Motivation and Contribution

This study evaluates peacekeepers' effectiveness in relation to one critical, but understudied security outcome in transitions from war to more peaceful politics: electoral violence. In doing so, it makes important contributions to three strands of literature.

First, our study informs research on international efforts to promote democracy through elections in conflict-affected countries. Researchers debate the role that these efforts play in facilitating war-to-democracy transitions. On the one hand, international election support is considered an integral part of facilitating the transition from bullets to ballots, allowing former warring actors to adopt non-violent modes of competing for power and inviting external enforcement mechanisms (Lyons Reference Lyons2004; Matanock Reference Matanock2017). On the other hand, international efforts to push for early elections in a volatile post-conflict phase have been criticized for running the risk of reinforcing existing divisions and precipitating violence, which may undermine the prospects for peaceful democratic rule (Brancati and Snyder Reference Brancati and Snyder2013; Flores and Nooruddin Reference Flores and Nooruddin2012; Paris Reference Paris2004). Contemporary peacekeeping operations are typically given strong mandates to counteract these dangers and safeguard the transition to a popularly elected regime. Yet evidence of their effectiveness in promoting democracy is mixed. Some studies find that the presence of UN peacekeepers is associated with improvements in overall democracy scores (Doyle and Sambanis Reference Doyle and Sambanis2000; Pickering and Peceny Reference Pickering and Peceny2006; Steinert and Grimm Reference Steinert and Grimm2015), but others find no such effect (Fortna Reference Fortna, Jarstad and Sisk2008; Fortna and Huang Reference Fortna and Huang2012; Gurses and Mason Reference Gurses and Mason2008). The evidence regarding whether a peacekeeping presence during post-conflict elections prevents the recurrence of civil war is similarly mixed (Brancati and Snyder Reference Brancati and Snyder2013; Flores and Nooruddin Reference Flores and Nooruddin2012; Joshi, Melander and Quinn Reference Joshi, Melander and Quinn2017). In light of these inconclusive findings, one way to advance our understanding of the UN's potential role in safeguarding political transitions in conflict-affected environments is to move away from aggregate relationships. By honing in on a particular aspect of the war-to-democracy transition – electoral security – and analysing a specific aspect of peacekeeping – sub-national deployment patterns – we offer empirical evidence that informs our understanding of the broader relationship between peacekeeping and post-conflict democratization.

Our study also contributes to the debate about how international actors can spearhead electoral integrity in fragile political contexts. Electoral violence represents one of the most blatant forms of electoral malpractice, and it severely undermines the credibility of elections (Norris Reference Norris2014). Extant studies of how the international community can promote high-quality elections have so far focused on two instruments: the role of international election observers and UN technical electoral assistance. The net impact of observers, in particular, is debated: they have been found to enhance electoral credibility, for example, by reducing manipulation (Hyde Reference Hyde, Alvarez, Hall and Hyde2008; Kelley Reference Kelley2012), but also increase the risk of post-election contention after fraudulent elections (Daxecker Reference Daxecker2012; Von Borzyskowski Reference von Borzyskowski2019a; von Borzyskowski Reference Von Borzyskowski2019b), to displace violent intimidation to less observed pre-election periods (Daxecker Reference Daxecker2014), or to unobserved districts within countries (Ichino and Schündeln Reference Ichino and Schündeln2012). Although they constitute a less prevalent tool for enhancing electoral integrity, UN peacekeepers deserve particular attention due to their critical role in overseeing political transitions in some of the world's most fragile countries, transitioning from civil war. Unlike election observers and electoral assistance, UN peacekeepers also target election-related security concerns head-on through their military presence.

