difference from the election year to the year after the elections and so on. Our analysis indicates a clear cycle in state repression; this begins in the year of the elections, where the change in the respect for human rights from the previous year is negative, being statistically significant at the 5 per cent level and qualitatively equivalent to approximately one fifth of the standard deviation on the dependent variable. However, in the next year, there is an improvement in the variable, which is statistically equivalent to the initial decline in the electoral year. This is verified by a test that never rejects the null hypothesis that the total effect in Election Year +1 is zero. In other words, the initial effect is cancelled out in the post-election year, that is, repression returns to its initial value. Accordingly, the effect becomes statistically equal to zero for the second year after the elections. In contrast, there is a new decline in the repression three years after the elections, possibly corresponding to a new pre-electoral year. However, this latter effect is only statistically significant at the 10 per cent level and does not appear to be robust across the various specifications of the model.
|Model||Treatment||Election year||Election year + 1||Election year + 2||Election year + 3||Obs|
|(1) Main model||Elections in autocracies||3.66** (2.28)||–4.04** (–2.10)||–1.44 (–0.68)||3.56 (1.63)||2,238|
|(2) Two lags||Elections in autocracies||3.10** (1.99)||–4.01** (–2.05)||–0.93 (–0.43)||2.37 (1.14)||2,119|
|(3) Democratic elections||Elections in autocracies (b1)||7.43*** (3.94)||–1.08 (–0.59)||–1.94 (–1.07)||4.83*** (2.58)||5,587|
|Elections in democracies (b2)||–1.46 (–1.25)||–2.38** (–2.42)||1.29 (1.43)||–1.06 (–1.02)|
|Test b1 + b2 > 0||3.99#||–1.97||–0.38||2.00#|
|(4) Only electoral regimes||Elections in autocracies||5.73*** (2.98)||–4.83** (–2.52)||–1.50 (–0.64)||2.06 (0.86)||1,793|
|(5) Non- truncated||Elections in autocracies||3.66** (2.28)||–4.04** (–2.10)||–1.45 (–0.68)||3.56 (1.63)||2,238|
|(6) No time effects||Elections in autocracies||3.49** (2.17)||–3.94** (–1.98)||–1.35 (–0.60)||3.59 (1.61)||2,238|
|(7) After 1990||Elections in autocracies||3.80* (1.92)||–2.96 (–1.23)||–1.30 (–0.50)||1.95 (0.71)||1,435|
|(8) Alternative coding of elections||Elections in autocracies||3.44** (2.22)||–3.56** (–2.00)||–1.24 (–0.59)||1.74 (0.84)||2,125|
|(9) Political Terror Scale||Elections in autocracies||0.07* (1.65)||–0.08 (1.42)||–0.04 (–0.69)||0.02 (0.37)||1,177|
|(10) Spending in autocracies||Elections in autocracies||0.13 (0.83)||–0.15 (–1.05)||0.07 (0.46)||–0.08 (–0.82)||2,019|
|(11) Spending and elections||Elections in autocracies (b1)||–0.12 (–1.07)||–0.01 (–0.14)||0.18 (1.30)||0.00 (0.04)||5,477|
|Elections in democracies (b2)||0.13* (1.65)||–0.16** (–2.31)||–0.12** (–2.34)||–0.06 (–1.06)|
|Test b1 + b2 > 0||0.06||–1.67||0.45||–0.67|
Notes: The table provides the estimation results for the model as in Equation 13. The dependent variable in the first eight panels is the change in repression as defined in the main text. In Row 9, the dependent variable is the change in the Political Terror Scale (see Wood and Gibney, 2010). Finally, in Rows 10 and 11, the dependent variable is the change in total government consumption as a share of GDP (taken from the World Bank’s World Development Indicators). Clustered t-statistics in parentheses. *, ** and *** denote statistical significance at the 10 per cent, 5 per cent and 1 per cent levels, respectively. Test b1 + b2 > 0 is a t-test of the hypothesis that the sum of b1 and b2 is positive.
