The effectiveness of mediation and peacekeeping for ending conflict

Mediation and peacekeeping are commonly used tools to manage conflict. To what extent are they complementary and effective instruments for ending violent conflicts? Generally, they are seen as distinct tools: mediation aims to facilitate negotiated settlements, while the goal of peacekeeping is to prevent agreements from collapsing. However, peacekeeping and mediation regularly occur simultaneously. Arguably, peacekeeping operations rely on continuing political processes, while peacekeepers create a context favorable for mediation and provide a valuable source of independent information. Using a variety of model specifications, including selection models, empirical evidence supports that (a) mediation rather than peacekeeping is key to halting hostilities, (b) mediation and peacekeeping are largely complementary, but (c) this complementarity is conditional: in the post-Cold War period, transformative peacekeeping boosted the effectiveness of mediation to halt civil wars. There is no evidence that peacekeeping on its own matters for ending conflict. Finally, counterfactual analysis shows the substantial impact of mediation and peacekeeping on the frequency of conflict.


Introduction
Are mediation and peacekeeping effective conflict management tools? To what extent are they complementary in ending armed conflict in the sense of stopping the fighting? Regardless of widespread skepticism, there is growing evidence that peacekeeping and mediation can be effective. 1 However, there have only been a few attempts to disentangle their relative contribution, or to assess the extent to which they influence each other (Greig & Diehl, 2005;Diehl & Regan, 2015). Therefore, our research addresses the following specific questions: (1) how effective are mediation and peacekeeping at ending conflict, and (2) how do mediation and peacekeeping shape each other's effectiveness?
Third-party efforts to end conflict encompass a broad spectrum of methods, but mediation and peacekeeping, primarily under guidance of the United Nations (UN), stand out in terms of their frequency and influence. In his opening remarks to the High-Level Debate on Peacekeeping in 2018, the UN Secretary-General Guterres commented, 'I call on Member States to sustain your political engagement and push for political solutions and inclusive peace processes, including through bilateral diplomacy and sanctions if necessary. A peacekeeping operation is not an army, or a counter-terrorist force, or a humanitarian agency. It is a tool to create the space for a nationally-owned political solution'. 2 Here, we examine whether there is indeed such complementarity between political and 'militarized' approaches. Figure 1 graphs the frequency of mediation and peacekeeping since 1946 and helps to identify relevant trends. Both instruments were rarely used until the early 1980s, and mainly in interstate rather than intrastate wars. From the 1980s onward, and especially in the post-Cold War period, mediation and peacekeeping in intrastate wars have become more common (Heldt & Wallensteen, 2007;Wallensteen & Svensson, 2014). Mediation is now arguably the most common form of conflict management during violent periods (Bercovitch & Gartner, 2006), though peacekeeping has grown dramatically in size and significance (Heldt & Wallensteen, 2007;Kathman, 2013). When in 2018 the United Nations marked 70 years of peacekeeping, it deployed 110,000 personnel serving in 14 peacekeeping missions. Moreover, over time, peacekeeping has been given broader mandates and become more robust (Doyle & Sambanis, 2006;Fortna, 2008). Such so-called transformative peacekeeping missions have only been deployed to civil wars. 3 Mediation and peacekeeping present distinct approaches to conflict management: as ideal types, mediation strives to end an ongoing conflict by facilitating a negotiated settlement, while peacekeeping aims to guarantee the stability of any agreement reached to reduce the likelihood of renewed fighting. In interstate conflicts, we rarely observe mediation during peace periods, or peacekeeping while fighting is ongoing. UNMOGIP, in the Kashmir region, is a rare example of a peacekeeping mission operating during interstate conflict, since it remained deployed during periods of open hostilities between India and Pakistan. In civil (or intrastate) wars, any clean delineation between mediation and peacekeeping is often inappropriate, since mediated agreements in civil wars regularly fail to halt fighting. Regularly, a ceasefire or peace agreement excludes potential spoilers with fighting continuing in parts of the country (e.g. in the Democratic Republic of the Congo). The situation on the ground may also deteriorate quickly because of failures in implementing the original agreement or unanticipated events (e.g. in South Sudan or the Central African Republic). Finally, humanitarian emergencies may necessitate the deployment of peacekeepers before a ceasefire is agreed (e.g. in the case of Bosnia), and peacekeepers become involved in ongoing conflicts. As Figure 1 shows, peacekeepers regularly operate while hostilities are ongoing and mediation efforts are continuing. This makes it meaningful to inquire into their relative contribution to halting actual fighting.
