Viewing Violence through a Partisan Lens: How Electoral Violence Shapes Citizens' Support for Democracy

Abstract

In many elections across the world, the regime in power uses violence to influence electoral dynamics and outcomes. What is the effect of such violence on citizens' attitudes to democracy? We argue that the effect of government-perpetrated electoral violence on citizens' democratic commitment will diverge depending on whether the individual supports the ruling or opposition party. While those affiliated with the opposition should become more likely to support democracy in the wake of government violence, we expect those affiliated with the incumbent to support more power concentrated in the hands of the executive. We examine these expectations using cross-national, geo-referenced survey data from the Afrobarometer, alongside event data on electoral violence. We find that while incumbent supporters generally display lower baseline support for democracy in the absence of violence, violent elections do not further erode their democratic commitment. Violence is, however, associated with increased support for democracy amongst opposition supporters.

Over the past decades, electoral democracy has continuously risen worldwide and the vast majority of countries now hold competitive elections to fill the highest political offices of the state. While formal institutions of democratic governance have spread to become the global norm, many components that make elections meaningful – such as the right of citizens to cast their vote without being intimidated, or for the opposition to field candidates without being harassed – are severely compromised (Lührmann and Lindberg 2019). In the majority of cases, the government in power is the main perpetrator of electoral violence and intimidation (Daxecker et al. 2019). What are the implications of such violence on people's overall support for democratic governance?

In this study, we examine the relationship between government-perpetrated electoral violence and citizen support for democracy. Government-perpetrated electoral violence includes coercive acts against humans, property or infrastructure that are carried out by state actors or their affiliates to deliberately influence the process and outcome of elections (e.g. Birch et al. 2020).¹ Governments might, for example, engage in voter repression to reduce turnout in opposition strongholds, violently detain opposition candidates, harass journalists critical of the government or open fire on citizen electoral protests. Government-perpetrated electoral violence accompanies about a quarter of all elections worldwide (Hafner-Burton et al. 2014). Despite its prevalence, we have quite limited knowledge about how such violence influences people's political attitudes. Existing studies, which often focus on specific countries, suggest that electoral violence frequently leads voters to withdraw from the electoral process and could even make voters less supportive of elections overall (see, for example, Bratton 2008; Gutierrez-Romero 2014; Gutiérrez-Romero and LeBas 2020; Höglund 2009; Rosenzweig 2023). Two recent cross-national studies, drawing on survey data from Africa, corroborate that respondents who report being fearful of electoral violence are less likely to support democratic principles and more likely to favour a return to autocracy (Borzyskowski et al. 2021; Burchard 2015).

We build on these important studies but draw attention to one potential source of heterogeneity that we believe shapes how citizens respond to election violence, namely partisanship. We argue that individuals' partisan affiliation conditions the effect of government-perpetrated electoral violence on citizens' democratic attitudes. When violence is perpetrated along partisan lines, electoral violence will enhance partisan polarization, with different threat perceptions and different cost–benefit calculations about which institutions are most desirable, depending on citizens' relationship with the current regime. In short, we expect that government-perpetrated violence strengthens opposition voters' support for democracy, as these institutions constrain the discretionary powers of the executive and safeguard the opposition's political and civil rights. Incumbent supporters, on the other hand, should become more likely to tolerate more authoritarian governance as a way to ensure their safety and stability in an insecure political environment.

We examine the relationship between government-perpetrated electoral violence and citizen support for democratic governance using survey data from the Afrobarometer covering 18 countries in Africa between 2008 and 2012.² We measure our independent variable using geo-referenced event data on electoral violence from the Electoral Contention and Violence dataset (ECAV) (Daxecker et al. 2019), matched to the geo-location of the Afrobarometer survey respondents.³ Our empirical results support the argument that opposition supporters tend to be even more resolute supporters of democracy in a local context with government-perpetrated electoral violence. For incumbent supporters we do not identify any clear effect: they display lower baseline support for democratic institutions in a non-violent context, but electoral violence does not shift these attitudes.

Our results show that the partisanship of the individual is important for understanding how citizens' attitudes towards democracy are influenced by election-related violence. This contribution is particularly important at a time when there is growing concerns about democratic recession and autocratization worldwide (Lührmann and Lindberg 2019). Citizens' support for democracy can help underpin democratic longevity and shape the prospect of democratization (Claassen 2020; Lipset 1959; Mattes and Bratton 2007). Where citizens are not willing to check leaders who undermine democratic processes and punish them at the ballot box, it creates space for aspiring autocrats to further dismantle checks and balances and stifle political competition (e.g. Levitsky and Ziblatt 2018; Svolik 2020). It is hence encouraging that our results suggest that violence might not uniformly translate into reduced support for democracy, but actually make some segments of the population more astute defenders of democratic processes. At the same time, it is worrying that incumbent supporters are not marked by electoral violence in the same manner. These findings align with several recent studies suggesting that electoral winners might be more tolerant of violations of democratic norms by the sitting government, as long as it serves their political goals or favours their own party (e.g. Mazepus and Toshkov 2022; Singer 2018, 2023).

