The conditions for reducing electoral violence through constitutional reform

Abstract

Constitution-making is promoted by the international community as a means to encourage democratization and peaceful politics. However, there is a dearth of studies investigating the conditions under which constitutional replacement can induce such positive outcomes. Notably, many constitutional reforms enhance the power of semi-autocratic leaders or institutionalize democratic backsliding. Since party representatives often participate as negotiators in constitution-making processes, and political parties are key perpetrators of electoral violence, this study suggests that constitutional reform can constrain electoral violence if the process involves negotiated agreement between political elites and the reform includes measures that increase executive constraints. I test this claim through a matched analysis of 155 constitutional reforms from 1946 to 2015. The analysis shows that constitutional replacements that result from negotiated agreements between representatives of distinct groups and constrain the executive’s ability to enact a state of emergency are associated with lower levels of ensuing government-perpetrated violence. By contrast, these measures have no association with electoral violence carried out by non-government actors, suggesting that reform of formal institutions may be less effective in constraining violence by other actors in society. The findings have significant implications for the design of constitution-making processes that promote peaceful government conduct while highlighting the need for alternative measures to constrain other electoral violence.

Introduction

The risk of electoral violence is shaped by the institutional landscape of a country. Hence, targeted institutional reform could alter this risk. Constitution-making has been a prominent tool in the international community’s efforts towards democratization and peace-building for decades. During constitution-making processes, fundamental institutions regulating political competition and government are often negotiated and redefined (e.g. Elster, 1995). ¹ Although constitutional reform often happens in contexts marked by violence between partisan actors while also being negotiated by party representatives, our understanding of the interplay between these reform processes and electoral violence remains limited. Therefore, I investigate the following question: Under which conditions can constitutional replacement encourage a reduction in electoral violence? Given the higher frequency and distinct dynamics of electoral violence and constitutional reform in hybrid regimes compared to liberal democracies, I focus on the former. ²

Constitutional replacements should not be assumed to have a uniform effect on electoral violence. Semi-autocratic leaders commonly employ constitutional reform to strengthen their hold on power (Cameron, 1998; Negretto, 2013). I propose that both the content and the nature of the process are decisive for whether a constitutional replacement can reduce electoral violence. First, reforms that result from negotiated agreements between elite-level political camps are more likely to constrain violence. Second, I propose that reforms that enhance executive constraints are more likely to curb violence. I conduct a statistical analysis of constitutional replacements worldwide from 1946 to 2015 to assess these claims. To account for endogeneity resulting from selection effects, I use optimal full matching and conduct numerous robustness tests as well as a sensitivity analysis.

My main focus is on government-perpetrated electoral violence, since this is the most common form of electoral violence (Straus and Taylor, 2012), and arguably most likely to be constrained by institutional reform. In an extended analysis, I investigate how violence by actors without links to the government is impacted. The study finds that negotiated constitutional reforms are associated with a reduction in government-perpetrated electoral violence, while such reforms appear to have no clear relationship with levels of violence by actors unaffiliated with the government. The introduction of executive constraints, broadly defined, is not correlated with ensuing levels of either form of electoral violence. While the retraction of an executive’s ability to call a state of emergency is associated with a reduction in government-perpetrated electoral violence, this association is somewhat more sensitive.

This study contributes to literature on the causes of electoral violence and outcomes of constitution-making through three main avenues. First, I build on findings that political institutions shape the risk of electoral violence (Fjelde and Höglund, 2016; Hafner-Burton et al., 2014). Constitutions regulate numerous key political institutions. By exploring the effects that constitutional reform can bring, I introduce a dynamic perspective of the institutional roots of electoral violence. In so doing, I advance knowledge about the conditions under which a decline in electoral violence can be induced through institutional change. This ties into debates about the outcomes of constitution-making, and the importance of the process compared to the content itself (see e.g. Elkins and Ginsburg, 2021b; Ginsburg et al., 2009; Samuels, 2006). Second, departing from a handful of case studies linking electoral violence to constitutional reform in sub-Saharan Africa (Bekoe, 2012; Malik, 2018), this study offers a rigorous cross-national quantitative empirical investigation of the conditions under which constitutional reform is most likely to mitigate electoral violence across contexts and time. Third, by distinguishing between electoral violence carried out by the government and by other actors, the study sheds light on both the overall risks of violence and how actors might be impacted differently by constitution-making. This corroborates research indicating distinct patterns of electoral violence between governments and their opponents (Smidt, 2016; Straus and Taylor, 2012), and provides an understanding of the underlying reasons for these different patterns. This study’s findings can inform constitution-making practices that promote peaceful electoral competition and offer insights into situations where constitution-making might fall short in the pursuit of this goal.

The nexus between electoral violence and constitutional reform

A constitution-making process, here used interchangeably with constitutional replacement and reform, is a process that involves replacing a constitution. ³ Constitution-making is often a process that supersedes regular politics (Ackerman, 1991; Elster, 1995; Landau and Lerner, 2019). Constitution-making can thereby provide a unique forum for representatives of diverse perspectives – at times with a history of violence – to deliberate and resolve key incompatibilities relating to the fundamental structures of governance and political competition (see e.g. Brandt et al., 2011; Elster, 1995; Widner, 2008). While amendments are often piecemeal alterations in response to ‘social and technological change’ (Landau and Lerner, 2019: 2), constitutional replacements mark salient moments in history. These high-profile, often supra-political processes are more likely to yield noticeable behavioral change than amendments due to the ‘higher-lawmaking’ nature of the process, the scale of change, and the resulting higher degree of awareness of the changes among the population (see e.g. Ackerman, 1991).