Despite both methods being regularly used simultaneously, research has rarely evaluated any joint or conditional effects. 4 So far, the key contribution has been provided by Greig & Diehl (2005), who outline contrasting expectations on how peacekeeping could influence the likelihood and success of mediation. Their empirical evidence suggests that peacekeeping mainly undermines mediation attempts. More recently, Beardsley, Cunningham & White (2019) consider the relative impact of diplomacy and peacekeeping on the severity of conflict, and Joshi & Quinn (2015) and DeRouen & Chowdhury (2018) examine the interaction of peacekeeping and mediation in avoiding the recurrence of civil wars. These studies report positive and complementary effects. Kathman & Benson (2019) observe that the deployment of peacekeepers shortens the duration of conflict until a negotiated, that is, generally mediated, settlement. 5   ' and 'civil war' are used interchangeably. 4 A few studies explore the interactions, interdependencies, and complementarities between conflict management methods more generally (see Owsiak, 2014;Diehl & Regan, 2015;Greig, Owsiak & Diehl, 2019). 5 Kathman & Benson (2019) control for intervention on either the rebel or government side which is, however, insignificant in all their models. Interventions can mean offers to mediate, but often include Existing evaluations of the relative contribution of peacekeeping and mediation use a range of measures: (1) whether (Greig & Diehl, 2005) or how quickly (Kathman & Benson, 2019) a negotiated settlement is reached, (2) the reduction of battle-deaths during conflict (Beardsley, Cunningham & White, 2019;Bredikova, 2019), or (3) hostilities following a peace agreement (Joshi & Quinn, 2015;DeRouen & Chowdhury, 2018). We focus on the stopping of fighting, both before and after the signing of any peace agreement. We argue that this provides an appropriate and encompassing yardstick for conflict resolution. Firstly, although negotiated settlements aim to end hostilities, they often fail. The duration of peace following a settlement thus in effect measures the robustness of stopping the fighting. Negotiated settlement and conflict recurrence not only focus on different (but related) aspects of 'stopping the fighting', but also suggest distinct theoretical mechanisms. Settlements are more directly associated with mediation, while peace duration is more associated with peacekeeping. Since we are analyzing the relative contributions of mediation and peacekeeping, it is appropriate to select a dependent variable to which both mediation and peacekeeping may contribute without giving an obvious primacy to one over the other. Secondly, conflicts often witness multiple periods of intensified hostilities but also times of relative peace. In practice and in line with our dependent variable, conflict management means shortening such periods of hostilities as much as possible.
We argue that peacekeeping and mediation are complementary in ending hostilities. We empirically explore the conditions under which such complementarities exist, using conflict data from the Uppsala Armed Conflict Data Project (Pettersson & Wallensteen, 2015). Our analysis further evaluates different peacekeeping mandates, types of conflict, and time periods. We assess the statistical significance of mediation and peacekeeping using a variety of model specifications including selection models. We also explore their substantive impact on ending hostilities. Since both instruments could be statistically significant while actually contributing little to conflict termination, we use simulations to provide a counterfactual analysis showing their substantial impact on the frequency of conflict. 6 Key findings are that mediation rather than peacekeeping is key to halting hostilities. There is also some evidence that mediation and peacekeeping are complementary: in the post-Cold War period, transformative peacekeeping (i.e. large multidimensional forces, mandated under Chapters 6 and/or 7 of the UN Charter) boosted the effectiveness of mediation in halting civil wars. Finally, we show that together peacekeeping and mediation contributed to a substantially lower frequency of armed violence. However, we find no evidence that peacekeeping on its own mattered for ending hostilities. 7 The next section briefly summarizes the main insights of the literature that considers the effectiveness of mediation and peacekeeping separately. We then develop the argument why mediation and peacekeeping are best thought of as complementary instruments. We explicitly distinguish between the impact of mediation and peacekeeping on conflict termination in interstate and civil wars. Subsequent sections set out the empirical strategy and present the key results. We conclude with discussing how peacekeeping relies on political engagement.

Literature review: Mediation or peacekeeping
As the main forms of non-coercive international intervention, mediation and peacekeeping have received extensive scholarly attention. Notably, a large number of quantitative studies find that both policy instruments reduce the severity and duration of conflict, as well as the likelihood of conflict to reignite, even though the shortcomings and inefficiencies of many mediation and peacekeeping attempts have also been documented extensively.
Mediation occurs when disputants accept the assistance of a third party to resolve their differences without resorting to physical force or invoking the authority of law (Bercovitch, Anagnoson & Wille, 1991). The nonbinding nature of mediation appeals to disputants wishing to retain control throughout negotiations, and international actors seeking a relatively cheap means through which to influence a conflict (Clayton, 2013). Beardsley et al. (2006) identify facilitation (i.e. improving the flow of information), formulation (i.e. process more extensive military engagement. Bredikova (2019) also considers the conditional impact of mediation, peacekeeping, and intervention and highlights how intervention and mediation are correlated with higher, and peacekeeping with lower fatalities. design), and manipulation (i.e. offering positive and negative inducement) as key forms of mediation. Via improving the flow of credible information and helping belligerents to reveal their true preferences, mediation improves the quality of dialogue between the parties and assists them in resolving their conflict (Beardsley et al., 2006). 8 Mediation also counters fears of defection and offers positive and negative inducements to push belligerents towards peace (Stedman, 1997). Accordingly, it counterbalances incentives for exploitation that could lead the conflict resolution process to falter. For example, mediators can reward continued cooperation, provide diplomatic support for the implementation of any agreement, hold parties accountable, and provide security guarantees to reduce belligerents' fears. In short, mediation offers a bridge towards self-enforcing arrangements and makes it more likely that open hostilities end (Gurses, Rost & McLeod, 2008;Beardsley, 2011a).
