Electoral violence prevention: what works?

ABSTRACT Elections are in theory democratic means of resolving disputes and making collective decisions, yet too often force is employed to distort the electoral process. The post-Cold War increase in the number of electoral authoritarian and hybrid states has brought this problem into relief. In recent years the prevention of electoral violence has played an increasingly large role in the democratic assistance activities undertaken by international agencies, following increased awareness within the international community of the specific security challenges that elections entail. However, there has to date been little systematic evaluation of the success of different electoral violence prevention (EVP) strategies in reforming electoral institutions so as to enable them to maintain the peace during the electoral period. This article assesses the effectiveness of two common types of international EVP activity. Using a new global dataset of EVP strategies between 2003 and 2015, this article finds evidence that capacity-building strategies reduce violence by non-state actors, whereas attitude-transforming strategies are associated with a reduction in violence by state actors and their allies. The findings are relevant both for understanding the dynamics of electoral violence, and also for policymakers and electoral assistance providers in the international community who have responsibility for the design of democratic assistance projects in states at risk of electoral violence.

has been little systematic analysis of whether electoral violence prevention (EVP) interventions are effective, and if so which strategies work in which contexts.
Electoral violenceunderstood as coercive force, directed towards electoral actors and/or objects, that occurs in the context of electoral competitioncan occur before, during or after elections and it can target a variety of actors, including candidates, activists, poll workers, election observers, journalists and voters. Recent analysis of patterns and trends in electoral violence have found that it is a global phenomenon affecting mainly electoral authoritarian or hybrid states, particularly those in Asia, the Middle East and Africa. 3 Interventions designed to prevent and/or mitigate electoral violence include a range of activities targeted at electoral actors: police training, security planning, electoral management body capacity building, peace messaging, codes of conduct, stakeholder fora and grassroots peace advocacy by civil society groups. This article assesses the effectiveness of two of the most common categories of EVP: capacity building through training and education on the one hand, and attitude alteration through "peace messaging", pacting, dialogue and mediation on the other. 4 Some scholars have argued that more "technical" approaches to EVP such as capacity building can engender confidence in electoral authorities and increase the ability of security services to manage potentially violent situations. 5 Others have held that capacity building alone is insufficient to prevent violence, and that "political" approaches that alter attitudes are also needed. 6 The effectiveness of the two strategies in curbing violence is tested here by means of a new dataset of EVP strategies undertaken by the United Nations Development Programme (UNDP) between 2003 and 2015. We find that the reality is more nuanced than either of these positions: while technical training is linked to a reduction in non-state violence, attitude-transforming approaches are most likely to be associated with a reduction in violence by state actors.
The analysis proceeds as follows. Given that EVP is a relatively new object of academic study, the next section devotes attention to defining and classifying EVP interventions. The second section then goes on to consider factors that might affect which EVP strategies are most likely to be successful. The following section details the data collection and analysis employed to assess the efficacy of EVP, while the fourth section presents the results of the empirical analysis. A final section concludes.

Electoral violence prevention: the concept
Electoral violence prevention has been recognized only relatively recently as a distinct subtype of international electoral assistance, by which we mean organized programmes, undertaken or managed by bodies not based in the state in question, that seek to improve electoral integrity. Electoral conflict was for many years understood by assistance providers mainly in terms of "security" and thus treated largely as a policing issue. Yet there has in recent years been increased emphasis on assistance specifically targeted at the use of force during electoral periods. 7 Work undertaken by organizations such as the UNDP, the International Institute for Democracy and Electoral Assistance (International IDEA), the United States Agency for International Development (USAID), the International Foundation for Electoral Systems (IFES) and the United States Institute for Peace (USIP) to specify the risk factors for electoral violence paved the way for the development of a range of EVP techniques that have been fielded either as standalone projects or integrated into existing electoral assistance programming by practitioners working in this field. 8 This change has been partly a matter of practice and partly a matter of the language used to describe EVP activities; in recent years EVP has become a mainstream component of electoral assistance in many contexts, and electoral assistance providers have begun to talk regularly of "electoral risk", recognizing that violence poses a fundamental threat to electoral conduct.