This study investigates the influence of constitutional reform processes on violence associated with electoral competition. Electoral violence is commonly defined as the use of harassment, repression, physical violence, intimidation, or threats with the intention of influencing the process or outcome of an election (see e.g. Birch et al., 2020: 4). Electoral violence can entail clashes between supporters of different candidates, deliberate attacks against voters or members of a political party, or attacks on politicians. It can also consist of violence toward electoral institutions, targeting polling stations and electoral administration (Harish and Toha, 2019). I distinguish between electoral violence that is perpetrated by the government and that which is carried out by non-state actors. These other actors can be opposition political parties and their supporters, or other armed actors outside the security apparatus, commonly criminal groups. Organized armed actors challenging the government through armed uprising are not included. Government-perpetrated violence refers to violence carried out by the police or military, the ruling party, or their agents. ⁴

Political institutions that are often regulated by the constitution have been shown to shape the risk of electoral violence. Fjelde and Höglund (2016), for example, find that countries with majoritarian electoral institutions are likely to experience more instances of electoral violence, theorizing that this is due to the high stakes around elections in such environments. Hafner-Burton et al. (2014) further show that institutional executive constraints can mitigate incumbents’ use of violent tactics, even when they face a threat at the polls. Further, an autonomous electoral management body reduces the likelihood of electoral intimidation (Ruiz-Rufino and Birch, 2020). Given these studies’ collective findings indicating the importance of formal institutions in shaping the risk of electoral violence, I argue that high-profile institutional reforms could alter this risk. To date, there have been no quantitative cross-national studies exploring the conditions under which changes in institutions can impact the levels of electoral violence that follow. This study addresses this gap.

Several qualitative studies explore how constitutional reforms can mitigate electoral violence. Drawing on insights from Togo and Zanzibar, Bekoe (2012) shows that post-election constitutional power-sharing reforms can lead to advances that, in the short term at least, mitigate further electoral violence by establishing forums for interaction and mutual influence between political opponents. Similarly, Aeby (2018) suggests that constitutional power-sharing reduced the risk of electoral violence persisting in Zimbabwe. Malik (2018) investigates the impact of constitutional reform on electoral violence following the 2010 reform in Kenya, which was promulgated in response to the extreme electoral violence of the 2007 election. In regions where the 2013 presidential election had high stakes, Kenya’s new electoral threshold, which favored interethnic cooperation, facilitated electoral peace. ⁵ These studies demonstrate the importance of constitutional reform in response to electoral violence.

Constitutional reform processes and electoral violence cluster in the same countries, motivating further investigation into their impact on each other, within the scope of the contexts that experience them most frequently. Both are most common in hybrid regimes. Of all replacements between 1946 and 2015, only 2% of these were in liberal democracies or in countries that did not have any electoral violence during this time frame. A country’s number of elections containing violence and its number of constitutional replacements are highly correlated. ⁶

Changes in electoral violence in the aftermath of reform

The institutional framework of a country shapes political behavior (Olsen, 2009). It follows that a reform of key institutions has the potential to alter trajectories of political behavior. I expect that the degree of election-related violence will change in response to the reformed political landscape and the process that shaped the reform.

The conditions under which constitutional replacement induces more peaceful electoral competition

Constitution-making processes share similarities with peace negotiations in terms of conflict resolution opportunities in contexts fraught with electoral violence. Nathan (2020) argues that constitution-making processes can incorporate peace-making characteristics that are similar to the peace negotiations following a civil war. Political party representatives are often essential actors in constitution-making, yet they are often the same actors who engage in election-related violence. Hence, the potential for constitution-making to promote peace can apply not only to post-conflict situations, but to any context that experiences political violence, including electoral violence. The following section outlines the two conditions proposed to impact the degree of electoral violence: negotiated agreement and increased executive constraints.

Negotiated agreement

Participation in constitution-making processes takes place in a wide range of formats. Sometimes a broad-based constituent assembly is established, consisting of representatives of a variety of political parties or interest groups (see e.g. Ginsburg et al., 2009). In other cases, a smaller drafting commission is appointed. However, the drafting processes often exclude relevant actors, as was the case in Egypt after the military had removed Morsi from power following the Arab Spring (Jamal and Kensicki, 2016). As such, a reform is at times the result of some degree of negotiation and agreement between political factions, but often it is not. This distinction matters for the outcomes that follow after a reform. Eisenstadt and Maboudi (2019) show that processes including representatives of distinct interest groups are more likely to advance democracy. Similarly, Negretto and Sánchez-Talanquer (2021a) show that when constitutional reforms are the result of agreement between political elites with distinct bases of social support, the degree of liberal democracy that follows tends to be higher. These studies reveal that the dividends of constitution-making processes are shaped by the presence or absence of inter-elite bargaining and negotiation.

Building on these insights, I argue that constitutional reform processes can promote peaceful political competition through the inclusion of a variety of political perspectives, and the practice of negotiation. Where the drafting body consists of representatives from various political factions and no party can dominate the process to determine the content of the reform, and a sufficient number of delegates must agree on the reform for it to be produced, negotiation between distinct groups is required. Mere inclusion is insufficient if the drafting process can be controlled by a dominant actor. Such processes are likely to be perceived as ‘stolen’ by the opponents of the dominant actor. Processes entailing constructive negotiation practices, and in turn successfully enacted reforms, are more likely to yield increased perceptions of legitimacy, and a reduction in violence.