Apart from stopping the fighting, a core criterion to evaluate the effectiveness of conflict management is the durability of the achieved 'peace'. In this respect, the impact of mediation is not necessarily positive. Whereas mediators can help belligerents to identify a mutually satisfying outcome more quickly, over time the incentives provided by mediation are likely to wane, and in combination with changes in the demands of the (former) combatants, mediated agreements are possibly less durable than those achieved without mediation. Accordingly, Beardsley (2008) argues that mediated agreements risk breaking down due to the time inconsistency problem, and reports evidence that, at least in interstate conflicts and crises, mediated agreements are less enduring. Arguably, incomplete information and time inconsistency also affect mediated outcomes in intrastate wars. Mediation, however, does not necessarily halt when a peace agreement has been reached. Figure 1 shows that mediation also takes place post-conflict, and continued engagement by mediators may still limit the extent and duration of renewed hostilities.
The relevance of peacekeeping for conflict resolution is also increasingly recognized. UN peacekeeping relies on authorization by the Security Council and provides a credible signal of a shared concern of the main powers about the need to end or at least contain a conflict. Peacekeeping has evolved to cover a broad range of activities and to include a wide variety of military and civilian actors. Doyle & Sambanis (2006: 11-18) distinguish between 'traditional' and 'transformative' forms of peacekeeping (see also Hegre, Hultman & Nygård, 2019). Traditional peacekeeping missions involve lightly armed troops and operate with the consent of the parties. According to Heldt & Wallensteen (2007: 10), these peacekeeping operations 'share basic tasks that are included in the traditional understanding of the term, namely separation of the conflict parties, ceasefire monitoring, and control of buffer zones'. Traditional peacekeeping also assists with non-invasive tasks such as monitoring troop withdrawal, ensuring that disputants observe the terms of a ceasefire or partial agreements, and preventing escalation of isolated incidents. Traditional peacekeeping thus lessens informational and commitment problems in the implementation of peace agreements (Walter, 2001). Showing an interesting overlap with mediation, traditional peacekeeping also regularly supports negotiations between (former) belligerents; examples include MINURSO in Morocco, and UNFI-CYP in Cyprus. Crucially, traditional peacekeeping is ill equipped to respond to large-scale resumption of hostilities. Perceived ineffectiveness, often coupled with the withdrawal of consent from the host countries, makes it less likely for traditional peacekeeping missions to continue during large-scale hostilities.
Peacekeeping has increasingly gone beyond monitoring and verifying the actions of belligerents. Transformative peacekeeping aims to alter the context of a conflict by becoming an active participant in the peacemaking process. This includes 'second-generation' peacekeeping operations that seek to address the roots of the conflict by promoting and assisting with economic and political reconstruction, such as the UNMIT mission in Timor-Leste. Transformative missions also encompass 'thirdgeneration' peace enforcement missions that do not require the consent of the parties, but instead seek to protect populations and apply force more robustly -for example, UNPROFOR in Bosnia. 9 Transformative peacekeeping missions equipped with appropriate mandates and supported by sufficient troop numbers have proven to reduce the intensity of conflict, to protect civilians (Hultman, Kathman & Shannon, 2013Kathman & Wood, 2016), to contain the 8 More generally on the role of information and commitment in the process of civil war bargaining, see Walter (1997). 9 Doyle & Sambanis's (2006: 11-18) original method of categorization separated peacekeeping operations into four groups: observer, traditional, multidimensional, and enforcement. We follow Hegre, Hultman & Nygard (2019) in aggregating these categories into traditional and transformative operations, as prior research has suggested transformative missions tend to be more effective than the more limited traditional missions. spread of violence (Beardsley, 2011b;Beardsley & Gleditsch, 2015), and to limit the duration of conflict (Ruggeri, Dorussen & Gizelis, 2017;Kathman & Benson, 2019). Transformative peacekeeping can assist local parties to progress from a partial agreement towards a comprehensive settlement. Peacekeepers thus not only try to 'keep' but also 'make' peace (Kathman & Benson, 2019). It is common for transformative peacekeeping to take place during ongoing hostilities, further blurring the line between peacekeeping during and after conflict (DeRouen & Chowdhury, 2018). Peacekeepers, in particular UN police, also assist with local-level policing and help to prevent the escalation of small-scale violent events. Using subnational data on intrastate conflict in sub-Saharan Africa, Ruggeri, Dorussen & Gizelis (2017) provide evidence that the presence of peacekeepers shortens conflict episodes locally.

Mediation and peacekeeping complementarities in ending conflict
So far, most research has evaluated the effects of mediation and peacekeeping separately and has largely ignored that both instruments are regularly applied simultaneously. 10 Consequently, empirical analyses have failed to control for the possible confounding influence of the other form of conflict management. Moreover, the presence or absence of a particular form of conflict management is likely to condition the effect of the other method.