Existing literature

Over the past decade, the relationship between electoral politics and violence has been the subject of a growing literature (see e.g. Staniland 2014). Much research has examined when and where violent strategies to shape electoral contests take hold (for a recent review, see Birch et al. 2020). We still know less, however, about how electoral violence shapes citizens' attitudes and political behaviour. A recent surge in micro-level work on the consequences of civil war and other forms of violent contention has enhanced our knowledge about the social and political legacies of violence more broadly (Davenport et al. 2019). Yet, due to its specific link to the electoral process and its direct political bearings, electoral violence might have especially profound effects on citizen engagement and support for democratic political institutions (e.g. Höglund 2009).

In this study, we are concerned with the micro-level effects of government-perpetrated electoral violence and how violent electoral politics shape individual support for democratic government. Most existing work on the political consequences of electoral violence has studied behavioural outcomes such as turnout and vote choice. These studies report divergent results, positing that violence can have both mobilizing and demobilizing effects on the exercise of democratic citizenship. On the one hand, evidence from Nigeria and Sri Lanka shows that voters threatened by violence were less likely to vote (Bratton 2008; Collier and Victente 2014; Höglund and Piyarathne 2009), and Sebastian van Baalen (2024) finds that electoral violence depressed turnout in Cote d'Ivoire, particularly in areas where the incumbent was confident of winning. Yet, Dorina Bekoe and Stephanie Burchard (2017) find no significant relationship between pre-election violence and voter turnout across Africa (see also Rosenzweig (2023) on Kenya). Emilie Hafner-Burton et al. (2018) even report higher turnout following government intimidation and threats, a pattern consistent with qualitative evidence of coercive strategies for voter mobilization in, for example, Ethiopia (Human Rights Watch 2010), Egypt (Blaydes 2011) and Burundi (Travaglianti 2014). Burchard (2020) similarly reports evidence from Africa, based on survey data, that voters supporting opposition candidates may be more intent on voting if they fear electoral violence, indicating that violence could have a mobilizing effect. Since electoral violence may be used both to depress voter turnout among the opponents' supporters and to persuade voters to vote in favour of their own party, the overall effects of violence committed by an actor may be hard to identify in the aggregate.

Studies examining electoral violence's downstream effects on political attitudes are markedly fewer. Burchard (2015) finds that individuals who are fearful of electoral violence are not only less likely to rate their own country as democratic but also display lower levels of support for democracy in the abstract. Based on similar data, Inken von Borzyskowski et al. (2021) find that survey respondents who are fearful of electoral violence generally express lower support for and satisfaction with democracy in their country, rate the quality of their democracy lower and are more supportive of a return to autocratic rule. When individuals come to associate elections with threat, intimidation and use of force, they argue, it diminishes their support for electoral competition, as well as democracy more broadly. The proposed mechanism is informed by existing research showing that threats and negative campaigning create disillusionment with the political process and lead to disengagement (cf. Ansolabehere et al. 1999).

We build on this important work but explore a potentially important nuance: the effects of government-perpetrated election violence may not be uniform across the electorate, but conditioned by whether the individual supports the regime or opposition. In short, we expect electoral violence to polarize the electorate along partisan lines and produce divergent threat perceptions that lead incumbent supporters to endorse more political power in the hands of the incumbent, and opposition supporters to oppose it. We develop this argument and the expectations that follow from it below.

The polarizing effect of electoral insecurity

An important building block that is not fully recognized in most existing theorizing is the role of social group identities in shaping political attitudes in the wake of violent electoral contention. Group identities represent a central variable in research on, for example, distributive politics, violence and public opinion, and constitute a powerful prism through which citizens make sense of and engage with the political world (e.g. Vallone et al. 1985).⁴ Contexts of competition, violence and threat make in- and out-group identities even more salient (Bauer et al. 2014; Tajfel and Turner 1979). Whereas most existing research on political violence has focused on ethnicity as the most potent source of group belonging, partisan affiliations are also a powerful social identity that might be especially important for citizens' political attitudes in the wake of violent elections.⁵