Constitution-making initiatives are widely considered a core component in peacebuilding in the aftermath of armed conflicts. They provide an opportunity for political actors to engage in dialogue (Miller, 2013: 636), address disagreements over the regime, and shape political institutions that will govern the country moving forward (Samuels, 2006). Through this process, political actors can develop mutual trust and establish practices for conflict resolution through institutional means (Brandt et al., 2011; Chambers, 1998). Each of these practices fosters peaceful interactions between the political parties involved, which can be carried forward into the post-reform period. I suggest that reducing violence is more likely to be achieved through reforms that are negotiated and agreed upon by multiple political actors. Negretto and Sánchez-Talanquer (2021a) argue that the core mechanism through which negotiated agreements yield higher levels of liberal democracy in the short term is the relative balance of power among different actors, and their ability to keep each other in check. Similarly, I expect this balance of power alongside the depolarizing dynamics of negotiation to create a deterrent against violent misconduct around elections, particularly by elite-level national politicians. ⁷

Following negotiated agreements, the political elites are more likely to face pressure to continue positive interactions with political opponents and hold their side of the bargain. When opposition representatives are involved in the process, they can better act as watchdogs to ensure that the reformed constitution is implemented and interpreted as intended (Eisenstadt et al., 2015: 596–597). A negotiated reform is more likely to include provisions that protect each side and their interests (Negretto and Sánchez-Talanquer, 2021a). Further, the same political actors that were involved in negotiating the constitutional reform tend to remain in the political arena in the elections that follow (Negretto and Sánchez-Talanquer, 2021a: 4). The improved relationships between these actors are accompanied by a greater balance of power between them as a result of the reform. This ensures that the distinct sides are viewed as credible threats to each other if they engage in illegitimate actions. As such, the costs of electoral violence increase.

Moreover, a reform that brings together actors across the political divide is more likely to promote conciliatory interactions at the elite and societal levels, and to be perceived as legitimate by a breadth of the population and political elite (cf. Miller, 2010). These processes require compromise and coalition-building between a sufficient number of actors to meet the procedural requirements to pass the constitution or put it to a referendum (Negretto and Sánchez-Talanquer, 2021a). Such constitution-making processes can present an opportunity to construct a unifying narrative that mitigates polarization (Eisenstadt and Maboudi, 2019; Ludsin, 2011; Wallis, 2016: 375). Inclusive negotiation can alter the dynamics between the negotiators. These changes affect how political representatives communicate with their constituents. Once agreement is achieved, the resulting reform is more likely to promote depolarization and peaceful democratic interaction between political sides than a non-negotiated reform (cf. Jamal and Kensicki, 2016). This depolarization is likely to yield more peaceful electoral competition through relational and norm changes, both at the elite level and among party supporters.

• Hypothesis 1: Constitutional reforms resulting from negotiated agreements are more likely to be associated with a decline in electoral violence after the reform than those that are unilaterally drafted.

Increased executive constraints

A new constitution can establish regulations that decrease the incentives and opportunities to employ violence or other illegal means to win elections. Constitutional reforms in conflict-ridden settings often focus on creating institutions that aim to mitigate violence (Ludsin, 2011; Samuels, 2006; Widner, 2008). I propose that one central mechanism for reducing electoral violence is the introduction of constraints on the executive’s powers. Hafner-Burton et al. (2014) show that government-perpetrated electoral violence is lower in countries that have higher executive constraints, attributing this to the higher costs of violence faced by incumbents under such circumstances. Building on this finding, I suggest that the implementation of additional constraints will lead to a decrease in electoral violence, particularly by the government and its agents. ⁸ This institutional change influences the incentives as well as the opportunities to employ violence in elections, resulting in a decline in violence compared to before the reform was implemented.

Increasing executive constraints can lead to a reduction of violence through three mechanisms. First, as executive constraints increase, the overall stakes of the elections decline, lowering the incentives for violence. In hybrid regimes, where electoral violence is most common, neopatrimonialism often shapes the political landscape (Bratton and van de Walle, 1994; Lust-Okar, 2009). Where power is concentrated in the hands of the executive, and benefits accrue through personal connections or ties to those in power, the stakes of electoral victory become high for politicians, their affiliates, and voters (Fjelde and Höglund, 2016: 298). By constraining the amount of power held by the executive and enhancing oversight mechanisms, the stakes of elections decline. The executive’s ability to leverage their position of power for personal gain and the gain of their supporters becomes more limited, and they become more likely to face repercussions for corruption, favoritism and misconduct. Winning an election becomes less of a zero-sum game (Fjelde and Höglund, 2016: 303), reducing the incentives among executive power contenders to employ violence due to the diminished opportunities to gain additional benefits when in office. The reduced stakes imply a lower willingness to incur costs to maintain and achieve executive power. Since electoral violence is among the costliest methods of electoral manipulation, I expect a reduction in electoral violence by national-level political parties and their agents as a result.

Second, removing a government’s special privileges constrains the government’s ability to use violence. Executives often have special privileges that enhance their opportunities to use violence, such as the command of state security forces combined with the right to declare a state of emergency or rule by decree. However, if the executive no longer has special powers like the mandate to declare a state of emergency, they will have fewer options to resort to violence as a strategy. By declaring a state of emergency, the executive is authorized to restrict human rights and deploy security forces to use violence in the name of security (Neumayer, 2013). These powers not only increase the likelihood of violence by the government, but also provide an incentive for opposition supporters to resort to violence in response to the curtailment of democratic liberties. When such executive rights are constrained, fewer avenues open for government-sanctioned violence.

Finally, increased accountability measures reduce the opportunity for violence. Electoral violence is often carried out in a context of impunity (Höglund, 2009: 421). Actors employing violence to shape elections face few direct costs under such circumstances. By establishing mechanisms for accountability through an independent judiciary, increased legislative oversight, and the removal of government immunity, the costs of using violence as a strategy increase, as the likelihood of pro-government actors getting away with violence diminishes.

In sum, I expect that after constraints on the executive increase, political candidates will use less violence compared to before the reform. The expectation is strongest for violence by the government and its agents, given that the executive is more constrained.

• Hypothesis 2: Constitutional reforms that implement increased executive constraints are more likely to be associated with a decline in electoral violence after the reform than those that do not.