In Figure 1, we present the estimated effect over time to get a visual idea of the main results. Figure 1 indicates a clear pattern of electoral repression cycles. Even though there is an apparent increase in repression in the electoral year period (t = 0), in the years that follow, there is an evident cyclical effect, that is, a decline in t = 1, and an insignificant effect in the years that follow.
In the rest of the rows in Table 1, we examine the robustness of our results. In the second row, we introduce a second lag in both stages of the estimated model. Likewise, in the next row, we estimate the model for the full sample of countries, including democratic ones, introducing an additional treatment effect for the election year in democracies. This later effect is insignificant in the election year and positive and statistically significant in the year after the elections, suggesting that elections in autocracies and democracies affect human rights differently. In Row 4, we examine the effect of dropping the countries with a non-electoral regime, as determined by Bjørnskov and Rode (2020). There are no expectations for elections under the incumbent regime in these countries. Thus, excluding them should have no bearing on the estimated effect. In Row 5, we examine the effect of not employing a truncation in the propensity score, as in Imbens (2004). Then, in Row 6, we look at the effect of not including time effects in the model. In all cases, our main results remain unchanged. Finally, in Row 7, we estimate the model for the years after 1990, that is, after the third wave of democratisation, to establish whether the repression cycle holds after the fall of the Soviet Union and the subsequent collapse of many other authoritarian socialist regimes. Compared to the baseline model, the only change is that the treatment effect in the year after the elections becomes statistically insignificant. However, its magnitude remains the same; thus, we cannot rule out the possibility that the non-significant effect is due to the lower number of observations.
Row 8 provides a robustness test regarding the coding of the election variable (as in Franzese, 2000; Shi and Svensson, 2006). Following our theoretical argument, as repression serves as a pre-electoral mechanism in the hands of the autocratic leader, if elections are held early in year t, we should expect the increase in repression to occur mainly in year t – 1. Therefore, in Row 8, we code our main independent variable to take the value of 1 if elections occur after June of year t. In contrast, elections before this month are coded to appear in year t – 1. The findings in this exercise indicate that the main result is not sensitive to the coding of the election time. Moreover, all estimated effects have the same sign, statistical significance and magnitude as the baseline results.
In Row 9, we examine the robustness of our results to the choice of the repression variable. Thus, we use as the outcome variable the Political Terror Scale. This latter variable takes higher values when there is more repression. The sign of the effect in the election year follows our priors. However, the estimated effect turns out to be marginally insignificant at the 10 per cent level of statistical significance, with a p-value of 10 per cent. Even though this result provides only weak evidence in favour of our main hypothesis, this may be due to the lower number of observations of the Political Terror Scale. Moreover, as already highlighted in the previous section, the Political Terror Scale does not capture as many aspects, especially unobserved characteristics, of state repression as the latent variable we use in the main specification.
In the final two rows of Table 1, we perform the same analysis as in Equations 1 and 2; however, the outcome variable this time is the share of government consumption in GDP, as taken from the Penn World Tables. Again, the analysis indicates that an electoral effect on spending is not present in autocracies. Instead, when we include elections in democracies as the treatment, we find a small but statistically significant effect of elections in the current electoral period. These findings indicate that even though we do not find evidence of an electoral spending cycle in autocracies, we cannot rule out the presence of electoral spending cycles in democracies.
The theoretical model outlined earlier assumes that the electoral cycle of terror in autocracies exists due to its effect on voter turnout. In other words, according to our theoretical priors, the autocratic ruler intimidates the opposition in pre-electoral periods and, hence, individuals who oppose the ruler do not show up at the polls. It is, then, essential to examine whether such an effect, that is, repression and voter turnout, exists.