Across the world, partisanship constitutes a central cleavage in the electoral competition over political power and influence. In violent elections, specifically, partisan loyalties structure not only the patterns of voter mobilization but also patterns of threats, intimidation and coercion (Rauschenbach and Paula 2019). One well-known feature of the social transformation of societies exposed to violence is that the cleavage along which violence is committed becomes more salient (Wood 2008). Scholars have, for example, shown that violence perpetrated along ethnic group lines is associated with greater in-group cohesion and negative and distrustful evaluations of the out-group (Choi and Bowles 2007; Lupu and Peisakhin 2017; Tajfel and Turner 1979). Violence also leads political attitudes and behaviour to be expressed along group cleavages (e.g. ethnic voting in conflict-ridden contexts) (Hadzic et al. 2017; Lupu and Peisakhin 2017). Partisan identities, and in particular the distinction between regime and opposition supporters, thus become especially salient in contentious elections. In Zimbabwe, for example, Adrienne LeBas (2006) describes how electoral violence led political processes to be reconfigured along the incumbent/opposition divide and intensified partisan divisions. Whereas political parties and partisan attachment are often portrayed to be relatively weak in Africa, survey data suggest that a majority of African voters do feel close to a political party (Michelitch 2015) and that partisan affiliation is one of the strongest predictors of political participation (Kuenzi and Lambright 2011). Party support has also been shown to be relatively stable over time in African contexts (Lindberg 2005). Strong incumbency advantages and regime durability in the African context make the opposition/incumbent divide particularly entrenched.

Based on this reasoning, electoral violence should harden group divisions along the incumbent/opposition divide. Electoral violence has explicit ties to the electoral contest; it tends to have political parties as the main protagonists; and the risk of victimization and abuse also follows partisan lines. As a consequence, individuals' assessments of both the level of threat they face and their quest for protection become tied to their party alignment.⁶ Research corroborates the strength of partisan-motivated reasoning: how citizens seek out and evaluate information in ways that rely on and conform to pre-existing partisan beliefs and attitudes (e.g. Bolsen and Palm 2019; Druckman et al. 2013). We propose that these political identities will also influence how citizens perceive electoral violence and that supporters of the incumbent party in particular will be more willing to excuse, perhaps even legitimize, electoral misconduct by the regime. Some existing research points in this direction. Ursula Daxecker and Hanne Fjelde (2022) show that co-partisans are likely to see electoral violence by their own party as less consequential violations of electoral integrity than violence involving the rival party, consistent with the notion that they interpret partisan violence as legitimate and perhaps even necessary in a contested electoral setting. Based on a survey experiment, Annekatrin Deglow and Hanne Fjelde (2023) find that opposition supporters in Nigeria, when informed about electoral violence, become less likely to tolerate government violations of democratic rights. And Nathan Batto and Emilie Beaulieu (2020) find that extreme partisan supporters of the legislative minority in Taiwan evaluate both their party representatives and the workings of the institutions more positively following a brawl. Based on these observations, we argue that electoral violence overall should harden group boundaries and solidify existing attachments to political parties along the incumbent/opposition divide and that citizens' political preferences in the wake of violence will diverge, conditional on their political affiliations.

Existing research shows that incumbent and opposition supporters often have very different stakes when it comes to defending democratic institutions: electoral winners are more likely to accept the weakening of checks and balances and support the regime, even when it violates democratic norms (Deglow and Fjelde 2023; Mazepus and Toshkov 2022; Moehler and Lindberg 2009; Singer 2023). Especially in weakly consolidated contexts, political institutions are not seen as neutral arbitrators in struggles over political power between groups but represent structures that allocate power to some groups at the expense of others. As noted by Mazepus and Toshkov (2022), incumbent supporters – whose interests are represented by those holding power – overall have fewer incentives to propagate for the checks and balances that democracy entails. Opposition supporters on the other hand – excluded from power – are particularly vulnerable to executive abuse and violations of democratic rights. They thus have a vested interest in vouching for democratic institutions, which constrain the power of the regime. We argue that government-perpetrated electoral violence will reinforce these positions, leading opposition supporters to become more pro-democratic in their political preferences, whereas incumbent supporters become more accepting of authoritarian forms of government.

Opposition supporters

For opposition voters, we expect violence perpetrated by the incumbent to be associated with increased support for democratic institutions. Such violence will enhance general threat perceptions, and opposition voters are the most likely targets of executive abuse. Several studies show that electoral violence tends to be used strategically as a voter demobilization tool, targeted at opponents to make them refrain from casting their vote and to disengage from the political process (e.g. Gutierrez-Romero 2014; Rauschenbach and Paula 2019). Coercive tactics by the regime place opposition supporters at especial risk of being targeted, both relative to non-partisans and to incumbent supporters. In this context we expect opposition supporters to voice stronger support for democratic forms of government. Democracy entails strong mechanisms of accountability, providing mechanisms that mitigate threats to opposition voters and strengthening their chance of gaining political power. Core democratic features, such as competitive elections, legislative checks on the executive and political freedoms all serve to protect dissenting political voices and help to insulate the opposition from intimidation, harassment and violence (Davenport 2007; Hafner-Burton et al. 2014).