Finally, our study speaks to the literature on peacekeepers' impact on security-related outcomes and, specifically, their role in securing the post-war period. A number of recent studies suggest that UN peacekeepers can be effective in reducing and containing battlefield violence (for example, Beardsley Reference Beardsley2011; Hultman, Kathman and Shannon Reference Hultman, Kathman and Shannon2014) and shielding civilians from belligerents' attacks (Bove and Ruggeri Reference Bove and Ruggeri2014; Fjelde, Hultman and Nilsson Reference Fjelde, Hultman and Nilsson2019; Hultman, Kathman and Shannon Reference Hultman, Kathman and Shannon2013). We should be cautious, however, about extrapolating peacekeepers' effectiveness during civil wars to the enhancement of electoral security in post-war transitions. Recent work has questioned UN peacekeepers’ ability to deal with post-conflict violence that does not mimic the wartime master cleavage, such as crime or local power struggles (for example, Autesserre Reference Autesserre2014; Bara Reference Bara2020; Di Salvatore Reference Di Salvatore2019). As we discuss below, the qualitatively distinct characteristics of electoral violence might present particular challenges for peacekeepers' effectiveness.

Against this backdrop, peacekeepers' impact on electoral violence has been surprisingly understudied. The only cross-national study on this relationship shows that peacekeepers’ impact on the risk of electoral violence is conditional on their activity in assisting with electoral security and organizing elections (Smidt Reference Smidt2020a). Yet, country-level associations make it hard to distinguish the contributions of local peacekeeping deployments from the overall effect of international intervention – including, for example, diplomatic efforts and other forms of election assistance. Moreover, the effectiveness of peacekeeping deployment may also be masked in cross-national studies because sub-national electoral security can vary significantly within countries. Therefore, this study joins a recent sub-national turn in the peacekeeping literature to further our understanding of whether (and how) peacekeepers can enhance electoral security in countries emerging from war (for example, Di Salvatore Reference Di Salvatore2019; Fjelde, Hultman and Nilsson Reference Fjelde, Hultman and Nilsson2019; Ruggeri, Dorussen and Gizelis Reference Ruggeri, Dorussen and Gizelis2017).

Electoral Security in Post-Conflict Countries

Electoral violence is physical force used to influence the electoral process or its outcome (c.f. Birch, Daxecker and Höglund Reference Birch, Daxecker and Höglund2020). The explicit link to the electoral process means that manifestations of electoral violence (who is targeted, why and by whom) likely look very different from how wartime violence manifests itself (for extensive discussions of the concept, see, for example, Bekoe Reference Bekoe and Bekoe2012; Birch, Daxecker and Höglund Reference Birch, Daxecker and Höglund2020; Höglund Reference Höglund2009; Staniland Reference Staniland2014).

With regard to targets, electoral violence can be directed against voters to deter them from casting their vote, to coerce them into voting a particular way, or to entirely reshape the electoral geography through strategies of forceful displacement (Strauss and Taylor Reference Strauss and Taylor2012). Electoral violence can target political candidates and their campaigns to stifle competition, for instance, through violent crackdowns on party rallies or assassination of political rivals. It can also be directed at the electoral infrastructure itself to reduce voting by particular constituencies or to derail the legitimacy of the elections (Staniland Reference Staniland2014). After election day, electoral violence includes contentious behavior, such as violent protest and riots, to shape political developments after contested results. It can also be used by regimes to thwart popular mobilization, as seen after contested elections in Ethiopia in 2005 and Côte d'Ivoire in 2010, or to ‘punish’ rival constituencies after ballots are counted, as in East Timor in 1999.

With regard to perpetrators, electoral violence is sometimes levied by the same armed groups that waged war with the aim of delegitimizing the elections and discrediting the government's ability to uphold law and order (for example, Condra et al. Reference Condra2017). In many other cases, however, electoral violence is orchestrated by state and non-state actors as a strategic complement to regular campaigning in order to further their electoral aims (Dunning Reference Dunning2011; Matanock and Staniland Reference Matanock and Staniland2018). In these cases, the principals behind the violence are often legitimate electoral contenders, such as party officials or candidates for office.Footnote ¹ These principals, in turn, tend to outsource the implementation of coercion to violence specialists, such as special forces, militia groups, criminal gangs or mobs, to claim plausible deniability and evade sanctions (Staniland Reference Staniland2015).