First of all, we expect that mediation by third parties is valuable to ongoing peacekeeping missions. Mediation facilitates the broader political process on which the deployment of peacekeepers normally relies, and which is essential to enable peacekeepers to operate effectively. Armed factions routinely challenge, obstruct, and undermine peacekeeping operations (Duursma, 2019). 11 Retaining the ability of peacekeepers to operate effectively requires active and authoritative representation at the highest levels of power. The UN engages in diplomatic activities by means of its 'good offices' representation of the UN Secretary-General in support of and alongside its peacekeeping missions (Arnault, 2006). Mediators further support this process by means of advocacy. They raise the concerns of peacekeepers with the central government and rebels. Ideally, they communicate strong international support for the peace process and the peacekeeping operation. Moreover, the continued diplomatic engagement with the peace process helps to clarify the interests, positions, and intentions of the parties in conflict, providing valuable information to peacekeeping missions and helping them to fulfill their mandate; for example, Matanock (2017) demonstrates how international diplomacy and peacekeeping both contribute to successful post-conflict elections.
Peacekeepers may also contribute to the effectiveness of mediation attempts. Greig & Diehl (2005) contrast positive and negative spillover effects. Regarding the latter, they argue that peacekeeping limits the flow of credible information by sustaining a military stalemate, making it in effect more difficult for mediators to achieve a negotiated settlement. As long as the stalemate results in low-level hostilities, belligerents are under less pressure to reach an agreement. Beardsley, Cunningham & White (2019) indeed find that diplomacy complements peacekeeping (in particular when a mission deploys low numbers of troops) in reducing battlefield casualties in ongoing conflicts. Yet, containing hostilities could have as an unintended consequence that it takes longer to get the conflict resolved. Greig & Diehl's (2005) empirical evidence largely supports a 'pessimistic logic' with, if anything, a negative impact of peacekeeping on concluding a negotiated settlement.
Even though we acknowledge some of the concerns raised above, we argue that, overall, peacekeeping reinforces mediation in halting armed conflict for a number of reasons. Firstly, it is harder for a mediator to negotiate a solution when the context on the battlefield remains in flux. Peacekeepers help to stabilize a situation, creating a more favorable environment for the mediators. Cooperation -or at least coordination -between peacekeepers and mediators assists the latter in their search for a lasting solution to the conflict. Peacekeepers monitor the activities of (former) combatants and verify that their behavior accords with any preliminary agreement. In this way, they help to build trust that is essential for mediators to continue the peace process. The field reports of peacekeeping operations contextualize their observations and provide valuable information to the mediators, and more directly to the members of the Security Council. Mattes & Savun (2010) argue accordingly that thirdparty involvement improves the durability of intrastate war settlements by means of providing information and reducing uncertainties, but they do not link this explicitly to the continued involvement of mediators in the peace process. DeRouen & Chowdhury (2018) find a positive effect for the peacekeeping-mediation interaction in lowering post-agreement violence.
Contrary to Greig & Diehl (2005), we would argue that the value of battlefield performance as a source of information about relative strength is easily overstated, in particular in the case of irregular warfare. 12 Rather than mediation or peacekeeping, it is the inability of any party to achieve a decisive victory that leads to military stalemate. As a consequence of the military stalemate, parties increasingly engage in a war of attrition where force is mainly applied against civilians either as part of a counterinsurgency strategy, or as an attempt to undermine effective governance. In these cases, however, the information revealed by the use of force is often at best insufficient to resolve the dispute, and misleading at worst. Atrocities by government forces regularly only end up strengthening the hold of rebels over communities. Equally, sabotage can excuse poor governance rather than demonstrating an inherent lack of capacity or willingness to govern.
In comparison, monitoring and verification are core tools of peacekeeping operations, allowing them to create a useful supply of relatively unbiased information. They provide information on violations of agreements and progress on their implementation, and on military maneuvering. Mediators rely on a ready supply of credible, and relatively unbiased, information to be effective. The information provided by peacekeepers is often of direct relevance for effective mediation, in particular as research has shown that mediators are more effective when they can bring their own independent source of information (Kydd, 2006;Fey & Ramsay, 2010;Hörner, Morelli & Squintani, 2015). Kathman & Benson (2019) similarly show that peacekeepers speed up a negotiated settlement of conflict; they argue that peacekeepers accomplish this by increasing the costs of combat, facilitating information sharing among combatants, and providing security guarantees. Peacekeeping operations regularly support the work of mediators directly; for example, they routinely secure the personal safety of mediators working within conflict zones. The ability of peacekeeping to safeguard the capital and key infrastructure also facilitates mediation efforts. As Arnault (2006) notes, the need to provide physical security extends to all parties involved in the mediation process, especially when the successful implementation of the peace process is still very much in doubt. 13 Transformative peace enforcement operations are equally, and possibly better, equipped to provide information to mediators, as they have a larger deployment and broader mandate. Transformative missions not only include more military personnel but also police officers and a significant number of civilians. Collectively, they allow for more detailed gathering and processing of intelligence throughout the area of deployment. This is reflected in the United Nations Joint Mission Analysis Centre (JMAC) field handbook (United Nations, 2018), which suggests that there should be weekly meetings between JMAC and the diplomatic forum which includes members of the mediation team. For example, the JMAC situation reports in Darfur detail regular meetings between Dribrill Bassolé, the Joint Chief Mediator, and UNAMID staff, offering clear evidence of two-way information flows between the mediation and peacekeeping teams. 14 The Security Council has increasingly authorized missions to use force to protect civilians -in effect, constraining the ability of the armed actors to conduct a war of attrition. The ability of peacekeepers to contain conflict also facilitates mediation, because the re-emergence of civil society can provide a boost to mediation, as shown in Liberia. The impact of transformative peacekeeping on local stability is well documented : Hultman, Kathman & Shannon (2013 find that transformative peacekeeping reduces one-sided violence. Beardsley & Gleditsch (2015) observe that peacekeeping missions reduce the area of a country affected by armed conflict.