Some studies suggest that electoral violence induces fear and makes voters disillusioned with and less supportive of democratic processes (e.g. Borzyskowski et al. 2021). Although this might be true for the general population, we expect a more directional response from opposition supporters. Partisan ties are important for shaping both the perception of and the behavioural response to violence and repression. Indeed, some research on violence and voter turnout shows that government violence can mobilize and bring opposition voters to the polls (Burchard 2020). Studies also show that state repression, more generally, can induce anger, leading to increased mobilization and the growth of dissident movements (e.g. Aytac and Stokes 2019; Davenport and Inman 2012). S. Erdem Aytac and Susan Stokes (2019), for instance, find that the effect of state coercion on an individual's willingness to join an anti-government protest is conditional on partisanship. In addition to the anger-provoking effect, opposition supporters might, through their political ties, have a sense of collective efficacy that makes ‘resistance’ also seem more viable. Lauren Young (2020), for instance, shows that self-efficacy is a strong predictor of intention to take action in support of the opposition after episodes of pro-incumbent violence in Zimbabwe (see also Bratton 2008). In sum, for those identifying with the political opposition, anger and collective efficacy can become tools of mobilization, strengthening their pro-democratic preferences in the wake of violence.

Incumbent supporters

We expect incumbent supporters to react differently to state-perpetrated electoral violence and overall become more supportive of more authoritarian forms of government. First, due to the mechanisms of motivated reasoning discussed above, we expect incumbent supporters to be less likely to react negatively towards the use of violence by the government. Partisan identities provide powerful filters through which citizens assess and interpret information and, as already discussed above, these are exacerbated in times of conflict. Incumbent supporters are likely to interpret pro-incumbent violence more benignly compared to non-partisans of the regime, downplaying or justifying it, rendering them less likely to react with outrage. Some existing work has explored the partisan gap in political attitudes between incumbent and opposition supporters in reaction to violations of democratic norms, documenting clear partisan bias. Citizens who feel close to ruling parties are thus more likely to support a government even if it violates democratic principles and to downplay the severity of violations of electoral integrity (Aarslew 2022; Daxecker and Fjelde 2022; Flesken and Hartl 2018; Mazepus and Toshkov 2022; Moehler and Lindberg 2009; Singer 2018). Incumbent supporters might hence be willing to tolerate, perhaps even condone, violent electoral misconduct without it provoking a political backlash if it provides politically favourable outcomes to their group. Jean Lachapelle (2022) notes how government repression in many contexts not only serves to eliminate a threat but also serves as a legitimization strategy for the sitting regime, as it signals the regime's capacity and willingness to protect loyal bystanders against an outside threat. In this context, less democratic forms of government that concentrate more power in the hands of the executive might be seen as a tool to protect and advance the interest of the own group.

Even if incumbent supporters are not themselves at great risk of direct victimization from government-perpetrated violence, a contentious electoral environment will breed a general sense of insecurity and tension. Existing research links threat perceptions to political intolerance against the ‘other’ and makes citizens more likely to accept restrictions on the rights of political opponents and to trade off liberal democratic principles for personal safety and self-protection (e.g. Conrad et al. 2018; Davis and Silver 2004). Incumbent supporters are less likely to blame the increased tensions on the incumbent (Bisgaard 2015; Tilley and Hobolt 2011), instead calling for measures to rein in the opposition-based threat and rally support behind their own representatives (Lynch 2014; Wilkinson 2004). These theoretical arguments produce the following two hypotheses for empirical testing:

Hypothesis 1: Government-perpetrated electoral violence will increase support for democracy among opposition supporters.

Hypothesis 2: Government-perpetrated electoral violence will decrease support for democracy among incumbent supporters.

Data and research design

We examine the relationship between government-perpetrated electoral violence and citizens' support for democracy using survey data from the Afrobarometer (BenYishay et al. 2017).⁷ Specifically, we combine geo-referenced data on respondents surveyed in the fourth and fifth rounds of the Afrobarometer across 18 African countries over the years 2008 to 2013.⁸ We include countries that held national elections within an 18-month window before the fielding of a given round of the Afrobarometer survey. As such, the specific features of the election and the violence associated with it will be sufficiently close in mind to influence political attitudes. For the countries included and the survey period for each of them, see Appendix, Table B1, in the Supplementary Material.