In summary, electoral violence is often distinct from wartime violence. Thus, it may pose a particularly thorny challenge for peacekeepers. Previous research notes that peacekeeping operations’ mandates and limited capacity force peacekeepers to focus their efforts on containing violence by warring actors involved in the previous conflict (Bara Reference Bara2020). Since the actors behind electoral violence are often different from the warring actors, peacekeepers might struggle to maintain electoral security. Furthermore, military peacekeepers generally have more training and experience in detecting and deterring war-related violence against military targets (for example, army bases or buffer zones) compared to electoral violence against ‘softer’ targets (such as polling stations, voters or election rallies). Thus even in cases where the constellation of actors involved in electoral competition reflect the constellation of actors involved in wartime mobilization, peacekeepers may find it difficult to deal with the particular nature of electoral violence.

While electoral violence poses new challenges for peacekeeping, we point to two important drivers of electoral violence that could also make this type of violence particularly amenable to peacekeepers' intervention. First, precisely because electoral violence is generally linked to the pursuit of electoral aims, the principals who command and organize electoral violence may be particularly sensitive to the reputation costs associated with being called out and exposed. Principals who participate in elections and aspire to be politically legitimate actors will use electoral violence selectively. They are more likely to refrain when the risk and costs of exposure are high and a backlash from international actors and domestic electoral constituencies is likely. Thus peacekeepers' capacity to detect and report electoral violence and to amplify the associated reputation costs provides them with a source of leverage that is particularly useful for maintaining electoral security.

A second driver of electoral violence is the presence of agents who are able and willing to carry out intimidation and coercion. Even in contexts where electoral incentives for violence might be present, it is ultimately the availability of these agents that determines where and when electoral violence actually occurs (Colombo, D'Aoust and Sterck Reference Colombo, D'Aoust and Sterck2019; Höglund Reference Höglund2009). These agents will be particularly sensitive to interventions that manipulate implementation costs, including the costs for arms acquisition or the costs associated with forceful reprisals in retaliation for violence. Thus their ability to heighten the implementation costs of violence gives peacekeepers another source of leverage to maintain electoral security.

In the context of post-conflict elections, and given the legacy of the previous civil war, peacekeeping interventions likely have an added value because, at baseline, the reputation and implementation costs associated with electoral violence are usually small (Höglund Reference Höglund2009; Lyons Reference Lyons2004). In war-affected countries, principals face few risks of exposure and backlash for ordering electoral violence because electoral oversight bodies, such as the electoral commission and civil society groups, have a limited capacity to monitor and sanction principals who rely on threats and coercion (Condra et al. Reference Condra2017; Weidmann and Callen Reference Weidmann and Callen2012). Moreover, the legacy of civil war also heightens the availability of agents of violence. Several political parties that stand for election have their roots in armed organizations (Matanock Reference Matanock2017; Zukerman Daly Reference Zukerman Daly2019), while scores of former combatants may readily offer their violence services to new parties and politicians (Christensen and Utas Reference Christensen and Utas2008). Since the enforcement of electoral laws by the police, military and judiciary might be weak and politicized, violence specialists can also expect to go unpunished for acts of violence (Condra et al. Reference Condra2017; Weidmann and Callen Reference Weidmann and Callen2012). Below we explain how peacekeepers might strengthen electoral security by raising the reputation and implementation costs of pursuing violent electoral strategies.

Data and Research Design

We examine our argument using regression analyses across second-tier administrative units and over months during electoral periods in all sub-Saharan African countries that hosted a UN peacekeeping operation at some point between January 1994 and December 2017. Building on prior research, we define electoral periods to last from six months before to six months after election day (Daxecker, Amicarelli and Jung Reference Daxecker, Amicarelli and Jung2019). Our sample comprises sixteen unique elections in eight countries. There are in total 730 unique administrative units. Election dates are taken from the National Elections Across Autocracies and Democracies database version 5 for the period 1946–2015, which we extended to the year 2017 (Hyde and Marinov Reference Hyde and Marinov2012).