To summarize, we expect peacekeeping and mediation to be complementary in halting hostilities. Peacekeeping matters for mediation in part by helping belligerents not only to reach a negotiated settlement (Kathman & Benson, 2019) but also to implement such an agreement (DeRouen & Chowdhury, 2018). Furthermore, by reducing and controlling hostilities, the impact of peacekeeping goes beyond the belligerents. Peacekeeping also allows nonviolent political actors, such as civil society organizations, to become relevant, and creates a context more conducive to mediation. The complementarities of peacekeeping and mediation 12 For a more detailed discussion of how battlefield violence can provide a source of information see, for example, Fearon (1995) and Werner & Yuen (2005). 13 Arnault (2006), however, also cautions against assuming that the United Nations is always able to guarantee physical security. 14 See, for example, JMAC SITREP -01 SEP 08; JMAC SITREP -12 Oct 08; JMAC SITREP -22 Oct 08; JMAC SITREP -26 Oct 08. therefore extend beyond their impact of negotiated settlements and should apply to terminating hostilities more generally. Accordingly, we expect: Hypothesis: There is a higher likelihood of conflict to end when mediation and peacekeeping are used simultaneously, in comparison to cases with either no conflict management or individual deployments of peacekeeping or mediation.

Research design
To assess the hypothesis, we evaluate the separate and conditional effects of mediation and peacekeeping on conflict termination using binary time-series crosssectional (BTSCS) models with 'conflict year' as the unit of analysis. 15 The models include the number of years since conflict started as temporal controls (Carter & Signorino, 2010). In our setup, BTSCS models essentially evaluate what factors affect the duration of conflict.
Since it is plausible that not all conflicts are equally likely targets for conflict management, the analyses are supplemented with Heckman models to account for possible selection effects (Fortna, 2008). The mediation and peacekeeping variables are lagged by one year to limit possible reverse causality. We run separate models for inter-and intrastate conflicts, which also allows us to include controls that specifically apply to either type of conflict. Finally, the international community has developed norms about the desirability and feasibility of conflict management that are distinct for inter-and intrastate wars. As a consequence, selection models are likely to be different for both types of conflict. To assess the substantive impact of peacekeeping and mediation, the best-fitting models are used to identify conflicts (ex post) where mediation and peacekeeping were critical to their ending. Based on this information, the counterfactual frequency of conflict is simulated. In effect, the counterfactual analysis suggests how much conflict the world would have experienced without mediation and peacekeeping.

Dependent variables
The Uppsala Conflict Data Program (UCDP) provides the information on inter-and intrastate conflicts from 1946 to 2013. UCDP defines conflicts based on a territorial or governmental incompatibility involving at least one government and applies a low threshold for inclusion in the data. In the UCDP data, conflicts become inactive when they fall below 25 fatalities in any particular year. Our dependent variable, Conflict termination, is coded accordingly, and allows us to consider multiple periods of hostilities associated with a particular conflict. 16 Conflicts terminate either through peaceful agreement (i.e. ceasefire or peace settlement), military victory (i.e. either the rebels or government are defeated), or simply by falling below the fatality threshold required for a conflict to remain in the UCDP conflict data. The analyses include all different types of conflict termination. The advantage of this approach is that it allows us to assess the influence of conflict management on the full universe of cases, not just those that end with a peace agreement. It is also a hard test for mediation and peacekeeping, since the influence of conflict management is expected to be stronger when conflicts end with a peace agreement. Therefore, if anything, the results underestimate the conflict management effectiveness of mediation and peacekeeping. 17

Independent variables
The key independent variables are the application of mediation and the deployment of a peacekeeping mission to a particular conflict. Mediation is coded using information from Civil War Mediation data (DeRouen, Bercovitch & Pospieszna, 2011) and International Conflict Management data (Bercovitch, Anagnoson & Wille, 1991). Both datasets use a broad definition of mediation as 'a process of conflict management where disputants seek the assistance of, or accept an offer of help from, an individual, group, or state, or organization to settle their conflict or resolve their differences without resorting to physical force or invoking the authority of law' (Bercovitch, Anagnoson & Wille, 1991: 8). To deal with the 15 We use yearly rather than monthly observations for a number of reasons. Most importantly, our dependent variable evaluates the cessation of hostilities on a yearly basis. Most control variables are also measured annually rather than monthly, and we do not want to introduce artificial variation. Information about peacekeeping and mediation is available at the monthly level but with relatively little within-year variation. As part of the robustness tests, we consider the duration and sequencing of peacekeeping and mediation, but they have no effect. 16 The analysis accounts for the history of conflict and uses the unique UCDP conflict-sequence identification to generate robust standard errors. We also control for the number of previous mediation efforts as well as the duration of peacekeeping. 17 This is supported by supplementary analysis in which we assess the influence of mediation and peacekeeping on each type of termination individually. In line with our theoretical expectations, the analysis reveals that both forms of conflict management only influence the likelihood of a peace agreement. temporal limitations of the original data, we extend the data from 2000 to 2013. Relying on the information on the start and end date for each mediation episode, our dataset identifies conflict-years with mediation.