For the purposes of this analysis, EVP will be defined as any type of electoral assistance activity that aims to prevent or mitigate some form of electoral violence. In theory, EVP can be (and is) undertaken by a variety of actors, both domestic and international (including transnational). In this study, we confine ourselves to consideration of EVP activities carried out during the post-Cold War period. We also restrict our analysis to forms of EVP whose documentation is in the public realm or can be freely accessed; we therefore do not include preventive diplomacy, although we recognize that this can be a fruitful strategy under certain circumstances, for example in endemically violent environments where post-electoral agreements can be negotiated. 9 The insights of previous literature allow us to outline a working theoretical understanding of the causes of electoral violence. Violence affects elections in all regions of the world, but it is particularly prevalent in three regions: sub-Saharan Africa, the Middle East and North Africa, and Asia, and there was an increase in the number of violent elections coinciding with the post-Cold War liberalizations that took place in many countries that had not previously held competitive electoral contests. 10 We also know that violence is more likely if major groups experience or fear exclusion from political decision-making due to ethnic discrimination, electoral fraud or other deviations from democratic norms. 11 The perceived risk of exclusion raises the stakes of elections and makes groups more willing to resort to violence. 12 It follows that the risk of violence at elections can be lowered by increasing political inclusion through, among other things, improving electoral conduct and also by altering perceptions of likely exclusion and of the intentions of actors in the electoral process. Building the capacity of electoral institutions and transforming popular attitudes about the electoral context are therefore both potentially viable means of addressing electoral violence.
As noted above, there is within the recent literature on electoral assistance a debate as to the relative merits of, on the one hand, capacity-building forms of assistance such as training and institution-building activities designed to build the capacity of emerging democracies and semi-democratic states to hold credible elections, and, on the other hand, attitude-transformation strategies that aim to shape the preferences of relevant actors. It is worth pausing to examine these arguments in greater detail.
The distinction between capacity building and attitude transformation is evident in a number of focused discussions of EVP. In a 2016 Policy Directive on "Preventing and Mitigating Election-Related Violence", the United Nations Department of Political Affairs noted the importance of adopting an approach that "combines what is sometimes seen as merely 'technical' assistance on the one hand, and political engagement on the other". 13 In its guide to electoral conflict prevention, the UNDP makes a similar distinction when it notes that "UNDP and its partners also understood that translating lessons from past experiences into careful design in both the 'hardware' (technical, resource and logistical support) and 'software' (conflict transformation) approaches is critically important". 14 A 2013 guide to electoral security published by USAID also distinguishes between "technical" and "political" approaches to EVP, together with "peace-building" activities relevant in post-conflict contexts. 15 This distinction between technical and political electoral violence assistance strategies thus provides a useful analytic framework for assessing interventions designed to reduce electoral violence.
The principal goal of technical assistance is typically to increase the capacity of electoral actorsincluding electoral management body (EMB) and security agencies, but also civil society organizations, voters and other groupsto enable them to deliver a credible election whose outcome will be recognized as fair by winners and losers alike. This is partly due to the constraints and mandates under which international electoral assistance providers operate, but there are also sound theoretical reasons for this approach. The logic behind capacity building is that in under-developed and underdemocratized settings, lapses in electoral procedures are often the result of poor logistical planning and lack of skills, but those who believe that they have been disadvantaged by such lapses almost invariably attribute to them a political motive. Grievances of this type can spill over into violence, especially in a close race. 16 Bolstering confidence in electoral institutions is thus a powerful means of ensuring that electoral procedures remain the sole mechanism for resolving political differences. Typical of this view is the claim, made by Darnolf and Cyllah in a handbook published by the International Foundation for Electoral Systems, that "the most valuable asset any EMB can possess is credibility, which can significantly increase election security. The credibility of the electoral process often leads to increased legitimacy of the government and, thereby, reduced risk of political turmoil". 17 Increasing the technical skills and capacity of electoral actors to run credible elections and to cope with potentially conflictual situations ought therefore to reduce both election-related tensions and the likelihood that such tensions will result in violent outbreaks. Engaging with civil society and voter groups to inform them of the political options, of voting procedures and of how their vote will contribute to the outcome of the contest also goes some way towards building popular capacity to participate in the electoral process meaningfully, which should in theory increase the legitimacy of that process.