We use the Voter Turnout Database of the IDEA to measure voter turnout. For the associated analysis, we use all countries where elections took place in the sample. The variable of interest this time is the voter turnout as a share of the total population eligible to vote. The elections we consider are all national elections, either for the parliament or the president. In Column 1, we first estimate a simple model with OLS, where the dependent variable is the voter turnout and the respect for human rights and time effects are the controls. Then, in Column 2, we estimate the same model by adding control variables (see Geys, 2006). These are: (1) the log of population; (2) the population density; (3) the GDP per capita growth rate (from the World Bank’s World Development Indicators); (4) a dummy for the political regime (Bjørnskov and Rode, 2020); (5) the average district magnitude (Bormann and Golder, 2013); and (6) a dummy that takes the value of 1 if voting in the elections is compulsory by law (also taken from the IDEA project). Finally, we also experiment with an Instrumental Variables (IV) estimator, where we use as external instruments for democracy the mortality rate of settlers during the colonial era (Acemoglu et al, 2001). In low-disease/low-mortality regions, European settlers tried to replicate the good institutional quality of their home country. Thus, they placed a strong emphasis on the creation of institutions to protect human rights against state repression. These institutions persisted in the post-colonisation era. Moreover, in the later period, initial settler mortality did not constrain the state from using repression through other channels besides the development of institutions. Hence, this is a valid instrument in the current setting (Rørbæk and Knudsen, 2017).
Table 2 presents the results of this exercise.18 According to Testable Hypothesis 2, state repression is negatively correlated with voter turnout. Columns 1 to 3 verify these priors when we consider all elections, add the controls and examine only elections for the legislature. The same thing applies when we use voter turnout as the dependent variable, as in Armingeon et al (2021). Finally, even in the case of the IV estimator, the theoretically expected effect of respect for human rights on voter turnout is verified.
Repression and voter turnout
|Main model||Adding controls||Only legislative elections||Voter turnout Comparative Political Dataset||IV model||IV with covariates|
|Repression||–0.04*** (–3.16)||–0.07*** (–5.22)||–0.06*** (–4.54)||–0.13*** (–6.72)||–0.22*** (–3.82)||–0.51*** (–3.42)|
Notes: The table provides fixed-effects estimation results for the effect of repression on voter turnout. The dependent variable in all columns is the share of voters that voted in the legislative or executive elections to the total number of registered voters taken from IDEA, except Column 4, which is the same variable taken from Armingeon et al (2021). Clustered t-statistics in parentheses. *, ** and *** denote statistical significance at the 10 per cent, 5 per cent and 1 per cent levels, respectively. The full set of results is presented in Appendix 6.
Elections are held in both democracies and autocracies. However, the rules of the game are not the same across political regimes. In this article, we have shown that autocracies experience a pre-electoral increase in repression. However, this increase exists only in the short run, and it appears that repression returns to its pre-electoral level rather quickly. Moreover, as expected, such an increase is absent in democracies and appears robust across various specifications. Most importantly, this effect is present when we consider the possibility of the endogeneity of elections.
Overall, the theoretical model and the empirical results presented here indicate pre-electoral repression cycles, where autocratic rulers use additional instruments to secure more votes. This is consistent with several findings in the literature, where the existence of institutional constraints limits executive discretion (Streb et al, 2009; Benito et al, 2013). The problem highlighted in the present article indicates that the absence of constraints on the government’s repressive power is crucial for the population’s welfare.
From a policy perspective, there are key considerations following these results. Specifically, special attention should be given by the international community not only to the occurrence of free and fair elections, but also to the increases in repression before the elections. Following our rationale, and given the endogenous timing of the election in non-democratic countries, one cannot always predict when these increases in repression will occur. This last feature further complicates the task of the international organisations that seek to protect human rights and the conduct of the foreign policy of democratic countries against repressive regimes.