Dependent variable

Our outcome variable of interest is Support for democracy. This variable is constructed as an index, based on four questions in the Afrobarometer data. The first of these questions asks the degree to which the respondent agrees with the following statement: ‘Democracy is preferable to any other form of government.’ The remaining three questions assess the degree to which the individual supports or rejects three common forms of non-democratic rule: one-party rule, military rule or presidential autocracy. For our index construction, we recode the four variables from the Afrobarometer to be on the same scale, where support for more autocratic forms of government is represented by a lower score and more democratic as higher; we rescale all components to range from 0 to 1 and then multiply the items. The measure thus captures whether an individual who states that he/she supports democratic modes of government also rejects autocratic alternatives (see Mattes and Bratton 2007). Notably, these questions are concerned with regime types ‘in the abstract’ – that is, whether the respondent supports the general idea of democracy as a principle, not their satisfaction with democracy or how it is working in their country. This is important in light of our theoretical argument that predicts changes in individuals' appraisals of democracy as a form of government, and not declining support for a particular regime. For many of the new electoral regimes in Africa, this distinction is important. Many respondents may be discontented with how their elected regimes are performing, but nevertheless voice strong support for democratic forms of government. Government-perpetrated electoral violence may bring down the evaluation of democratic performance, while at the same time leading to stronger support for the importance of democratic institutions such as free and fair elections.

Independent variable

Our independent variable is the occurrence of Government-perpetrated electoral violence in an individual's geographical proximity. From the ECAV dataset, we consider all events where the initiator of the violence is affiliated with the incumbent (Daxecker et al. 2019). This includes events where the violence was carried out by state actors such as the military or police, but also violence that is carried out by militias or other informal groups ‘on behalf’ of or ‘in favour’ of the incumbent, as well as violence carried out by ruling party officials or party supporters. The ECAV notes whether or not a perpetrator was pro-government or anti-government, which forms the basis for our coding. We include all electoral violence events that occurred within the period covered by the ECAV for the national elections in our study. Previous research has shown that citizens' attitudes are more affected by violence in their proximity than by violence further away (Linke 2013). We combine the geographical information about the location of electoral violence with the geo-reference of survey respondents in the Afrobarometer data to create a measure of the occurrence of government-perpetrated electoral violence in relative proximity to the individual. To construct our independent variable, we establish a 50 km buffer zone around each respondent and create a variable taking the value of 1 if any event of government-perpetrated electoral violence occurred within this buffer, and 0 otherwise.

Conditioning variable

Our conditioning variable – Partisanship – is measured by a categorical variable with three categories: incumbent supporter, opposition supporter and non-partisan. The classification of respondents into each category is based on a question asking whether an individual feels close to a political party and, if so, which party that is. Incumbent supporters are those who state that they feel close to the incumbent's party. Incumbency is coded based on available information about which party holds executive power prior to the last national election in the country.⁹ Opposition supporters denote respondents who feel close to a political party that is not the incumbent party and non-partisans are those stating that they do not feel close to any party.

Control variables

We include a set of sociodemographic control variables that have been linked to individual support for democracy in previous research and can also be associated with partisan affiliation and exposure to violence. Modernization theory has long suggested that levels of affluence and education affect democratic support (Lipset 1959). These features likely also affect citizens' political positions, and how vulnerable they are to instability in society. We hence include a dummy variable indicating whether the individual has attained Secondary-level education or higher. We also include an index of household Poverty, constructed as an average of five survey items, including how often in the recent past the household has gone without food, cash income, fuel to cook food or clean water. We also control for the Gender of the respondent as this likely influences both how they experience the threat of violent acts as well as democracy given that such norms, beliefs and preferences could be shaped by gender norms prevalent within society (May et al. 2010). Due to likely differences in connectivity, infrastructure and values, we control for whether the respondent is a Rural or urban resident.

There might be important features beyond the characteristics of the individual that influence our relationship of interest. In some models, we account for contextual confounders by including a control for the history of electoral violence (EV past 10 years), measured as a variable counting the number of electoral violence events within a buffer of 50 km radius from the individual's site of interview in a 10-year window before our observation period. Given the potential of spillover, we also control for other forms of political violence using the number of state-based armed conflict events in the 50 km buffer 10 years prior to our observation period using data from the Uppsala Conflict Data Program (UCDP-GED) (Sundberg and Melander 2013). We also include a variable capturing contemporaneous armed violence, capturing all state-based UCDP-GED events in the buffer. These variables are denoted SBV past 10 years, and SBV current. Finally, we also include the degree of Electoral competitiveness within each region, calculated using the logged ratio of opposition supporters per incumbent voters within the survey data.