By contrast, the attitude-transformation approach to EVP rests on the premise that capacity building via training and technical assistance is not sufficient to address underlying grievances based on power relations in society. Like capacity building, attitude transformation may be undertaken with a variety of groups, including state and nonstate actors. At the grassroots, level, it includes activities, such as "peace messaging", that are designed to alter perceptions of the utility and viability of violence as an election strategy, as well as youth programmes that provide alternative activities for the youth who are often engaged as agents of violence. Attitude transformation also includes various pacting mechanisms (including roundtables, peace pledges and codes of conduct), dialogue fora and mediation that aim to build trust and provide dispute-resolution tools to elites. 18 In the words of Greg Kehailia: for many party supporters, joining a political party that uses violence as a part of its political strategy, is an informed choice driven by shared ideology or socioeconomic motivation. Being able to address reasons why supporters join such a party would take much more than disseminating information about the importance of voting through a voter education campaign. It would require understanding and addressing the contextual or structural drivers that lead to electoral violence. 19 Where actors instrumentally resort to violence at election time due to underlying grievances and/or beliefs in the efficacy of violence as a tool, training and capacity building is of limited help. Under these circumstances, mediation, dialogue and "social enforcement" strategies designed to alter attitudes towards violence are likely to be more effective in deterring actors from selecting violence as their preferred goal for addressing grievances. 20 In sum, technical assistance has in recent years come in for criticism for not getting at the root causes of the reasons why elections are less than credible. 21 The adherents of this view typically argue that it is necessary to undertake forms of assistance that involve "transformative" engagement with power relations, including political mediation, shaping strategies for addressing grievances, and attempting to engender greater genuine commitment to fair electoral competition. 22 Capacity building and attitude transformation are by no means mutually exclusive and in practice they are often used in combination or with other approaches. 23 Yet the broad distinction between these two approaches based on their underlying logics is useful as a means of framing the assessment of EVP.

The efficacy of electoral violence prevention
The core aim of this analysis is to determine the relative efficacy of EVP strategies based on the logic of capacity building and on the logic of attitude transformation. Previous research on this topic is limited, as the academic study of EVP has lagged behind work by practitioners. Yet there is a small and growing body of scholarly work examining technical assistance to EMBs as a means of EVP, 24 as well as political approaches such as mediation, pacting and peace messaging in local communities. 25 Several analyses have noted the success of capacity-building projects. Inken von Borzyskowski's study of the impact of UN technical assistance to EMBs on post-electoral violence in Latin America and Africa between 1990 and 2008 shows that EVP activities carried out by the UNDP can reduce electoral violence in the post-electoral period. 26 Von Borzyskowski argues that technical assistance improves electoral quality, and thereby reduces opposition concerns about fraud. In addition, state actors, by agreeing to take part in international electoral assistance projects, are sending a signal to nonstate actors that they are credibly committed to holding free and fair elections; this in turn increases trust and depresses opposition propensity to stage violent protests after the results have been announced.
A recent collection of qualitative studies edited by Jonas Claes provides a systematic assessment of the success of different types of EVP strategy in preventing electoral violence. 27 On the basis of the cases assessed in this volume, the most fruitful types of EVP appear to be based on the capacity building of state actors such as security sectors and electoral management bodies. Other promising approaches include activities whose target reaches beyond the state, such as long-term voter education activities. By contrast, many of the initiatives undertaken by civil society groups to change attitudes towards electoral violence proved less successful. In their conclusion to the volume, Claes and Macdonald note: "Despite the theoretically compelling logic, the measurable impact of citizen-or community-oriented instruments, like peace messaging, voter consultations, and even youth programming, remains small or unclear." 28 Other studies have found that interventions based on attitude-transformation are effective at reducing violence. In a Nigerian experiment carried out in 2007, Collier and Vicente found that community campaigns by non-governmental organizations (NGOs)including town meetings, street theatre, targeted printed materialsreduced voter tolerance of violence and increased voters' ability to resist intimidation. 29 The mechanisms believed to subtend this finding include the ability of the activities in question to induce (a) coordination in overcoming the collective action problem associated with resisting intimidation; and (b) perceptual change designed to reduce the legitimacy of intimidation. It should be noted that Nigeria is a state in which electoral violence is endemic, thus not one where we would expect EVP activities to have a significant impact, making this finding all the more noteworthy.