The analysis presented may also serve as the first step to a broader examination of the nature, scope and function of non-democratic elections. In fact, several implications of the theoretical model presented here can be examined. For example, we find that a higher share of opposition voters negatively affects the level of repression. Similarly, our theoretical predictions suggest that higher exogenous (second-period) rents are associated with increased repression. Also, as we show in Appendix 1, when the opposition is more heterogeneous, the autocrat employs lower repression. These three results provide clear, empirically testable hypotheses: repression is lower in tinpot regimes, where rents in the form of exogenous natural resources are lower and there is higher political fractionalisation. However, the empirical test of these hypotheses is, of course, beyond the scope of the present article, and we thus leave this for future research.
Egorov and Sonin (2021) show that a significant fraction of elections in non-democratic countries is, in fact, free. In the present article, our focus is on electoral autocracies. In fact, parties, elections and legislatures coexist with authoritarian governments in these hybrid political regimes (Schedler, 2002). Several reasons have been proposed for why elections in autocracies exist, for example: Gehlbach and Simpser (2015) and Little (2012) focus on issues of asymmetric information and signalling; Joseph (1997) argues that elections are used to increase the legitimacy of the leader; and Magaloni (2008) maintains that elections in non-democracies are employed to deter elite division.
However, this advantage does not come without a cost. The variable is not constructed using only events-based data. Instead, it also relies on standards-based data, as reported by, for example, the Central Intelligence Agency (CIA) and Amnesty International. The data, then, are available on a yearly, instead of a monthly, basis. Thus, we are not able to determine the exact timing of the repression cycle.
We do not model the underlying motives of the ruler for holding elections, but only examine their behaviour at the election years. Providing a formal model for these motives is, however, beyond the scope of this article.
Similarly, as with , higher corresponds to a higher density and higher homogeneity of preferences for the supporters.
Loss of legitimacy can, of course, have a dynamic effect by changing the number of supporters over time. Given the fact that we are interested in the short-run dynamics of the electoral cycle, we do not model this dynamic effect.
I would like to thank an anonymous reviewer for suggesting this argument.
Alternatively, we can assume that the probability of remaining in power and the vote share have an inverse U-shaped relationship. This latter assumption would only impose an additional constraint that the optimal level of R is below the level that attains the optimal share of votes that keeps the incumbent in power. However, the main result of the model regarding the increase in repression in election years will still hold.
Of course, in a richer model, R2 might have other effects. For example, it could change the elasticity of labour supply and, thus, increase total government revenues (as in, for example, Wintrobe, 1998: ch 8).
In Appendix 1, we discuss the additional results and insights of the theoretical model.
The countries in our sample are listed in Appendix 2.
As the repression variable is derived from a model of latent variables, it comes with an associated standard error. In our case, this uncertainty is incorporated in the standard error of the first-stage model (see Fariss, 2019).
We follow Imbens (2004) and truncate the estimated propensity score to 0.05 and 0.95, so that no observation takes a very high weight, ensuring in this way that our results are not driven by a specific observation. However, we examine the robustness of our results in a model with a non-truncated propensity score.
To estimate the ATE, the model relies on three assumptions: (1) conditional independence, that is, after conditioning on the covariates, the outcomes are conditionally independent of the potential outcome; (2) overlap, that is, each treated observation has a positive probability of being allocated to each treatment level; and (3) independent and identically distributed (i.i.d.) sampling, which, in our setting, rules out interactions between countries in each period (for more details on the assumptions, see Angrist and Pischke, 2009; Imbens and Wooldridge, 2009). To inspect visually whether the overlap assumption holds, in Figure A1 in Appendix 4, we present the smoothed densities of the estimated propensities of the treatment and control units, using a standard Epanechnikov kernel. As the reader can verify, considerable overlap is found among treated and control propensities, with the control observations covering almost all truncated estimated probabilities of the treated observations. This evidence provides support for the required overlap assumption and gives suggestive evidence in favour of our empirical strategy.
The first-stage results, as presented in Appendix 5, indicate that the presence of elections in the past is a consistent and strong predictor for present elections. This suggests that in electoral regimes, even though the timing of elections is endogenous, in most cases, autocrats appear to be constrained in holding elections.