Cross-national analysis

My study focuses on the immediate, short-term effect of constitutional reform. The unit of analysis is the constitutional reform – the promulgation of a new constitution. ⁹ This includes information about relevant variables for the periods immediately before and after the reform. The initial data are in a panel format, which is then condensed into cases that pertain to constitutional reform containing pre- and post-reform information. To obtain pre-reform values for the variables analyzed, I rely on data from the years preceding the reform. At maximum, the variables may cover a ten-year window before or after the reform (specified further below). Post-reform values are the values from the first election year following the reform.

The study’s scope is restricted by three criteria, which narrows down the analysis to 155 cases. The first requirement is that, during the first year of the pre-period, at least ten years must have passed since the most recent constitutional reform. Second, the scope of the theoretical argument is limited to hybrid regimes. Third, only countries that have held elections within ten years before or after the reform are included. The temporal scope of the study stretches from 1946 through 2015. ¹⁰

Data

To identify instances of constitutional reform, I use a binary variable that is coded as 1 for the year when a new constitution is promulgated (comes into legal effect). This variable is used to create a cross-sectional dataset, with one observation per constitutional reform. These data are from the Comparative Constitutions Project (CCP) (Elkins and Ginsburg, 2021a). To ensure that only new constitutions are considered, I do not include reforms that simply re-introduce a previous constitution. This adaptation is based on the coding by Negretto and Sánchez-Talanquer (2021b).

The main outcome of interest is the change in the degree of government-led electoral violence. In an extended analysis, I also explore the shift in violence perpetrated by other actors. I measure this shift using the difference between pre- and post-values of violence. ¹¹ The violence variables are indices provided by V-Dem that capture the level of election-related violence carried out by government and other actors separately (Coppedge et al., 2023). ¹² Government-perpetrated electoral violence is the main focus of the article and encompasses electoral violence, harassment, and intimidation carried out by security forces, incumbent party actors, or other actors with ties to the government. All other electoral violence is grouped together in the extended analysis. The second index of outcomes measures the level of violence committed by actors that are not affiliated with the government during the campaign period, election day, and post-election process. While the theory only produces expectations for national level political elites contending in elections, perpetrators in the variable of the extended analysis may include a broader range of actors. While it covers agents and supporters of an opposition party, it also includes violence by loosely organized citizens or illicit groups, motivating its status as an exploratory extended analysis rather than an explicit test of the theory. Importantly, the theoretical argument primarily concerns the incentives and opportunities of parties and candidates within formal political opposition and their allies, rather than non-state groups involved in civil war. The coding of these variables explicitly excludes violence that relates to civil war.

The variable capturing whether the reform was the result of a negotiated agreement is coded based on replication data from Negretto and Sánchez-Talanquer (2021b). The variable is binary and denotes whether at least two different political actors were required to reach an agreement during the negotiations in order for the constitution to pass. The authors state two criteria for coding the variable, which they call plural approval: ‘(1) two or more distinct and independent political parties or groups achieved representation in the constitution-making body, either through elections, appointment, or mixed selection methods and (2) the collaboration between at least two of these parties or groups was necessary to decide on the constitution’s content and enact it, according to the decision rule governing the constitution-making body’ (Negretto and Sánchez-Talanquer, 2021a: 5). The coding of this variable is determined by the decision-making rules of each reform process, in combination with the representation of various delegates in the drafting committee. Processes that were dominated by one actor, or that excluded alternative views from the negotiations, are coded as 0, while the negotiated agreements are coded as 1. ¹³ If one political party is able to determine the outcome of the drafting process without needing to reach an agreement with another party, this process is deemed to be dominated by that party. It is not coded as a negotiated agreement, even if representatives of other parties were at the table.

The second argument posits that if constitutional reforms introduce constraints on the executive, it will encourage more peaceful political behavior. To measure increased executive constraints in the constitution, I have developed an additive index that incorporates six component variables to capture the constitution’s content. Next, I code a binary variable as 1 to indicate whether there was an increase in the index around a constitutional reform.

The components of the index are coded according to variables from the CCP data that indicate whether the constitution stipulates (1) the legislature’s right to have oversight over the executive, (2) the legislature’s power to investigate the executive’s actions, (3) judicial independence, (4) whether the executive can declare a state of emergency unilaterally, (5) whether the executive has the power to rule by decree, and (6) whether the executive is granted some degree of immunity. Given that it is an increase in executive constraints that is expected to induce the hypothesized change, it is a change upward in conjunction with a constitutional reform process is coded as 1. Otherwise, the variable is coded as 0. The downside of using the aggregate change in an index is that there may be variations that are hidden if there is an increase in one measure and a decrease in another. Theoretically, these components are distinct. The first three introduce accountability mechanisms, while the last three, which are measured differently, represent a removal of strong executive powers. I also empirically test whether there is a distinction between the effects of the subcomponents of executive constraints.

Design of the study

The distribution of constitutional reforms with their varying characteristics across countries and time is not random. Matching is an established method for addressing these problems. It involves creating sets of comparable cases, where a subset of cases has received a ‘treatment’ while others have not, and calculates the average estimated effect within these sets. The test statistic is the coefficient of the average treatment effect (ATE). I estimate the ATE by running an OLS regression with robust standard errors on the post-minus-pre values of the levels of each form of violence. Hence, the dependent variable is the difference between the level of electoral violence after the reform and the level of violence before the reform. The treatment variables are binary measures of whether a unit has been treated with either of the conditions. ¹⁴ Specifically, I use optimal full matching (see e.g. Austin and Stuart, 2015; Rosenbaum, 1991). This approach uses propensity scores to quantify the probability of an observation receiving a treatment, conditional on covariates. The two conditions – negotiated agreement and increased executive constraints – are the treatments in separate matching procedures. ¹⁵ This approach aims to minimize the distance between the observations within all sets and find matches for all observations. The sets are then used as fixed effects in the estimation models.