The first question we are interested in asking in our analysis is which type of electoral violence EVP strategies are best at addressing, that committed by state agents or that undertaken by non-state actors. As noted above, the existing systematic studies of EVP have largely focused on violence committed by oppositions, and it could well be that this is the context in which EVP activities are most successful. International EVP programmes always require at least some level of official buy-in for them to operate effectively, and under most circumstances, state actors are going to be more likely to support programmes designed to prevent opposition violence than they are to support activities that seek to prevent violence from within their own ranks. Yet we also know that the majority of violence that takes place during the electoral period is committed by state actors. 30 Moreover, many technical assistance programmes, especially those involving the security sector, are designed to enable state agents to diffuse tense situations without resorting to violence. It is certainly plausible, therefore, that EVP activities should reduce state-sponsored violence.
The second question we ask is which types of EVP strategies work best? This is the question to which electoral assistance practitioners most urgently require answers. In his introduction to his edited volume, Claes notes that "Too often the choice among preventive measures is made intuitively or impulsively, rather than based on empirical evidence and risk assessments." 31 We might hypothesize that EVP activities which involve raising awareness of electoral violence, increasing knowledge and providing information will be more effective than approaches whose aim is more transformational. As noted above, Claes' own study found that EVP programmes that entailed working with state actors, such as the security sector and electoral management bodies, were relatively successful, as were long-term voter education and monitoring and mapping, which are often undertaken by civil society organizations.
EVP measures that involve enhancing knowledge and information, such as security sector and EMB training as well as voter education, are capable of altering perceptions and expectations by increasing capacity and reducing uncertainty. Providing police with effective crowd control skills and increasing the competence of the electoral management bodies builds popular confidence in the ability and willingness of the state to hold credible elections. 32 If state bodies prove inept or partisan, by contrast, this can provoke outbreaks of violence, as happened in Kenya in 2007/8, 33 Malawi in 2014 and Uganda in 2011. 34 Not only does training make such state bodies more adept at EVP, it also provides them with the skills needed effectively to communicate their competence and their commitment to maintaining the electoral peace. This helps to overcome the coordination problems that often trigger sporadic violence in contexts of high uncertainty. Another common scenario is one in which fraud is higher than what the opposition had expected and was prepared to tolerate. Emily Beaulieu sees post-electoral protestwhich typically has a violent element to itas a failure of the leadership and the opposition to coordinate on a mutually acceptable level of electoral fraud. 35 Improved ability by all actors to communicate capacity and intentions is likely to boost coordination and reduce the likelihood of violence.
Likewise, long-term voter education increases citizens' understanding of the electoral process, and in so doing it can enhance confidence in electoral conduct and reduce uncertainty as to what voters can expect at different points in the electoral cycle. 36 Knowledge of registration and polling station procedures, vote-to-seat conversion formulae and the timetable for announcing results can be expected to allay fears and address misunderstandings.
By contrast, transformational approaches, such as discussion and mediation (dialogue fora, pacting), are designed to alter actors' core motives in the electoral sphere. There are good reasons to believe that perceptions and short-term expectations are far easier to alter than underlying motives, which may well account for why the Claes study found that deliberative activities involving civil society groups such as voter consultations, peace messaging and youth programming were relatively unsuccessful at preventing electoral violence. 37 The problem with such approaches does not appear to be with the implementer per se, as the authors note that "NGO-led prevention, in total, works, but some elements are more important than others. It appears better to have stronger civic education and monitoring efforts than, for example, peace messaging or youth programming". 38 We can potentially explain this divergence by noting that many of the motives of the elites who orchestrate violence as part of the electoral process are associated with underlying power relations and the high stakes that electoral outcomes entail. These factors are associated with deep-seated interests that cannot easily be swayed by messaging campaigns and similar activities; though of course this does not mean that such interventions cannot be successful in certain contexts and with certain groups.