Variable definitions, sources and descriptive statistics are given in Table A2 in Appendix 3.
In Appendix 5, we present the first-stage results for each panel of Table 1.
The full set of results are presented in Appendix 6.
This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.
I would like to thank Sofia Tsarsitalidou, Achilleas Vassilopoulos, Michalis Chletsos, Nikos Mylonidis, Stamatia Ftergioti, Evi Tsavou and Sotiris Papaioannou for their valuable comments and suggestions. I am also grateful to two anonymous reviewers and the editor of the journal, whose comments improved the substance and exposition of the article.
Conflict of interest
The author declares that there is no conflict of interest.
Acemoglu, D., Johnson, S. and Robinson, J.A. (2001) The colonial origins of comparative development: an empirical investigation, American Economic Review, 91(5): 1369–401. doi: 10.1257/aer.91.5.1369
Alesina, A., Cohen, G.D. and Roubini, N. (1993) Electoral business cycle in industrial democracies, European Journal of Political Economy, 9(1): 1–23. doi: 10.1016/0176-2680(93)90027-R
Angrist, J.D. and Kuersteiner, G.M. (2011) Causal effects of monetary shocks: semiparametric conditional independence tests with a multinomial propensity score, Review of Economics and Statistics, 93(3): 725–47. doi: 10.1162/REST_a_00109
Angrist, J.D., Jordi, O. and Kuersteiner, G.M. (2018) Semiparametric estimates of monetary policy effects: string theory revisited, Journal of Business & Economic Statistics, 36(3): 1–17.
Apolte, T. and Gerling, L. (2018) Youth bulges, insurrections and Labor-market restrictions, Public Choice, 175(1): 63–93. doi: 10.1007/s11127-018-0514-8
Armingeon, K., Engler, S. and Leemann, L. (2021) Comparative Political Data Set 1960–2019, Zurich: Institute of Political Science, University of Zurich, www.cpds-data.org/.
Austin, L. (2011) The politics of youth bulge: from Islamic activism to democratic reform in the Middle East and North Africa, The SAIS Review of International Affairs, 31(2): 81–96. doi: 10.1353/sais.2011.0019
Benito, B., Bastida, F. and Vicente, C. (2013) Creating room for manoeuvre: a strategy to generate political budget cycles under fiscal rules, Kyklos, 66(4): 467–96. doi: 10.1111/kykl.12032
Bhasin, T. and Gandhi, J. (2013) Timing and targeting of state repression in authoritarian elections, Electoral Studies. Special Symposium: The New Research Agenda on Electoral Integrity, 32(4): 620–31.
Bjørnskov, C. and Rode, M. (2020) Regime types and regime change: a new dataset on democracy, coups, and political institutions, The Review of International Organizations, 15(2): 531–51.
Bormann, N.C. and Golder, M. (2013) Democratic electoral systems around the world, 1946–2011, Electoral Studies, 32(2): 360–9. doi: 10.1016/j.electstud.2013.01.005
Bratton, M. (2008) Vote buying and violence in Nigerian election campaigns, Electoral Studies, 27(4): 621–32. doi: 10.1016/j.electstud.2008.04.013
Bueno de Mesquita, E. (2010) Regime change and revolutionary entrepreneurs, American Political Science Review, 104(3): 446–66. doi: 10.1017/S0003055410000274
Cheibub, J.A., Gandhi, J. and Vreeland, J.R. (2010) Democracy and dictatorship revisited, Public Choice, 143(1): 67–101. doi: 10.1007/s11127-009-9491-2
Collier, P. and Vicente, P.C. (2012) Violence, bribery, and fraud: the political economy of elections in sub-Saharan Africa, Public Choice, 153(1): 117–47. doi: 10.1007/s11127-011-9777-z
Davenport, C. (1997) From ballots to bullets: an empirical assessment of how national elections influence state uses of political repression, Electoral Studies, 16(4): 517–40. doi: 10.1016/S0261-3794(97)00032-2
Deckert, J., Myagkov, M. and Ordeshook, P.C. (2011) Benford’s law and the detection of election fraud, Political Analysis, 19(3): 245–68. doi: 10.1093/pan/mpr014
Drazen, A. and Eslava, M. (2010) Electoral manipulation via voter-friendly spending: theory and evidence, Journal of Development Economics, 92(1): 39–52. doi: 10.1016/j.jdeveco.2009.01.001
Egorov, G. and Sonin, K. (2014) Incumbency advantage in Non-democracies, National Bureau of Economic Research, NBER Working Paper No. 20519.