For the local presence of peacekeepers, we draw on information from the Geo-PKO dataset (Cil et al. Reference Cil2020). Geo-PKO covers all UN peacekeeping missions in sub-Saharan Africa for the period 1994–2018. Table 1 provides details on the elections held during the lifespans of the peacekeeping operations in our sample. We construct measures for the presence and strength of peacekeeping deployment. For the strength, we count the number of military bases in a given administrative unit and month during the electoral period. To capture presence, we use a dichotomous version of this variable.

Table 1. List of UN peacekeeping missions and elections in our sample

For the dependent variable, we use information from the Deadly Electoral Conflict (DECO) dataset (Fjelde and Höglund Reference Fjelde and Höglund2021). The dataset is compiled based on published, as well as previously unpublished events, in the UCDP geo-referenced event database, covering the years 1989–2017 (Sundberg and Melander Reference Sundberg and Melander2013). DECO defines electoral violence as violence that is ‘substantially linked to an electoral contest’. A central part of this definition is that the violence is directly tied to features of the electoral process such as political parties, voters, candidates, polling or the institutional arrangements surrounding elections (Fjelde and Höglund Reference Fjelde and Höglund2021). The dataset focuses on lethal violence with at least one fatality. While this dataset omits threats and other forms of physical violence, it arguably captures the most severe forms of electoral conflict that military peacekeepers ought to address. In addition, lethal electoral violence suffers from less under-reporting than less severe violence (Sundberg and Melander Reference Sundberg and Melander2013, 12).Footnote ⁷ To examine Hypothesis 2, we split the sample into the pre- and post-election periods and test the effect of UN military personnel in the two samples. To test Hypothesis 3, we construct two additional dependent variables that count violence by anti-government actors (for example, opposition militias and rebel groups) and violence by pro-government actors (for example, soldiers, police forces, and government militias), respectively.

Figure 1 shows the geographic distribution of violent events in the eight sub-Saharan African countries included in our analysis. In total, there are 130 events of lethal electoral violence. When we aggregate events to our units of analysis (administrative units and months), then 63 out of 16,671 observations exhibit one or more events of electoral violence. In Appendix C we show similar results for a less sparse measure of electoral violence and contention (Daxecker, Amicarelli and Jung Reference Daxecker, Amicarelli and Jung2019). In Figure 1, administrative units with at least one UN peacekeeping base are coloured in blue. Comparing the geographic distribution of peacekeeping bases and electoral violence incidents suggests that UN peacekeepers are deployed in places where electoral violence is more likely to occur.Footnote ⁸

Figure 1. Geographical distribution electoral violence and UN peacekeeping bases

Identification Strategies

One challenge associated with estimating the effect of peacekeeping on electoral security is that the UN likely deploys to places that are already at higher risk of violent electoral contention. From previous research, we know that UN military forces set up bases in civil war hotspots (Ruggeri, Dorussen and Gizelis Reference Ruggeri, Dorussen and Gizelis2017) and, as noted above, the occurrence of previous war-related violence likely increases the contemporary availability of means and manpower for organizing electoral violence. Moreover, peacekeepers continuously gather strategic intelligence on the spatial distribution of threats and adapt deployment patterns to fulfil their mission (Fjelde, Hultman and Nilsson Reference Fjelde, Hultman and Nilsson2019; Ruggeri, Dorussen and Gizelis Reference Ruggeri, Dorussen and Gizelis2017; Townsen and Reeder Reference Townsen and Reeder2014). As such, violent tensions in the early pre-election period are not only a precursor of worse electoral violence; they also likely shape a mission's decision to relocate UN military forces during an ongoing electoral period. If there are generally more peacekeepers in times and places at greater risk of electoral violence, we likely underestimate their violence-mitigating impact.