Yet there are contexts where mediation and the initiation of dialogue between groups have been found to be effective at diffusing tensions and engendering pre-commitment on the part of actors to refrain from violence. In an analysis of southern African elections, Opitz, Fjelde and Höglund found that when EMBs adopted inclusive practices that involved working together with other electoral actors to discuss problems and build trust, less violence resulted. 39 Thus, under certain circumstances, the orientations of competing parties towards violence can be altered, and peaceful outcomes of disputes can be made more likely. Fischer's analysis of UNDP-founded National Peace Council and Regional and District Peace Advisory Councils in Ghana and the Inter-Party Consultative Committee in Sierra Leone also identified beneficial effects of attitude-transforming interventions that included bringing electoral actors together in stakeholder dialogue fora. 40 The studies detailed in this section represent important contributions to our understanding of EVP; the large-N analysis undertaken in this article is useful in ascertaining the extent to which the findings of the case studies hold in a larger global dataset. The next two sections are devoted to testing these competing perspectives. We now detail the data collection and estimation strategies employed to carry this out.

Electoral violence prevention: data and coding procedures
Most studies of EVP are based on the qualitative analysis of cases, individually or in groups. Useful though such analyses are, they struggle to demonstrate causal relationships, and the cases chosen may not be representative. Some of the most interesting recent political science work on the topic of electoral assistance has taken the form of experimental studies that are adept at overcoming the challenges associated with demonstrating causality when using observational data, especially in an area as vulnerable to selection effects as electoral assistance. 41 Experimental work in the field of EVP includes that of Collier and Vicente, Guttiérez-Romero and LeBas, Rosenzweig, and Gonzales Ocantos et al. 42 However, experiments do not enable the formulation of general comparative conclusions about how well given techniques operate in general across a range of contexts, or the relative efficacy of different interventions in given contexts. In order to make general claims like this, we must rely on comparative data that, due to the limits of current data collection techniques, are perforce observational rather than experimental.
While it is undoubtedly the case that state actors carry out many (if not most) initiatives designed to ensure that elections remain conflict-free, 43 there are no specific institutional structures or classifiable practices that can readily be identified a priori as being associated with such activities, and it would not be practicable to assess the efforts of all states to prevent violence from erupting at election time. Instead, data are collected on international electoral violence assistance programmes. Such activities typically involve close collaboration with state actors to guarantee electoral security, but viewing state efforts through the lens of international assistance makes the task of conceptualizing EVP more straightforward, and it makes data collection tractable.
To measure EVP, we focus on electoral assistance activities undertaken by the UNDP between 2003 and 2015. The UNDP is the largest multinational provider of electoral assistance; its work is not systematically linked to the foreign policy priorities of any given state or group of states, and its programme documentation during the period in question is publicly accessible. Though it would in theory have been desirable to collect data on the activities of a wider range of organizations, this was precluded by lack of systematic reporting and/or unwillingness of other global providers to release full documentation they hold on their activities.
Our unit of analysis is the election and our dataset includes all national-level lowerchamber or executive elections held in independent states with populations over one million between 2003 and 2015. Given that EVP is often integrated into broader electoral assistance programming, the corpus we use to code EVP includes documentation on all electoral assistance projects undertaken between 2003 and 2015. We then extract from these documents information on the EVP activities in which we are interested. We coded all electoral assistance programme documentation for the period in question for the implementation of, one the one hand, electoral conduct training programmes (of state and non-state actors including EMBs, security sector, parties and civil society groups), and peace messaging, mediation and pacting interventions aimed at the same groups on the other. An extract from the codebook for the resulting dataset is included in the online appendix together with details of the coding procedure. Each document was coded by two coders with reading fluency in the language of the report (the reports were in English, French, Spanish and Russian).