The coding of UN peacekeeping missions is based on UN information on past and ongoing peacekeeping missions. In addition to a binary indicator capturing the presence or absence of peacekeepers, we also account for the type of peacekeeping mandate. Following the categorization of Doyle & Sambanis (2006: 11-18), and coded by Hegre, Hultman & Nygård (2019), we distinguish between traditional and transformative peacekeeping mandates. We extend their original data to interstate conflicts and civil wars taking place in 1946-60 and 2013-14.
To explore the conditional effects of mediation and peacekeeping, we create binary variables for each of the different combinations of conflict management techniques. In other words, one of following six binary variables applies uniquely to each conflict-year: No conflict management Mediation without peacekeeping Mediation with traditional peacekeeping Mediation with transformative peacekeeping Traditional peacekeeping without mediation Transformative peacekeeping without mediation Table I describes the frequency of the various combinations of conflict management in inter-and intrastate conflict. Table I further distinguishes between years in which conflict continues and final years of conflict. The vast majority of years -77% of intrastate and 80% of interstate conflict-years -do not witness any form of conflict management. Mediation alone is relatively common in both forms of conflict, appearing in around 16% of conflict-years, and becomes more common in the final year of conflict; mediation takes place in nearly 30% of all final years of interstate conflicts. In years with continuing conflict, peacekeeping is about as common with or without mediation, but mediation and peacekeeping are much more common in the final year of conflict. 18 Both traditional and transformative peacekeeping missions are deployed in intrastate conflict and regularly alongside mediation. Notably, there are no transformative peacekeeping missions associated with interstate wars.

Confounding factors
Arguably, both mediation and peacekeeping are applied to more challenging cases, in particular those with a high intensity of violence (Fortna, 2008). Belligerents often only agree to mediation or peacekeeping once they fully appreciate the costs associated with conflict. Given that both mediation and peacekeeping may be particularly applied to more challenging (so-called 'hard') cases, our research design provides a hard test. We also test for selection effect by means of Heckman selection models.
The models include a number of additional variables that are expected to affect the likelihood of conflict termination. First of all, we expect that more intense and violent conflicts are more difficult to end. Accordingly, we use the UCDP data to identify Major wars, which exceed the threshold of 1,000 casualties. Secondly, during the Cold War, the great power rivalry sustained the fighting efforts of belligerents making it less likely for conflicts to end. The Post-Cold War variable identifies conflicts taking place after 1989. The models for the ending of intrastate conflict include as additional control Territorial incompatibility (from UCDP data) to indicate territory as dispute issue rather than governmental control. The expectation is that territorial conflicts tend to last longer. Conflicts with more rebel groups should be more difficult to resolve; the Rebel groups count variable (Pettersson & Eck, 2018) controls for the complexity of the intrastate conflict. Both of the variables are not applicable for interstate conflicts. The interstate models include the number of Defense and entente pacts (Gibler, 2009) of the belligerents as a crude measure of international interest in the dispute. The mean Global Polity score (Gleditsch, 2017) and the Global count of concurrent Mediation and Peacekeeping missions control for the supply of mediation and peacekeeping respectively. Since interstate conflict is relatively rare in our sample (with 258 interstate conflict-year observations), we are limited in the type of analysis that is feasible; specifically, Heckman and bivariate probit models are ill suited for small samples and commonly fail to converge. There are 1,523 observations of intrastate conflict-years making it feasible to run Heckman selection models. The Heckman model requires identifying assumptions, and at least one additional explanatory variable that influences selection but not the outcome variable. We expect that existing commitments to conflict management influence the capacity and readiness to offer assistance, but do not necessarily influence the eventual success of any attempt. The mean global polity score, and the global count of concurrent mediation and peacekeeping missions are selection variables in the intrastate Heckman models for the supply of mediation and peacekeeping respectively. Finally, the selection model includes an indicator of previous mediation or peacekeeping, as we expect earlier conflict management efforts to increase the likelihood of mediation and/or peacekeeping. 19

Peacekeeping, mediation, and conflict termination
Between 1946 and 2013 there were 439 instances of conflict termination. Focusing on instances of conflict termination in Table I, we observe that mediation and peacekeeping regularly occur together, and this holds particularly in the final year of conflict. Mediation was present in 25% of all conflict terminations prior to 1989, rising to 35% in the period 1989-2013. Peacekeeping is less common than mediation in both periods but was still present in a notable 8% of all terminations prior to 1989, and 19% of terminations since 1990. In line with the complementarity hypothesis, terminations involving peacekeeping often also involve mediation (74%).