To address this challenge, our models control for a variety of potentially confounding factors that may influence both peacekeeping deployments and electoral violence. First, we control for civil war-related violence two years previously – that is, we count all incidents of state-based violence and one-sided violence in the thirteenth to the twenty-fourth month prior to the specific observation.Footnote ⁹ In addition, we control for trends in state-based violence and one-sided violence, respectively. To do so, we subtract the violent events count two years previously from the count one year previously (the first to the twelfth month prior to the specific observation). A negative value indicates that violence was already decreasing before the electoral period. All four variables come from the UCDP–GED dataset (Sundberg and Melander Reference Sundberg and Melander2013).

Secondly, we control for characteristics of the peacekeepers' operating environment. The mean distances to the capital and the next international border, as well as the lengths of paved roads, capture logistical challenges associated with acquiring military and electoral supplies. Infrastructural challenges could also influence electoral violence, for example by making it difficult for election workers to reach the area. Measures for distance to the capital and international borders come from the PRIO-Grid (Tollefsen, Strand and Buhaug Reference Tollefsen, Strand and Buhaug2012), while roads data are available from the NASA Socioeconomic Data and Applications Center (CIESIN 2019). Thirdly, we control for the geographic size of the administrative unit and population size because larger and more populous departments may experience more contentious events and host more UN peacekeepers.Footnote ¹⁰ Fourthly, we include the infant mortality rate to proxy for the level of poverty, since poor localities may experience more electoral violence as people have lower opportunity costs of participating in violence. Data for these variables are retrieved from PRIO-Grid (Tollefsen, Strand and Buhaug Reference Tollefsen, Strand and Buhaug2012). Our robustness test models also include a proxy for political grievances by using an indicator for whether the department is inhabited by a politically relevant ethnic group that is excluded from government power. Our results are substantively the same (see Appendix B).

We use three additional strategies to mitigate biases arising from the non-random deployment of peacekeepers locally: fixed-effects models, controls for time trends in our dependent variable and matching. First, our fixed-effects models account for unobserved heterogeneity in the risk of electoral violence across administrative units and electoral periods. In essence, we thus estimate whether changes in peacekeeping deployments influence variation in electoral violence within a locality. In another specification, we also add monthly fixed effects to control for unobserved shocks during an electoral period, such as a shifting power balance after the announcement of results. Though our dependent variable is binary, we use linear regression with robust standard errors to estimate these models because fixed effects in binary choice models may introduce the well-known incidental parameter problem. Linear regression yields unbiased estimates of the marginal effects, and Monte-Carlo experiments show that these estimates differ little from those of binary choice models (Angrist and Pischke Reference Angrist and Pischke2008, 94ff).Footnote ¹¹

Secondly, we also account for time trends in electoral violence to ensure that a negative correlation between peacekeeping presence and electoral violence is not due to the possibility that peacekeepers deploy in locations that already trend towards peace and stability. Thus, we include control variables capturing time trends in our dependent variable over the past three months. For the first trend variable, we subtract the average level of electoral violence 2 and 3 months previously from the level of electoral violence one month previously. For the second trend variable, we subtract the level of electoral violence 3 months previously from the average level of electoral violence 1 and 2 months previously. A positive value indicates an upward electoral violence trend, while a negative value stands for a downward trend.Footnote ¹²

As a third strategy to prevent spurious results, we use matching to create a quasi-experimental sample, in which observations with and without local peacekeeping presence are similar in terms of influential covariates. The post-matching analyses allow us to estimate the effect of peacekeeping not only within but also across spatial units. Therefore it neatly complements the longitudinal, within-units comparisons of the fixed-effects models. Matching can only improve balance on observed factors. Yet, given the absence of a suitable instrument and our knowledge of the determinants of local peacekeeping deployment, it appears to be an appropriate technique (Ruggeri, Dorussen and Gizelis Reference Ruggeri, Dorussen and Gizelis2016). We show that our results are robust after pre-processing the sample with two matching methods: propensity score (PS) and coarsened exact matching (CEM).