A total of 99 elections in this dataset benefitted from UNDP electoral assistance, representing 14.7% of the total number of elections held during this period. These elections include 36 in sub-Saharan Africa, 21 in Asia, 17 in Latin America, 7 in Eastern Europe and Central Asia and 3 in the Middle East and North Africa. Of the elections benefitting from UNDP assistance during this period, 46 received programming involving capacity building via training and 26 received programming involving attitudetransformation strategies. 44 The database of UNDP EVP activities is used in conjunction with a range of existing data, including indicators from the Varieties of Democracy (V-DEM) dataset, 45 from the Countries at Risk of Electoral Violence (CREV), 46 and the National Elections across Democracies and Autocracies (NELDA) dataset. 47 Elections are coded according to their NELDA election ID, or for elections not in the NELDA dataset, an ID formed on the same basis. We have collapsed data on concurrent electionsfor example, legislative and executive races occurring at the same timeto avoid double counting of EVP programmes in the data.
Our dependent variable is coded on the basis of data from the V-DEM and CREV datasets. 48 We used two separate measures of electoral violence to add credibility to our results; given that the two sets of indicators are based on different data collection procedures, the fact that the two sets of models reported below yield similar results suggests that our findings are robust to variations in measurement strategy. Two indicators in the V-DEM dataset were used. The first is "Election government intimidation", v2elintim, a variable which reflects answers to the question "In this national election, were opposition candidates/parties/campaign workers subjected to repression, intimidation, violence, or harassment by the government, the ruling party, or their agents?" The highest level of state intimidation on the v2elintim scale represents a situation where "repression and intimidation by the government or its agents was so strong that the entire period was quiet". 49 In order to be able to use this indicator as a measure of state violence, the highest category was recoded as "0" (no violence), and the ordinal version of the indicator was inverted. The result is a four-point scale where higher levels of violence are designated by higher values. The second useful V-DEM indicator is "Election other electoral violence", v2elpeace, which represents answers to the question In this national election, was the campaign period, election day, and postelection process free from other types (not by the government, the ruling party, or their agents) of violence related to the conduct of the election and the campaigns (but not conducted by the government and its agents)?, 50 and is thus a good measure of non-state violence. The ordinal version of this variable was also inverted so as to generate a five-point scale where higher levels of violence are reflected in higher values.
The corresponding CREV measures for state-initiated and non-state-initiated electoral violence are based on counts of the number of violent physical events during a ten-month window around each election. 51 These data are coded from event data contained in the Integrated Crisis Early Warning (ICEWS) dataset. 52 These granular data are well-suited to measuring changes in the level of electoral violence since they are continuous counts of the number of reported events, and information is not lost via aggregation or the introduction of thresholds. In the CREV dataset, states at risk of electoral violence include only those whose "Polity" score from the Polity IV dataset has at some point during the period in question fallen below 10, 53 indicating that democracy is not fully consolidated in that state. The CREV data divide electoral violence events into "threats and coercion", where "threats, incitement and coercion falling short of actual bodily harm" and "physical violence" where "violence taking the form of assaults and violent physical confrontations". 54 In this analysis we rely exclusively on variables designating "physical violence", as we are interested in the most serious forms of violence which pose the greatest threat to electoral integrity. Two indicators from the CREV dataset are used as dependent variables in this analysis: state_nonstate_attacks as a measure of state violence and a variable designating non-state violence composed by adding nonstate_state_attacks and nonsta-te_nonstate_attacks. 55 Descriptive statistics for these and the other variables in our analyses can be found in Table A1 of the online appendix.