Methodology

We test our hypotheses empirically using hierarchical regression models. Given the continuous scale of our dependent variable, we use Ordinary Least Square linear regression models with robust standard errors. We include region-level random intercepts to account for the differences between subnational units, as well as within them. Temporal trends are also likely to influence our relationship of interest, so we include round-fixed effects in our main models. Respondents in the Afrobarometer data are nested within countries with very different baseline risks both for electoral violence and support for democracy, due to, for example, historical legacies, political configurations and regime characteristics. We, therefore, include country-level fixed effects in all models, alongside the demographic and contextual control variables outlined above. As a robustness test, we also include a matched analysis using propensity scores, matching accounts for the non-random distribution of electoral violence, reducing endogeneity concerns. Using country, urbanity and events of armed conflict in the past 10 years within the 50 km buffer, we estimate the distinct likelihood of each respondent having government-perpetrated electoral violence in their proximity. These likelihoods are used as weights in regression models that estimate the average ‘treatment’ effect of experiencing government-perpetrated electoral violence in the most recent election. We split the data into three, based on the partisanship variable, and estimate the average effect of violence using the propensity score weights for each of the partisan groups. This ensures that the core comparison is between respondents of the same partisan affiliation residing in areas with similar propensity to experience violence. Balance tables and box plots showing the balance between the treatment and control groups before and after matching are shown in Table C4, Figure C1 in the Supplementary Material.

Results

We proceed to examine our theoretical expectations that the association between electoral violence and citizen support for democracy is conditional on whether the individual supports the incumbent or an opposition party. Our main results are presented in Table 1. Model 1 shows the relationship between the government-perpetrated electoral violence in the respondent's proximity and the respondent's degree of support for democracy interacted with the partisanship variables. We include the incumbent partisan and non-partisan variables in the regression and keep the opposition partisans as our reference category. With the introduction of interaction terms, the coefficient for the main constituent terms refers to the conditional effect when the value of the other constituent term is 0 (in our case an opposition supporter). The positive and statistically significant coefficient for government electoral violence thus supports our first hypothesis: that opposition supporters display more support for democratic institutions in contexts of government-induced electoral insecurity. On average, opposition respondents living in proximity to government-perpetrated electoral violence score between 0.039 and 0.060 points higher on our support for democracy index (which ranges from 0 to 1) than opposition supporters who do not. These results remain robust when including additional control variables. The estimated effect is quite large, also compared to other influential determinants of support for democracy such as urban residence or poverty, suggesting that proximity to government-perpetrated violence is – both in a statistical and substantive meaning – significantly linked to the degree of support for democracy among opposition supporters.

Table 1. Government-Perpetrated Electoral Violence and Citizen Support for Democracy

Notes: EV = electoral violence; SBV = state-based violence. Standard errors in parentheses. ⁺p < 0.10, *p < 0.05, **p < 0.01.

Turning to our second hypothesis, we do not find support for the notion that incumbent supporters come to display weaker support for democracy in the wake of government-perpetrated electoral violence. Incumbent partisans, at the baseline, score lower on the support for democracy index. The interaction term is also negative, as theorized, but only statistically significant at the 90% confidence level. The pattern is very similar for non-partisans, but here the interaction term also reaches the 95% confidence level when control variables are introduced (see Models 2 and 3). The magnitude of the total effects and their statistical significance is graphed across all three partisan categories in Figure 1 (based on Model 2 in Table 1). The estimated association between government-perpetrated violence and support for democracy is only statistically significant amongst opposition supporters. Hence, we do not find support for our second hypothesis, that government-perpetrated electoral violence would yield decreased support for democracy among incumbent supporters. Yet, the significant interaction terms in Model 4 suggest that the difference in the effect of violence for opposition partisans compared to non-partisans is statistically significant at the 95% confidence level (and at the 90% confidence level for incumbent partisans), providing evidence that partisanship moderates the association between government-perpetrated electoral violence and support for democratic government. It is also interesting to note that both non-partisans and incumbent partisans, in the absence of government violence, display weaker support for democracy compared to opposition supporters.

Figure 1. Marginal Effect of Government-Perpetrated Electoral Violence by Partisan Affiliation

In Model 3, we control for the presence of electoral violence in the past 10 years and armed conflict events in the proximity of the respondent. The main results remain virtually the same. In Model 4 in Table 1, we include a measure of opposition violence from ECAV, interacted with our partisan variables. The results for our main variables of interest remain very similar, whereas none of the relationships for opposition violence is statistically significant (see Figure D1, in the Supplementary Material). That results differ depending on the type of actor is not surprising. Government-perpetrated electoral violence is a fundamental violation of the social contract between a state and its citizens. Given that the government is formally tasked with a mandate to use force to protect the population, electoral violence at the hands of government actors could be viewed as particularly illegitimate.