Models 1-3 in Table II present the analysis of the termination of intrastate conflict. A series of dummy variables tests for the conditional effect of mediation and peacekeeping. Mediation without peacekeeping is the reference category in all models, making it easy to recognize any complementarity of peacekeeping. All conflict management indicators are lagged by one year to lessen concerns about reverse causality. Model 1 includes controls to account for the intensity of the conflict and temporal dependence. Model 2 includes additional controls to account for the post-Cold War period, territorial disputes, and the complexity of the conflict measured via the number of rebel groups involved in the conflict.
In both models, mediation has a positive influence on the likelihood of termination of intrastate conflict, regardless of whether mediation occurs independently or alongside peacekeeping. In contrast, peacekeeping only significantly improves the likelihood of termination when it happens alongside a mediation process. Without a mediation process, peacekeeping is not more associated with the ending of intrastate conflict than instances without conflict management. Yet, transformative peacekeeping missions alongside mediation significantly increase the likelihood that civil wars end -while traditional peacekeeping missions are only in Model 1 significant at 10% level.
Based on Model 2, Figure 2 shows the marginal effects for each of the conflict management variables relative to instances with mediation but without peacekeeping, with all other variables held at their mean or modal values. 20 Without conflict management, the probability of an intrastate conflict terminating is 0.09 lower than when mediation is used. The presence of a transformative peacekeeping mission alongside the mediation attempts increases the probability of termination by 0.16 (relative to only having mediation). In contrast, the presence of a traditional peacekeeping mission does not significantly increase the effect of mediation. The exclusive use of (traditional or transformative) peacekeeping does not have a significant effect. To summarize, mediation is an effective form of conflict management in intrastate conflict, but most effective when combined with a transformative peacekeeping operation. Peacekeeping, in contrast, has no effect on the likelihood of termination when occurring without mediation. The coefficients of the control variables are generally as expected. Major wars and territorial disputes are less likely to end. While there is some evidence of a structural break with the end of the Cold War, the effect is not significant. Finally, more rebel groups complicate the resolution process and reduce the chance of civil wars endings.

Selection bias
Previous research indicates that the selection of mediation and peacekeeping is not random (cf. Fortna, 2008;Clayton, 2013;Clayton & Gleditsch, 2014;Diehl & Regan, 2015). To account for possible selection bias, we reran our analysis using a Heckman selection model (Model 3 in Table II). In this analysis, the first stage of the model estimates the likelihood of conflict management (mediation or peacekeeping). The second step assesses the influence of conflict management on the likelihood of intrastate conflict termination, while controlling for selection using the residuals from the selection equation. As the second stage of the Heckman model is limited to cases in which conflict management occurs, this approach can only assess the different forms of conflict management relative to each other. The Heckman model requires identifying assumptions, and, as explained above, the selection equation includes Global Mediation, Global PKO, Lag conflict management, and Global polity. The selection equation also includes Major war, Territorial incompatibility, and Rebel groups count, based on previous work that has shown high intensity and complex conflicts to be more likely to receive conflict management.
The results are broadly equivalent with the prior analysis. Mediation is more likely to terminate a dispute when accompanied by transformative peacekeeping, but no longer significant at 95% (instead p ¼ 0.056). Mediation with traditional peacekeeping produces an effect in the anticipated direction but is statistically insignificant. The likelihood of termination decreases when either form of peacekeeping occurs without mediation, but the results are not significant. The negative and significant (at 5%) value of the rho-statistics indicates that factors that increase the likelihood of conflict management reduce the chance of termination. Put simply, conflict management seems to occur in the more challenging cases. Complex, high-intensity conflicts are the most likely to receive conflict management, and a history of conflict management is the strongest predictor of future conflict management efforts.

Robustness checks
To further assess the validity of our findings we consider a variety of model specifications. 21 Firstly, we included additional controls to account for state strength using GDP per capita, democracy (polity) score, population size, external support, and conflict management history (Conflict management). The results are robust across these specifications. We also ran a variety of different specifications of the Heckman model, with different controls in both the selection and outcome phases. Mediation with transformative peacekeeping becomes insignificant (at the .05 level) in some of these models.
The size and significance of an interactive effect is dependent on the value of the other covariates in a logit model (Ai & Norton, 2003). As a further check of the stability of our results, we reran the analysis focusing independently on low intensity conflicts, post-Cold War periods, and territorial conflicts. We also address possible problems of temporal dependence by including additional temporal controls for the number of years since the use of each form of conflict management, and control for the sequence of methods. The results remain consistent.
One disadvantage with the Heckman model is that the termination analysis is limited to those cases that involved conflict management, meaning that we lose the comparison with cases without mediation or peacekeeping. To account for the latter, we reran the analysis using a bivariate probit model, which can simultaneously evaluate two related processes, and a multivariate process that models the likelihood of peacekeeping, mediation, and termination simultaneously. In most specifications, rho is insignificant, indicating that the processes are not significantly related, and the original logit analysis is unbiased. The findings are robust across models.