Matching variables

Constitutional reform is sometimes part of conflict resolution in the wake of civil war or the institutionalization of significant historical events such as revolution or regime change (Elster, 1995; Landau and Lerner, 2019: 4). For example, after the end of Apartheid, political parties in South Africa collaborated on re-drafting the constitution. Following the end of military rule in Brazil 1988 and Nigeria 1999, constitution-making processes institutionalized regime change. Conversely, processes of constitutional reform are sometimes initiated by aspiring autocrats, or institutionalize military rule, as was the case in Hungary 2012, Turkey 1982, and Venezuela 1999. Yet, frequently, they respond to demands for democratization or modernization, or a lack of popular legitimacy, without being preceded by a major rupture (Landau and Lerner, 2019: 4). All of these features – regime type, regime change, the mode of regime change, civil war, democratization demands – are likely to shape whether a reform is negotiated and whether it constrains the executive, in turn determining its impact on violence. Therefore, this study matches on variables capturing these driving factors.

The matched analysis creates balanced sets based on the following variables. ¹⁶ First, using V-Dem’s regimes of the world variable, I match on the regime type that a country had the year before the reform. Second, I match on the size of opposition to the current regime – measured as the average of V-Dem’s continuous index on regime opposition group size in the ten years before reform. Faced by a threat of a strong opposition in society, an incumbent’s incentives to negotiate with opposition parties and increase executive constraints grow (Jung and Deering, 2015). Electoral violence is also significantly more likely where the incumbent is more threatened (Hafner-Burton et al., 2014). Third, I match on whether a country has had internal armed conflict in the ten years prior to reform. Many constitutional reforms are carried out in the aftermath of armed conflict, as a step toward implementing the content of a peace agreement (Nathan, 2020), and experience with armed conflict increases the risk of electoral violence (Höglund, 2009). I use data from the UCDP-PRIO dataset starting in 1946 (Gleditsch et al., 2002). I also code whether the cases have experienced a violent regime change (war, assassination, violent coup), or a nonviolent regime change (revolution, nonviolent coup) in the ten years preceding the reform, since nonviolent regime change is more likely to involve negotiated agreements and increased executive constraints. These variables are from V-Dem.

I match on a number of additional variables. First, to account for regional trends, I match on subcontinental region to account for spill-over (West Africa, Eastern Europe, etc.). Second, constitution-making has occurred in waves (Elster, 1995). To account for temporal trends, I match based on clusters of decades (1946–59, 1960–79, 1980–99, 2000–15). Third, I account for the country’s time since independence, since younger countries undergo more constitutional reform. Finally, I calculate the country’s accumulated democratic stock – using the same method employed by Gerring et al. (2012) – since prior experience with democracy encourages inclusive processes and executive constraints. This is the cumulative level of V-Dem’s polyarchy measure across years since the country’s independence. Moreover, I include a count of the cumulative number of constitutional reforms in a country since 1900 in order to account for regime volatility and salience of the reform process at hand.

To address ceiling effects and account for pre-treatment trends, I include the prior levels of electoral violence and executive constraints. The models estimating increased executive constraints also include labor union participation since this affects the likelihood of executive constraints being introduced (Abramson and Boix, 2019). This variable comes from V-Dem.

Results and analysis

The following section outlines the results of the analysis. First, I conduct balance tests to determine whether the data are evenly distributed between the treatment and control groups. I then investigate the relationship between the characteristics of constitutional reform and government-perpetrated violence before moving on to explore how these characteristics relate to violence by other actors. Finally, I run several robustness tests and a sensitivity analysis to further investigate the main results.

Using a balance test, I compared the distance between groups based on the specified covariates. ¹⁷ Both balance tests indicate that there is no statistically significant difference between the treated and untreated groups. While the unmatched samples are relatively balanced already, the matching procedure renders them significantly more balanced. Figures A1 to A6 in the Online Appendix show the difference in standardized differences between the matched and the unmatched sample for each of the main treatments. The figures indicate that there is an overall improvement after matching, even though the samples are relatively balanced to begin with.

First, I assess the change in government-perpetrated electoral violence. The main results are presented in Table 1. In line with Hypothesis 1, Model 1 in Table 1 shows that negotiated agreement is associated with a statistically significant reduction in government-perpetrated electoral violence. The coefficient is −0.622. Given the scale of the outcome variable ranging from −3.8 to 2.2, this is a significant change, not least given the short time span over which it occurs. Model 2 in Table 1 suggests that increased executive constraints have no clear association with changes in government-perpetrated electoral violence, when measuring increased constraints using the aggregate index. The R-squared is much higher in the model including the variable for negotiated agreement. This corroborates the low explanatory power of increased overall executive constraints. Given that electoral violence often follows subnationally local dynamics (Wahman, 2023), and given the distinctly national-level character of these constraints, this may indicate that more locally adapted constraints are needed.

Table 1.

Government violence, main models.

	Model 1	Model 2
Negotiated agreement	−0.622** (0.215)
Increased executive constraints		0.217(0.270)
Opposition size	−0.047(0.064)	−0.075(0.076)
Democratic stock	0.012(0.013)	0.027(0.017)
Prior reforms	−0.055(0.039)	−0.046(0.038)
Years since independence	−0.003* (0.001)	−0.003* (0.002)
Time since reform	−0.027(0.040)	−0.019(0.038)
Prior violent regime change	0.297(0.228)	0.447(0.234)
Prior nonviolent regime change	0.424(0.244)	0.489* (0.235)
Prior electoral autocracy	−0.400* (0.182)	−0.485** (0.182)
Prior electoral democracy	0.062(0.318)	−0.403(0.367)
Recent conflict	−0.289(0.178)	−0.054(0.203)
R²	0.485	0.271
Adj. R²	0.304	0.095
N	155	155

*p < 0.05, **p < 0.01, ***p < 0.001. Standard errors in parentheses.