In the matching algorithm we use all control variables that we have identified as potentially confounding factors. The sample after PS matching is reduced by 12,159 observations (about 73 per cent) and the post-CEM sample is reduced by 14,746 observations (just under 90 per cent). The algorithm prunes observations with and without a peacekeeping base that are too drastically different from each other in terms of the values of the matching variables. Figure 2 shows that the mean differences in influential covariates between observations with and without local peacekeeping presence are substantively reduced after matching. Yet imbalances remain, and therefore all matching variables enter in the post-matching regression analyses. Given the binary nature of the dependent variable and the exclusion of fixed effects, the post-matching regressions use a logit link function.

Figure 2. Mean covariate differences before and after matching

Analysis

The analyses support the expectation that UN peacekeeping deployments are associated with a lower incident risk of electoral violence in the areas where they are based. In Table 2, we report the results of the fixed-effects linear regression model. Model 1 is our baseline, and Model 2 adds the control variables. Model 3 further adds monthly fixed effects. Across all model specifications, the effect coefficient on UN military bases is consistently negative. In substantive terms, an additional UN base in an average sub-national unit decreases the incident risk of electoral violence by 0.5 percentage points. The reduction is substantively important. It corresponds to about one-tenth of the sample standard deviation in the risk of electoral violence.

Table 2. Fixed effects linear regression of electoral violence

Note: robust standard errors in parentheses. ***p < 0.001, ** p < 0.01, *p < 0.05, +p < 0.1

Models 4 and 5 include the time trends for violence. The effect coefficient on UN military bases remains negative and becomes even larger. The positive coefficients on the trend variables indicate that a prior upward trend in violent electoral conflict relates to a greater risk of electoral violence in future periods. The only control variable that is consistently significant across the models is the trend in state-based violence. The positive coefficient suggests that electoral violence is more likely if battles between government and non-state armed groups have been increasing over the past 2 years.

Next, we present the results from the post-matching logistic regression analyses in Table 3. The models control for all matching covariates as well as the size of the administrative area and the one-month temporal lag of UN military presence. In the matching algorithm, our ‘treatment’ variable is a binary indicator for at least one UN military base in a given administrative unit and month. Across all model specifications, UN peacekeeping presence is associated with a lower risk of electoral violence. While the coefficient on peacekeeping presence is highly significant (Models 6 and 8), the count of UN military peacekeeping bases barely reaches conventional levels of significance (Models 7 and 9). This is not surprising, because the matching process does not account for endogenous deployment of more than one military base to an administrative unit. Multiple bases are more likely to be deployed to locations that are prone to electoral violence.

Table 3. Post-matching logistic regression of electoral violence

Note: robust standard errors in parentheses. ***p < 0.001, ** p < 0.01, *p < 0.05, +p < 0.1

We note that the coefficient on concurrent UN peacekeeping bases would lose significance if we excluded the temporal lag of UN peacekeeping presence from our models. However, it is unlikely that our main findings are an artefact of this control variable. Instead, consistent with extant research, our interpretation is that areas with a military presence have a higher baseline risk of electoral violence (due to non-random deployment choices of peacekeeping operations). If we control for this higher baseline risk of electoral violence – either with unit fixed effects or the temporal lag of military deployment – we can better approximate the impact of a concurrent military presence on electoral violence.Footnote ¹³

Matching changes the distribution of the control variables and reduces the sample size. Thus there is no straightforward interpretation of the effect coefficients on these controls. Nevertheless, we note three interesting results. First, the consistently positive coefficients on one-sided violence suggest that this war-related violence heightens the risk of violent electoral conflict. Yet an upward trend in one-sided violence is associated with a lower risk of electoral violence. One explanation might be that targeted violence before the electoral campaign reduces the need for violence during the electoral period (Daxecker Reference Daxecker2014). Secondly, the spatial lag of UN military deployment that accounts for geographical spillover effects is negative across all models and is even significant in Models 6 and 7. Hence, peacekeeping presence also seems to benefit electoral security in neighbouring administrative units. Finally, the coefficient on peacekeeping presence in the previous month is positive and significant, which supports our assumption that peacekeepers tend to be deployed in the most violence-prone locations.