Eibl, F. and Lynge-Mangueira, H. (2017) Constraints, competition, and competitiveness: explaining the non-linear effect of democratization on political budget cycles, European Political Science Review, 9(4): 629–56. doi: 10.1017/S1755773916000205
Fariss, C.J. (2014) Respect for human rights has improved over time: modeling the changing standard of accountability, American Political Science Review, 108(2): 297–318. doi: 10.1017/S0003055414000070
Fariss, C.J. (2019) Yes, human rights practices are improving over time, American Political Science Review, 113(3): 868–81. doi: 10.1017/S000305541900025X
Fariss, C.J., Kenwick, M.R. and Reuning, K. (2020) Estimating One-sided-killings from a robust measurement model of human rights, Journal of Peace Research, 57(6): 801–14. doi: 10.1177/0022343320965670
Franzese, R.J. Jr (2000) Political cycles and the macroeconomy, Journal of Policy Analysis and Management, 19(3): 501–9. doi: 10.1002/1520-6688(200022)19:3<501::AID-PAM14>3.0.CO;2-5
Gehlbach, S. and Simpser, A. (2015) Electoral manipulation as bureaucratic control, American Journal of Political Science, 59(1): 212–24. doi: 10.1111/ajps.12122
Gehlbach, S., Sonin, K. and Svolik, M.W. (2016) Formal models of nondemocratic politics, Annual Review of Political Science, 19(1): 565–84. doi: 10.1146/annurev-polisci-042114-014927
Geys, B. (2006) Explaining voter turnout: a review of Aggregate-level research, Electoral Studies, 25(4): 637–63. doi: 10.1016/j.electstud.2005.09.002
Han, L. (2014) Are elections in autocracies a curse for incumbents? Evidence from Chinese villages, Public Choice, 158(1): 221–42. doi: 10.1007/s11127-012-0004-3
Howard, M.M. and Roessler, P.G. (2006) Liberalizing electoral outcomes in competitive authoritarian regimes, American Journal of Political Science, 50(2): 365–81. doi: 10.1111/j.1540-5907.2006.00189.x
Imbens, G.W. (2004) Nonparametric estimation of average treatment effects under exogeneity: a review, The Review of Economics and Statistics, 86(1): 4–29. doi: 10.1162/003465304323023651
Imbens, G.W. and Wooldridge, J.M. (2009) Recent developments in the econometrics of program evaluation, Journal of Economic Literature, 47(1): 5–86. doi: 10.1257/jel.47.1.5
Jordà, Ò. and Taylor, A.M. (2016) The time for austerity: estimating the average treatment effect of fiscal policy, Economic Journal, 126(590): 219–55.
Jordà, Ò., Schularick, M. and Taylor, A.M. (2016) The great mortgaging: housing finance, crises and business cycles, Economic Policy, 31(85): 107–52.
Joseph, R. (1997) Democratization in Africa after 1989: comparative and theoretical perspectives, Comparative Politics, 29(3): 363–82. doi: 10.2307/422126
Kaempfer, W.H. and Lowenberg, A.D. (1988) The theory of international economic sanctions: a public choice approach, American Economic Review, 78(4): 786–93.