To further probe the effects of executive constraints, I disaggregate the index. The results presented in Table 2 reveal that the null effect above is caused by the lack of impact from most of the component variables, except for one that stands out: when the constitutional reform entails taking away the executive’s authority to single-handedly declare a state of emergency. The estimated effect is negative and relatively strong, at −0.862, and is hence in line with Hypothesis 2. It is noteworthy that this specific index component has a significant estimated effect. This measure is the same one theorized by Hafner-Burton et al. (2014), who establish a link between institutional constraints on the executive and a lower risk of an incumbent’s pre-election violence.

Table 2.

Government violence, disaggregated executive constraints.

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Removed executive SOE	−0.862* (0.385)
Increased legislative oversight		−0.117(0.310)
Increased legislative investigation			−0.404(0.246)
Judicial independence introduced				−0.156(0.297)
Removed decree rights					−0.410(0.301)
Removed immunity						0.071(0.415)
Controls	YES	YES	YES	YES	YES	YES
R²	0.229	0.212	0.267	0.201	0.238	0.200
Adj. R²	0.134	0.088	0.132	0.108	0.137	0.087
Observations	155	155	155	155	155	155

*p < 0.05, **p < 0.01, ***p < 0.001. Standard errors in parentheses.

This particular power – to declare a state of emergency – is arguably most directly associated with the use of violence. The effectiveness of the other measures depends on the quality of other institutions such as the legislature and the judiciary, which tend to be flawed in hybrid regimes. Although granting the legislature the authority to oversee and intervene provides a formalized avenue for accountability, legislatures in these regimes commonly contain a co-opted opposition unlikely to counter the government (Gandhi and Lust-Okar, 2009). Instead, abolishing the executive’s power to unilaterally declare a state of emergency eliminates their ability to legitimize their use of repression. Under a state of emergency, governments often employ more repression (Bjørnskov and Voigt, 2019), using this as a tool to enable more autocratic behavior without overtly breaching the law.

Finally, Model 2 in Table 3 includes both of the estimated statistically significant conditions combined. Only the coefficient for negotiated agreement remains significant at the 95% level, while the coefficient for removed executive SOE rights only achieves statistical significance at the 90% level. This suggests that while negotiated agreement has an impact that is independent of the removal of executive SOE rights, the opposite is more uncertain. While this may be due to the lower variation – and hence statistical power – of the removed SOE variable, it may indicate that removed SOE rights is at times a result of negotiated agreement, which drives the ensuing decline in government violence. This is interesting in light of previous research. Comparative literature has been unable to show that procedural aspects are systematically liked to the resulting content of a reform (Landau and Lerner, 2019: 8). For example, processes that include diverse group representatives are not systematically more likely to yield constitutions that significantly advance democracy (Maboudi, 2020). However, some executive constraints may be more likely when new constitutions are negotiated between distinct elite group representatives.

Table 3.

Government violence, state of emergency and negotiated agreement.

	Model 1	Model 2
Removed executive SOE	−0.862* (0.385)	−0.707† (0.372)
Opposition size	−0.111(0.072)	−0.074(0.073)
Democratic stock	0.028(0.018)	0.024(0.018)
Prior reforms	−0.043(0.041)	−0.060(0.041)
Years since independence	−0.003** (0.001)	−0.002* (0.001)
Time since reform	−0.017(0.038)	−0.027(0.038)
Prior violent regime change	0.357† (0.211)	0.272(0.210)
Prior nonviolent regime change	0.279(0.224)	0.225(0.226)
Prior closed autocracy		0.032(0.340)
Prior electoral autocracy	−0.531*** (0.157)	−0.389(0.324)
Prior electoral democracy	−0.285(0.343)
Recent conflict	−0.001(0.176)	−0.036(0.174)
Negotiated agreement		−0.585** (0.214)
R²	0.229	0.284
Adj. R²	0.134	0.184
N	155	155

†

p < 0.1, *p < 0.05, **p < 0.01. ***p < 0.001. Standard errors in parentheses.

In sum, negotiated agreements are robustly associated with a reduction in government-perpetrated electoral violence, while an increase in executive constraints is not. Instead, only the removal of the executive’s right to single-handedly declare a state of emergency is associated with reduced electoral violence by the government, although this association is less robust when accounting for negotiated agreement. As such, I find strong support for the first hypothesis with regard to government-perpetrated electoral violence, but only weak support for the second hypothesis, and only for the state of emergency constraint.

Extended analysis

The analysis also covers violence carried out by actors who are not affiliated with the government. Table 4 presents the results for the estimation of changes in other electoral violence using the main measures for the treatment variables, and Table 5 shows the results for the disaggregated index. Unlike for government-perpetrated violence, neither negotiated agreements nor increased executive constraints are associated with changes in this violence. However, removal of the executive’s right to rule by decree is estimated to have a positive association with violence by actors other than the government. Tables A1 and A2 in the Online Appendix show the full regression tables for removed executive power to rule by decree in combination with negotiated agreement, replicating this relationship. However, while results from unmatched OLS regressions, including the matching variables as controls, are substantively in line with the results of Table 4, they show no association between removed decree rights and changes in levels of violence. They are presented in Tables A15 and A16 in the Online Appendix. The estimated association in Table 5 does not hold up in the alternative matching procedures presented in Table A24 in the Online Appendix, nor is there a statistically significant bivariate association in the unmatched data. Hence, I conclude that there is no support for either Hypothesis 1 or Hypothesis 2 regarding violence by actors other than the government. Despite the estimated positive relationship between the removal of executive decree rights and increasing levels of electoral violence by other actors, which runs counter to the expected relationship, these results are insufficiently robust to draw inferences from.

Table 4.