Per Hypothesis 2, we expect that a UN military presence has a stronger effect on pre-election violence than post-election violence because the reputation costs resulting from peacekeepers' exposure of violence work more strongly on the principals before their fate was decided at the ballot box. As Figure 3 suggests, however, there is no significant difference in the effect coefficients on UN military presence in the pre- and post-election periods (for tables, see Appendix G). Moreover, if we replicate the analyses using the Electoral Contention and Violence (ECAV) dataset, the effect coefficient on UN military presence is only negative and significantly different from zero in the models of post-election violence (for tables, see Appendix I).

Figure 3. Electoral violence in the pre- and post-election periods

Hypothesis 3 predicts that UN military has a stronger effect on anti-government electoral violence compared to pro-government electoral violence. Peacekeepers are more constrained in acting against the host government because their deployment and activity requires its consent and cooperation. As Figure 4 illustrates, UN military presence indeed has a slightly larger effect coefficient in the models of anti-government electoral violence compared to the models of pro-government electoral violence (for tables, see Appendix H). However, the difference is not statistically significant at conventional levels. We also re-estimate these models with the ECAV dataset, which includes a wider range of non-lethal violence and contention (for tables, see Appendix J). In line with our expectation, we find that UN military presence is negatively correlated with anti-government violence, but not with pro-government violence.Footnote ¹⁴

Figure 4. Anti-government and pro-government electoral violence

Using these fine-grained classifications of electoral violence by perpetrator or timing also implies that there are fewer cases of deadly electoral violence in each category (cf. summary statistics in Appendix A). While our initial results for different types of electoral violence point to interesting patterns, especially that peacekeepers might be more effective at mitigating violence sponsored by anti-government actors than pro-government actors, we require more data to corroborate their validity.

Robustness

We argued that a military peacekeeping presence reduces electoral violence by increasing the reputation and implementation costs. The reputation cost-related mechanism rests on peacekeepers' ability to expose violence. Yet, even in the absence of peacekeepers, killings will likely be made public by domestic observers and journalists. As robustness tests, we therefore examine whether our estimated effects of peacekeeping presence hold for non-lethal electoral conflict that usually receives less attention, using data from the ECAV dataset (Daxecker, Amicarelli and Jung Reference Daxecker, Amicarelli and Jung2019). ECAV compiles data on electoral contention, defined as ‘public acts of mobilization, contestation, or coercion by state or non-state actors used to affect the electoral process, or arising in the context of electoral competition’ (Daxecker, Amicarelli and Jung Reference Daxecker, Amicarelli and Jung2019, 716). Appendix C suggests that the presence of military peacekeepers also reduces the risk of electoral contention and less severe violence. Yet the results are weaker when we control for the trend in this dependent variable.

Furthermore, while models with cluster-robust standard errors account for temporal dependence between observations in the same administrative unit, observations and errors could be spatially clustered too. In Appendix D, we therefore re-estimate the fixed-effects models of electoral violence using spatial error models. The estimated negative effect of local peacekeeping presence becomes even stronger compared to the main analyses.

Moreover, rather than acquiescing to the constraints imposed by UN military presence, domestic actors may try to circumvent them and attack the peacekeepers. In locations where such attacks occur, subsequent electoral violence should be likely.Footnote ¹⁵ Thus, using data from Bromley (Reference Bromley2018), we control for the count of all fatalities from malicious acts against UN peacekeeping operations that occurred in the previous year. Since information is only available until 2009, our sample is less than half the original size. Nevertheless, Appendix E shows that the coefficient on UN military presence remains negative across all models. Only in specifications that do not control for electoral violence trends is the coefficient not significant at conventional levels.

Scholars and practitioners warn that attacks against peacekeepers may become increasingly common because UN peacekeeping operations are given increasingly robust mandates that even allow the use of offensive force (Bromley Reference Bromley2018). In Appendix F, we thus split the sample into countries hosting peacekeeping operations that have a mandate to use offensive force against specific armed groups (for example, MONUSCO, MINUSCA, MINUSMA and UNAMID) and those that do not host such missions. The coefficient on UN military presence remains negative and significant at conventional levels.