Table 2 shows additional robustness tests, with Model 3, Table 1 as the baseline. In Table 2, Model 1 we use an alternative coding of the independent variable where we include events where the initiating perpetrator could not be identified. The results remain substantively similar, yet the size and statistical significance of the electoral violence variable decline somewhat. The side of the perpetrating actor thus seems to matter. Model 2 maintains the original coding of the independent variable and instead includes a control for opposition violence and clashes where the initiator was unidentified, within 50 km of the individual. The results are strengthened in comparison with the original models, with a stronger association between the variables of interest. Individuals might have an incentive to misrepresent their stated partisan affiliation if they believe the enumerators to represent government agencies by controlling for whether the respondents believed the government to have sent them. Model 3 controls for this, leaving the results largely unchanged. Finally, Model 4 includes a measure for the level of liberal democracy in the country at the time of the survey, using V-Dem's liberal democracy index (Coppedge et al. 2020). The effect of violence on citizens' degree of support for democracy could be confounded by the level of democracy that is already enjoyed. Including this control variable, the results remain the same.

Table 2. Government-Perpetrated Electoral Violence and Citizen Support for Democracy: Robustness Checks

Notes: Standard errors in parentheses. ⁺p < 0.10, *p < 0.05, **p < 0.01.

Additional robustness checks are shown in Tables C1, C2 and C3 in the Supplementary Material. First, we ran separate models with different buffer sizes around the respondents to probe whether the proximity of violent events matters for the degree to which individuals' attitudes are affected. Not surprisingly, the size of the estimates and the precision by which they can be estimated decline as we expand the buffer size beyond 50 km. As violence happens further away, it becomes less salient for the political opinions of the citizens, resulting in a weaker and less uniform correlation between government-perpetrated violence and opposition support for democracy.¹⁰ In Table C2 we limit the sample to include only countries where any events of government-perpetrated electoral violence occurred. The results remain substantially unchanged. Finally, we use propensity score matching to address selection into violence category. We split the sample based on partisanship and test whether the results hold up. A discussion of the matching procedure, alongside the estimated average treatment effects, is presented in Table C3 in the Supplementary Material. If anything, the results are strengthened, with opposition supporters showing a stronger (0.08) and statistically significant positive association between government-perpetrated electoral violence in their proximity and their degree of support for democracy. In the matched analysis, incumbent supporters and non-partisans alike continue to show no statistically significant correlation between government-perpetrated electoral violence and their support for democracy.

Whereas our study has focused on the link between government-perpetrated electoral violence and support for democracy, there are additional variables in the Afrobarometer that can help us gain more insights into the attitudes underpinning this association. We employ the same models, yet exchange the outcome variable with satisfaction with democracy and trust in the electoral commission. If our interpretation holds, we should see a stronger negative association between government violence and these outcomes among opposition supporters, while not for the other partisan groups. The results, presented in Tables D1 and D2 in the Supplementary Material and Figure 2, show support for this notion. We find a negative and statistically significant association between government-perpetrated electoral violence and satisfaction with and trust in democracy among opposition supporters. The relationship is particularly strong for trust in the electoral commission, reaching an estimate of almost −0.4. These findings are significant since they show that citizens opposing the government react strongly to state-perpetrated violence, reducing their trust in state institutions while doubling their demand for genuine democracy. There is no clear relationship between these variables among incumbent supporters. However, non-partisans yield a smaller yet negative and statistically significant estimated relationship between government-perpetrated electoral violence and trust in electoral commissions. The use of violence thus risks undermining the support for electoral institutions also among those who do not feel close to any political party.

Figure 2. Marginal Effect of Government-Perpetrated Violence by Partisan Affiliation on Additional Outcomes. Left panel: Trust in electoral commission. Right panel: Satisfaction with democracy

Conclusion

In many countries around the world, the institution of democratic elections is under threat from political leaders who use violence, coercion and intimidation to advance their electoral agendas. Electoral violence violates the rights of voters to form and express their political preferences at the ballot box without fear of threat and intimidation. Yet besides undermining electoral integrity, can such violence also be a force in democratic backsliding? We study the association between government-perpetrated electoral violence and popular support for democratic governance, combining fine-grained geo-referenced data on election violence with survey data on citizens' political attitudes. We find that those who support the political opposition become more supportive of democratic principles in areas where these are threatened by regime engagement in violence and intimidation. We suggest that in a contentious electoral environment, democratic institutions are seen as particularly important tools to safeguard the political influence of the opposition and protect their political rights. Supporters of the regime do not seem to act in the same manner. Although generally displaying weaker support for democratic governance in the absence of violence, compared to the opposition and non-partisans, their political attitudes, in our data, are unaffected by regime-perpetrated electoral violence.