Kaya, R. and Bernhard, M. (2013) Are elections mechanisms of authoritarian stability or democratization? Evidence from postcommunist Eurasia, Perspectives on Politics, 11(3): 734–52. doi: 10.1017/S1537592713002119
Kuvshinov, D. and Zimmermann, K. (2019) Sovereigns going bust: estimating the cost of default, European Economic Review, 119(C): 1–21. doi: 10.1016/j.euroecorev.2019.04.009
Levitsky, S. and Way, L.A. (2010) Competitive Authoritarianism: Hybrid Regimes after the Cold War, New York: Cambridge University Press.
Lipset, S.M. (1959) Some social requisites of democracy: economic development and political legitimacy, American Political Science Review, 53(1): 69–105. doi: 10.2307/1951731
Little, A.T. (2012) Elections, fraud, and election monitoring in the shadow of revolution, Quarterly Journal of Political Science, 7(3): 249–83. doi: 10.1561/100.00011078
Magaloni, B. (2008) Voting for Autocracy: Hegemonic Party Survival and its Demise in Mexico, 1st edn, Cambridge: Cambridge University Press.
Mandon, P. and Cazals, A. (2019) Political budget cycles: manipulation by leaders versus manipulation by researchers? Evidence from a meta-regression analysis, Journal of Economic Surveys, 33(1): 274–308. doi: 10.1111/joes.12263
Nordås, R. and Davenport, C. (2013) Fight the youth: youth bulges and state repression, American Journal of Political Science, 57(4): 926–40.
Nygård, H.M. (2020) Timing matters: the impact of regularity of election cycles on autocratic stability, Electoral Studies, 66: 102167.
Richards, D.L. and Gelleny, R.D. (2007) Good things to those who wait? National elections and government respect for human rights, Journal of Peace Research, 44(4): 505–23. doi: 10.1177/0022343307078946
Robinson, J.A. and Torvik, R. (2009) The real swing voter’s curse, American Economic Review, 99(2): 310–15. doi: 10.1257/aer.99.2.310
Rørbæk, L.L. and Knudsen, A.T. (2017) Maintaining ethnic dominance: diversity, power, and violent repression, Conflict Management and Peace Science, 34(6): 640–59.
Schedler, A. (2002) Elections without democracy: the menu of manipulation, Journal of Democracy, 13(2): 36–50. doi: 10.1353/jod.2002.0031
Schuknecht, L. (1996) Political business cycles and fiscal policies in developing countries, Kyklos, 49(2): 155–70. doi: 10.1111/j.1467-6435.1996.tb01390.x
Shi, M. and Svensson, J. (2006) Political budget cycles: do they differ across countries and why?, Journal of Public Economics, 90(8–9): 1367–89. doi: 10.1016/j.jpubeco.2005.09.009
Streb, J.M., Lema, D. and Torrens, G. (2009) Checks and balances on political budget cycles: cross-country evidence, Kyklos, 62(3): 426–47. doi: 10.1111/j.1467-6435.2009.00444.x
Veiga, F.J., Veiga, L.G. and Morozumi, A. (2017) Political budget cycles and media freedom, Electoral Studies, 45: 88–99. doi: 10.1016/j.electstud.2016.11.008
Veiga, L.G., Efthyvoulou, G. and Morozumi, A. (2019) Political budget cycles: conditioning factors and new evidence, in K.H. Goetz (ed) The Oxford Handbook of Time and Politics, New York: Oxford University Press.
Williamson, S. (2021) Elections, legitimacy, and compliance in authoritarian regimes: evidence from the Arab world, Democratization, 28(8): 1483–504. doi: 10.1080/13510347.2021.1929180
Wood, R. and Gibney, M. (2010) The Political Terror Scale (PTS): a re-introduction and a comparison to CIRI, Human Rights Quarterly, 32(2): 367–400. doi: 10.1353/hrq.0.0152