Results
International organizations tend to undertake electoral assistance projects in contexts where problems with elections, including violence, are anticipated. In order to prevent this selection effect from biasing our findings, we employ a Heckman section model to account for those factors that condition the provision of electoral assistance by the UNDP. The selection model includes as indicators level of democracy and level of electoral authority capacity, factors which have been found in previous research to be strongly associated with the provision of UN electoral assistance. 56 The resulting random effects probit model is shown in Table 1. These findings partially confirm the results of previous research; electoral authority independence is negative and significant at the 0.028 level; democracymeasured by means of the V-DEM polyarchy (v2x_polyarchy) indicatoris not significantly related to the provision of electoral assistance, once electoral authority capacity is controlled. This suggests that states which have weaker electoral authority capacity are more likely to receive UNDP electoral assistance. 57 The next step in the analysis is to model the impact of different types of electoral assistance. The second stage model, which takes the selection effect into consideration, includes dummy variables for the types of EVP under consideration here: capacity building and attitude transformation. We report separate models for state and nonstate violence based on both the V-DEM and CREV indicators.
We also included control variables for a variety of factors found consistently in previous research to be associated with electoral violence: electoral fraud, ethnic exclusion, corruption, type of election (presidential or concurrent, with legislative as the baseline category), and world region. 58 With the exception of the fraud and election type indicators, which relate to the election under consideration, all the control variables are taken from the previous year or from 1 January of the year of the election in question, which means that they are lagged relative to the dependent variable. We also include a lagged dependent variable in each model to provide a further control for possible endogeneity and to absorb country-specific effects. Finally, population (logged) is included in the models based on CREV data, due to the fact that these are incident count models and greater numbers of electoral violence incidents can be expected to take place in larger countries, all else being equal.
The results of the second-stage models are presented in Table 2. These results provide support for the arguments in favour of both capacity-building and attitude- transformation strategies of EVP, but with an important nuance: whereas attitudetransforming interventions are linked to lower rates of state-initiated violence, capacity-building approaches are associated with lower levels of non-state-initiated violence. The result for attitude-transformation interventions such as pacting, mediation and peace messaging holds in both the models based on V-DEM and CREV data, whereas the result for capacity-building reaches statistical significance only in the model based on the more finely-grained CREV data. There is also some evidence that attitude-transformation may actually increase non-state electoral violence; though the coefficient is only significant in the CREV model and only weakly so (at the 0.044 level), there is a hint that international efforts to alter the attitudes of nonstate actors towards election-related violence may backfire in some contexts. It thus appears we have uncovered an interesting inverse relationship. Though some of the capacity building programmes were targeted at non-state actors, and some of the attitude-transformative programmes were targeted at state actors, it seems that generally speaking working with one actor reduces the propensity of the other actor to resort to violence. 59 The role of capacity building in reducing violence initiated by non-state actors can be plausibly explained by the fact that higher capacity EMBs and more credible electoral institutions engender trust on the part of others in society, and lack of confidence in the integrity of elections is one of the most common triggers of violent conflict during the electoral process. 60 This result accords with findings of von Borzyskowski, who identified technical electoral assistance as a key means of increasing confidence in electoral authorities and reducing electoral violence by non-state actors. 61 The finding that attitude-transforming interventions are successful in reducing violence by state actors can potentially be explained with reference to the ability of such interventions publicly and credibly to bind state actors morally to peaceful strategies, thus reducing the perceived threat of elections to oppositions. Negotiation fora may also provide opportunities for behind-the-scenes deals that lead to co-optation of opposition elites or enable such elites to press for key grievances to be addressed. Yet as hypothesized, this effect is less pronounced and less consistent than that for capacity building.
The substantive impact of these effects is noteworthy. Model A implies that an election benefitting from attitude-transformation interventions of some kind is likely to have a score 0.859 points lower on the four-point V-DEM state-violence scale, all else being equal. Model B suggests that such an election will be associated with on average approximately one (0.663) less violent attack during the ten-month electoral period. Model D indicates that an election in which capacity-building training interventions were deployed can be expected to experience one (1.262) less non-state initiated attack event on average during the electoral period.
Outbreaks of physical violence that occur during the electoral process can seriously undermine the legitimacy of that process, and they can have a chilling effect on popular engagement with electoral competition. The fact that EVP interventions can be shown to be associated with a reduction in the number of attacks in these models, despite the small number of cases included, is a striking finding that provides evidence for the effectiveness of EVP activities commonly deployed in electoral assistance programmes. As these cross-national results are among the first to be generated on the impact of EVP interventions, they also warrant further probing and testing in other studies employing different data sources and research designs.