Non-government violence, main models.

	Model 1	Model 2
Negotiated agreement	0.167(0.173)
Incr. executive constr.		0.146(0.246)
Opposition size	−0.004(0.055)	−0.049(0.049)
Democratic stock	−0.008(0.013)	−0.008(0.012)
Prior reforms	0.064(0.041)	0.015(0.034)
Years since independence	−0.002(0.001)	−0.001(0.001)
Time since reform	−0.086* (0.033)	−0.062* (0.031)
Prior violent regime change	−0.268(0.212)	−0.338(0.173)
Prior nonviolent regime change	0.284(0.217)	0.272(0.180)
Prior electoral autocracy	−0.125(0.163)	−0.021(0.141)
Prior electoral democracy	0.086(0.301)	0.258(0.268)
Recent conflict	0.310(0.157)	0.232(0.141)
R²	0.370	0.250
Adj. R²	0.110	0.112
N	155	155

*p < 0.05, **p < 0.01, ***p < 0.001. Standard errors in parentheses.

Table 5.

Non-government violence, disaggregated constraints, matched.

Variable	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Removed executive SOE	−0.152(0.251)
Inc. legisl. oversight		−0.269(0.327)
Inc. legisl. investigation			0.128(0.303)
Judicial indep. introduced				0.480(0.536)
Removed decree rights					0.561* (0.277)
Removed immunity						−0.220(0.355)
Controls	YES	YES	YES	YES	YES	YES
R²	0.142	0.185	0.172	0.160	0.150	0.184
Adj. R²	0.035	0.056	−0.004	0.063	0.044	0.069
N	155	155	155	155	155	155

*p < 0.05, **p < 0.01, ***p < 0.001. Standard errors in parentheses.

These null-results should be put in perspective with the broad scope of this outcome variable relative to the scope of the theory, which specifically regards national-level politicians. The theoretical expectations do not extend to actors other than these, and the breadth of actors included in this variable risks concealing outcomes specific to national-level opposition actors. Electoral violence often follows more localized subnational dynamics, or is carried out by loosely organized illicit armed actors (Berenschot, 2019; Turnbull, 2021; Wahman, 2023; Wilkinson, 2004). Political actors often exacerbate these dynamics by seeking plausible deniability for violence, which causes the dispersal of violence among a range of actors (see e.g. Carey et al., 2015; Matanock and Staniland, 2018; Taylor et al., 2017). The presence of non-state agents of illicit violence who employ violence may require alternative approaches beyond reforming legal institutions, and localized dynamics of violence may require distinct types of institutional constraints than those included in this study. Electoral violence, as well as other violence carried out by illicit armed groups, is inherently illegal. Hence, the actors involved have already proven their disregard for formal legal frameworks. Moreover, national-level negotiation and cooperation, as well as the introduction of national-level executive constraints, is likely insufficient to curb violence that is rooted in locally rooted dynamics.

Constitutional reforms in hybrid contexts have at times been argued to be toothless (Hale, 2011). Notably, the bureaucracy remains the same (Hassan, 2015), as do the informal institutions and networks that underpin electoral violence by actors who are not overtly tied to the state. However, some constitutional reforms may still affect some actors that are unaffiliated with the government, while not others. The effect of negotiated agreement and increased executive constraints on specifically national-level opposition party actors remains a question open for further research. Importantly, the mechanisms for the negotiated agreement are most likely to play out among the parties to the agreement. While these almost invariably include the government, their counterparts in obtaining a negotiated agreement are often only a subset of opposition parties, and never illicit groups who carry out violence, who may instead act as spoilers (cf. Stedman, 1997). Further analysis into the implications of constitutional replacements resulting from negotiated agreements must disaggregate the effects by actor more specifically.

Robustness checks and sensitivity analysis

Further tests probe the robustness of the main results of the analysis. I focus on the estimated negative correlation between a change in the level of government-perpetrated electoral violence and negotiated agreement, as well as the executive’s removed ability to declare a state of emergency. Since the balance tests show that the sample is relatively balanced between the values of the treatment variables, I run regressions omitting the matching procedure. The results show substantially similar results to the main analysis, although with sustained statistical significance of the removed SOE coefficient when controlling for negotiated agreement, are presented in Tables A13 to A14.

Second, I alter the combinations of matching variables in the matching process, and check whether coarsened exact matching yields different results. The results for negotiated agreements remain largely unchanged, while the estimate for the removed executive SOE variable is statistically significant in all of the tests. Moreover, it held its statistical significance beyond the 95% level in two of the four matching iterations when controlling for negotiated agreement, and maintained statistical significance at the 90% level for the two additional alternative the matching procedures. These results suggest that while the level of statistical significance of removed SOE is somewhat sensitive to the specific variables included in the matching procedure, it holds for most combinations. The results are presented in Online appendices G and H. Finally, I run a sensitivity analysis based on Cinelli and Hazlett (2020) to test for any degree of omitted variable bias that might affect the results. For negotiated agreement I use opposition size as a reference variable, while for removed ability to declare a state of emergency I use democratic stock. Additionally, I explore sensitivity to omitted variables of the same strength as prior government EV. The results are presented in Online Appendix I. The sensitivity analysis tests how the estimate of the main treatment variable would change as a result of a hypothetical confounder up to three times as strong as the reference confounder.

Figure A7 in the Online Appendix shows that when including a confounder with three times the strength of opposition size, the estimate of negotiated agreement would shift downward, yet maintain a negative association with government electoral violence. The figure also shows that the t-value would maintain statistical significance for an omitted variable of the same magnitude as opposition size yet not for a variable of twice its impact. The additional sensitivity analysis using the prior levels of government-perpetrated electoral violence as a reference point yields more robust results, with the estimate maintaining both strength and statistical significance. ¹⁸ Figure A9 in the Online Appendix shows that the estimate of removed executive SOE would maintain the strength of the correlation with a decline in government electoral violence when including a confounder three times the strength of democratic stock, yet the statistical significance would drop below conventional levels for omitted variables the size of both reference variables. As such, even though the sign of the estimate remains consistent, the level of certainty is lower.