Our findings tie in with studies suggesting that supporters of winning and losing parties have very different stakes when it comes to defending democratic institutions. Set against the backdrop of a global wave of democratic recession – a process often spearheaded by popularly elected leaders who undermine mechanisms of accountability and constraints – recent research has probed the conditions under which citizens will try to curtail leaders with authoritarian ambitions. Studies have shown that both the tolerance for violations of democratic norms, the willingness to dismantle checks and balances, and satisfaction with ill-performing democracies are higher among those who support the winning party, compared to losers (e.g. Mazepus and Toshkov 2022; Moehler and Lindberg 2009; Singer 2018, 2023). Citizens are more likely to support undemocratic actions by politicians from their political base (e.g. Svolik 2020). Our findings similarly suggest the presence of a wedge between opposition supporters and the rest of the electorate in terms of their support for democratic governance, a wedge that is widening in the presence of electoral violence.

That opposition supporters rally behind democratic principles in situations when these are threatened is an important finding. Much research has highlighted the importance of opposition support – losers' consent – for democratic legitimacy and consolidation (Anderson et al. 2005) and opposition actors are often an important counter-force to democratic backsliding (Arriola et al. 2023). Therefore, the finding that opposition supporters show resilience in their support for democracy in the face of violence has positive ramifications for the persistence of democratic ideals in a country. However, since opposition politicians, candidates and voters are often the targets of electoral intimidation and coercion, government-perpetrated electoral violence might still serve to severely undermine the possibility for opposition actors to mobilize to defend democracy. Cross-national research shows that government-perpetrated electoral violence can increase the probability that the regime remains in power (Hafner-Burton et al. 2018). In this context, the indifference we observe among incumbent supporters to electoral violence by the regime is especially worrisome. Those who voted for the incumbent regime are also those who can directly hold it accountable for democratic violations by placing their vote elsewhere. But this requires that they are willing to call the regime out when it violates democratic norms and condemn the use of coercion and repression. Our results suggest that electoral winners might not be susceptible to shifting their political preferences in the face of violence. Support for democracy might thus increasingly become a ‘partisan issue’ (Singer 2023).

Our study establishes a polarizing effect of electoral violence on support for democracy, leaving important avenues open for future research. First, we study the association between electoral violence and support for democracy, but we do not have the data to unpack the evaluations and trade-offs that underlie the overall patterns. Is there a partisan lens through which opposition and incumbent supporters interpret the motivation and legitimacy of the regime's use of violence? Or do they diverge in the inferences they draw regarding the role democratic checks and balances can play in constraining violence and safeguarding democratic rights? Answering these questions requires more fine-grained data on how citizens perceive electoral violence and how citizens assess these. Our results, moreover, do not reveal the motivation underpinning the shift in democratic support among opposition supporters. Whereas some research suggests that citizens' political attitudes in the wake of violence are driven by fear, other research highlights the role of anger and outrage (Aytac and Stokes 2019). Do these emotions differ across partisans depending on their association with the regime in power? Indeed, an interesting aspect of our findings is that the group of voters who are not close to a political party is the one where the difference to the opposition voters comes out most starkly. One thing that sets this group apart from the opposition voters is a sense of group efficacy afforded by a clear partisan affiliation. An important avenue for future research would be to study how individual responses to violence differ depending on the degree to which individuals are embedded in social and political networks (see Young 2020). Moreover, previous research on Sub-Saharan Africa has found that, on average, fear of electoral violence is negatively correlated with support for democracy (Borzyskowski et al. 2021). The partisan differences in how fear of electoral violence is affected by the occurrence of electoral violence, taking into account the perpetrator of the violence, would be an important avenue for future research.

In this study, we have focused on heterogeneity in citizens' reactions to electoral violence based on whether they support the winning party or the opposition or do not feel close to a political party. Yet political parties vary, not only in their juxtaposition to political power but also in their ideological orientations. Previous research has shown that political violence in democracies can breed support for right-wing parties, increased conservativism and more hawkish policies (see e.g. Getmansky and Zeitzoff 2014). An interesting avenue for further research is whether ideological orientations also shape citizens' political responses to violence. Finally, we study the association between government-perpetrated electoral violence and democratic support in Africa, a geographical region where such violence is quite prevalent and democratic institutions in many places remain weak and unconsolidated. An interesting avenue for future research is exploring the polarizing effect of electoral violence in other contexts, including Western countries where democracy is more consolidated, yet might still be vulnerable to backsliding. The stark partisan divisions in US citizens' views of the riots at the Capitol on 6 January 2021 are, for example, well documented (Pew Research Center 2022). We know less about the long-term consequences of violent events like these for citizens' support for and trust in democratic processes and how these differ across the electorate.