Conclusion
Electoral violence is a significant problem in the contemporary world. When it occurs, violence can undermine participation and confidence in electoral processes, and it can derail emergent democratic institutions. The use of force at election time can also threaten fragile peace-building processes, and it can have negative consequences for the quality of life of victims. It is therefore not surprising that there has been considerable recent effort devoted to addressing this harmful phenomenon, especially by international organizations keen to sustain peace and foster democratic consolidation.
Until now we have had scant systematic evidence of the effects of the EVP activities undertaken by such bodies, and a number of recent studies have cast doubt on the ability of researchers and practitioners alike to establish robust causal efficacy of electoral assistance programmes. 62 Yet this article demonstrates that with careful attention to selection issues and to the use of control variables, comparative analysis of a corpus of programme documents can unearth patterns linking EVP interventions to the actual experience of conflict at polling time. The analysis undertaken in this article suggests that such interventions can be successful at reducing violence; both EVP intervention strategies based on capacity building and those based on attitude transformation are associated with reduced levels of electoral violence, even controlling for selection effects and a host of other factors known to be associated with violent elections.
Our headline finding is that attitude-transforming interventions are associated with a reduction in violence by state actors, whereas capacity building appears to reduce violence by non-state actors. Training and other capacity-building activities play a major role in many electoral assistance programmes; indeed, training is by far the most common intervention in the UNDP documents employed in this analysis. The results of this analysis suggest that this sort of intervention goes a considerable way towards reducing the incentive of non-state actors to challenge the state violently. However, we know that most electoral violence around the world is undertaken by state actors, 63 and when violence by state actors is expected to be particularly problematic, strategies involving mediation, pacting and other attitude-transforming techniques have a larger role to play in fostering a context where actors can be pressured to credibly commit to non-violent forms of electoral competition. An additional contribution of this study is the systematic coding of electoral assistance documentation to generate an original dataset that can potentially be used by other scholars.
Future research could usefully examine the relative effectiveness of a wider range of different types of EVP, as well as the relative success of different programme implementers. It would also be of interest to consider the stage of the electoral cycle in which interventions were most effective at quelling conflict, as well as the electoral actors with whom it was most useful for EVP initiatives to engage (grassroots or highlevel). And it would be of benefit to have greater insight into the causal mechanisms subtending the findings reported here. Future studies could also explore the effects observed here over a longer time period; the stated aim of UNDP programming documents is to influence electoral conduct over the course of the period of the programme and shortly thereafter, but it might also be hypothesized that certain effects of interventions also have long-term impacts years after the programmes in question have finished. A further area of useful research would be the collection and analysis of the activities of a wider range of EVP providers. The study of EVP is still in its infancy, and much work remains to be done to fine-tune our understanding of the conditions under which concerted efforts to prevent and mitigate electoral violence are successful. This article has hopefully helped to lay the groundwork for future work in this emerging field. Notes 1. Global Commission on Elections, Democracy and Security, Deepening Democracy. 2. Dundas and Ojo, "Nigeria's Experience with Electoral Violence." 3. Birch and Muchlinski, "The Dataset of Countries at Risk of Electoral Violence"; Birch and Muchlinski, "Electoral Violence." 4. We recognize that these are not the only strategies that could have been chosen, but they reflect our theoretical focus on capacity building versus attitude transformation, and they are also relatively under-studied in comparison to other approaches such as monitoring which have already been widely assessed in the literature on electoral violence as well as the broader literature on electoral integrity. Beaulieu and Hyde, "In the Shadow of Democracy Promotion"; Von Borzyskowski, "Sore Losers?"; Daxecker, "The Cost of Exposing Cheating"; Kelley, Monitoring Democracy; Hyde, "The Observer Effect in International Politics"; Hyde, The Pseudo-Democrat's Dilemma; Sjoberg, "Autocratic Adaptation." It is not possible to elaborate a full typology of