Conclusion

The characteristics of constitutional replacement vary significantly. This, I argue, shapes how constitutional reform impacts the resulting political environment of a country. Focusing on both the process and the content of the reform, I build on insights from previous literature on constitution-making and electoral violence to investigate the conditions under which constitutional replacement can contribute to a decline in electoral violence in hybrid regimes.

In this article, I propose that constitutional replacement can establish conditions for a reduction in electoral violence when they are a result of negotiated agreement between representatives of different groups, and increase constraints on the executive. I find that constitutional reforms resulting from negotiated agreements between distinct elite groups are associated with a decrease in government-perpetrated electoral violence, but do not have an effect on electoral violence committed by other actors. Further, the analysis shows that only the restriction of the executive’s ability to declare a state of emergency is associated with a decline in governments’ electoral violence following the enactment of a new constitution. Other constraints on the executive do not appear to have an effect in the immediate aftermath of a reform. Introducing constraints on executive power is not associated with a change in levels of violence by actors other than those with ties to the government. The results underscore that the characteristics of the process itself have an impact on the aftermath of constitutional replacement.

The effect of negotiated agreement speaks to research showing that the level of inclusiveness during the negotiation process and the practice of elite bargaining have a positive impact on the development of liberal democracy following constitution-making (see e.g. Eisenstadt and Maboudi, 2019; Negretto and Sánchez-Talanquer, 2021a). It also speaks to arguments that constitutional designs fostering elite cooperation promote peace (Lijphart, 1977), though a further investigation into the designs stemming from negotiated agreements is required. It is an important finding, given how commonly government actors and affiliates are implicated in electoral violence. However, the limited impact of negotiated agreement on electoral violence by others than the government suggests that inclusive constitution-making processes may have limited potential to spur pro-democratic changes in political behavior writ large.

The findings corroborate that institutions play a significant role in government-perpetrated electoral violence (see e.g. Hafner-Burton et al., 2014), yet they provide additional nuance. Importantly, the results demonstrate that not all institutional constraints are equally effective in limiting the incumbent’s ability to exert excessive power. Prohibiting the executive from unilaterally declaring a state of emergency appears significantly more effective than other national-level accountability measures, and it is the only constraint that, when introduced, has an estimated effect on the government’s tendency towards violence. This is relevant to discussions on the impact of institutional reforms that constrain executive powers (e.g. Chemin, 2021; Maboudi et al., 2023), and highlights that accountability measures designed for democracies do not necessarily function similarly in semi-authoritarian regimes (cf. Gandhi and Lust-Okar, 2009; Yadav, 2023). The scope of the findings is limited to such hybrid regimes.

The findings also tie into the debates surrounding the effects of institutional reform in hybrid regimes, highlighting the varying impacts on governments and other actors involved. Sceptics have raised concerns over the true impact that institutional reforms can bring in hybrid regimes (see e.g. Glinz, 2011; Hassan, 2015). Often, informal rules and institutions shape the political landscape and interactions to a greater extent than formal institutions do (Schedler, 2013), not least at the local level. While reforms of formal institutions can constrain leaders, they appear less efficient in constraining actors outside the state apparatus. Other actors are varied, and while the same mechanisms may apply to opposition parties that were involved in the agreement and seek national-level political office, they may not apply to actors who opposed the new constitution, were excluded from the process, or use violence due to localized dynamics. Electoral violence by locally active criminal or militant groups is, for example, likely less affected by the institutional landscape. Moreover, as state officials, governments are under greater pressure to uphold the promises made during a negotiated constitutional reform given that they face greater public scrutiny and risk losing legitimacy in the eyes of the populace as well as the international community should they breach the new rules or atmosphere of dispute resolution. As such, the government’s need for legitimacy, and pressure stemming from international and domestic public scrutiny may explain why these types of processes yield a decline in ensuing governmental electoral violence specifically.

The mixed findings invite further research to better understand why negotiated agreements primarily affect the government’s use of violence, and whether the null-findings for other actors are due to the breadth of actors captured in the measurement of the variable, or the fact that they are not in government. Importantly, the data used here rely on expert coders’ perceptions of levels of yearly violence in a country. These are aggregate measures that conceal the diversity of actors active in a country and their distinct actions. Further research could disaggregate further than the data used in this study allow by using available event-based datasets in combination with qualitative analysis, tracing the dynamics of violence by different sets of actors before and after a constitutional replacement.

The findings also raise questions for further research to explore regarding the time frame under which constitutional reforms might mitigate electoral violence. Literature on civil war recurrence has suggested that negotiated peace agreements are more prone to conflict resurgence than military victory or partition (see e.g. Downes, 2004; Toft, 2010). While the dynamics of electoral violence differ from civil war, the findings from this study highlight the peace-promoting potential of negotiated settlements of electoral political conflicts in the short term specifically. The findings by Downes (2004) and Toft (2010), and the empirical persistence of electoral violence in many settings, raise the question of whether electoral violence risks increasing again further down the line. Importantly, the content of a new constitution may take time to implement and initial incentives for peaceful interactions following a negotiated agreement may decline, imploring further research into the development of electoral violence over time in new institutional landscapes.

These insights are relevant for policymakers, governments, and legal advisors involved in constitution-making and the prevention of electoral violence. They can help identify which features are most likely to yield a more peaceful and democratic outcome. They also indicate that there could be challenges in achieving the desired impact through the introduction of new national-level accountability mechanisms, highlighting the importance of taking an approach beyond de jure measures and taking localized dynamics into account.