Conflict management and ending interstate conflict
In the sample, there are more instances of interstate conflict endings prior to 1989, which is unsurprising given the more frequent nature of interstate conflict in this period. Both before and after 1989, interstate conflicts generally terminated with the assistance of mediators and/or peacekeepers -47% prior to 1989, and 62% after. To assess the influence of conflict management on interstate conflict termination we replicate the above analysis with a few crucial changes. First of all, as there are no observations of transformative peacekeeping during interstate conflict, we evaluate only No conflict management, Mediation without PKO, Mediation with traditional PKO, and Traditional PKO without mediation. Instead of measures of incompatibility and number of rebel groups, we use the aggregate count of the number of defense and entente pacts held by the belligerents as a crude measure of the international interest in the dispute.
As Model 4 (Table II) shows, also in the case of interstate conflict, mediation significantly increases the likelihood of conflict termination. Mediation is effective regardless of its occurrence with or without peacekeeping. This is shown in Figure 3, which depicts the marginal effect of each combination of variables. While mediation increases the chances of termination, traditional peacekeeping applied alongside mediation does not make it more likely that fighting stops compared to mediation alone. Peacekeeping without mediation reduces the likelihood of termination. Possibly, traditional peacekeeping with lightly armed observers lowers the incentives of participants to halt low-level hostilities. However, peacekeeping without a mediation process is very rare in case of interstate conflict, so the result needs to be treated with caution.
The Heckman and bivariate probit models failed to converge in the interstate models. Instead, Model 5 includes a number of additional controls to account for selection bias in the supply of conflict management. None has a significant effect on the likelihood of conflict termination, and the results for mediation and peacekeeping remain robust. The inclusion of conflict management history does not affect the key results either. In summary, mediation is an effective conflict management tool in interstate conflict, both with and without traditional peacekeeping. Traditional peacekeeping has a negative effect on the likelihood of conflict termination when occurring without mediation and is comparable to having no conflict management at all.

Substantive effect of conflict management on conflict termination
The analysis so far has shown conflict management to have a statistically significant impact on the ending of conflict, but in itself this does not demonstrate that conflict management matters substantially: does it halt conflicts that otherwise would have continued, and if so, how many? We use counterfactual predictions to assess the substantive effect of conflict management on aggregate conflict trends.
The details of the counterfactual analysis are provided in the Online appendix, which also lists the historical cases where, following the counterfactual analysis, conflict management shortened ongoing hostilities. We first determine the probability that the conflict would have ended in a particular year for all cases where we observed either mediation or peacekeeping on the basis of Model 2 (intrastate conflict) and Model 5 (interstate conflict). To provide a counterfactual, we then calculate the predicted likelihood of conflict termination assuming that mediation and peacekeeping had not occurred for the same cases. It follows from our empirical analysis that conflicts are less likely to end without mediation and peacekeeping, but the impact of conflict management may not have been decisive. It is crucial to exclude conflicts that would have ended or continued regardless of any conflict management. For example, the conflict in Morocco would very likely have ended in 1959 regardless (the expected probabilities declined from 0.94 to 0.77). In contrast, in 1993 the civil war in Ethiopia was extremely unlikely to have ended (the likelihood of termination declined from 0.04 to 0.02 in the simulated absence of mediation). In other words, conflict management is decisive when in its absence the probability of conflict termination drops below a specific threshold, where we consider a range of thresholds. Finally, for each case where conflict management was 'decisive', we calculate how much longer the conflict would have been expected to last in the absence of the third-party assisted termination, based on the type of conflict (e.g. intra-or interstate), and the observed duration of the conflict at that point. In Figure 4, these simulated additional conflictyears are added to the actually observed conflicts using the UCDP data. Here, the dark grey fill represents the total number of intrastate conflicts, and the light grey fill the number of interstate conflicts. The solid, dashed, and dotted grey lines indicate our simulated number of interstate conflicts in the absence of conflict management for three different thresholds. The solid, dashed, and dotted black lines simulate the number of intrastate conflicts using the same thresholds.
We thus estimate that around 20 intrastate conflicts would have continued without conflict management (using 0.5 as the probability threshold value). This equates to 123 additional conflict-years between 1946 and 2013. During the Cold War, the difference is relatively marginal, as is clear from the manner in which all three threshold lines closely mirror the UCDP conflict trend line in Figure 4. Arguably, only in a few cases was mediation decisive to the ending of conflict; for example, when the Organization of American States (OAS) mediated in the Dominican Republic in 1965, but also deployed the Inter-American Peace Force. The most significant difference is in the post-Cold War period, in which our model predicts that without conflict management there would have been on average five additional conflicts per year between 1990 and 2013. A notable success was the mediation resulting in the Good Friday Accords which ended the conflict in Northern Ireland. The eventual halting of hostilities in Bosnia following the Dayton Agreement illustrates how mediation and UN peacekeeping can be seen as complementary. Extensive transformative UN peacekeeping missions were also deployed to the African Great Lakes area where, following our counterfactual analysis, they mattered in ending hostilities.
Regarding interstate conflict, we also estimate that 20 conflicts (again assuming the 0.5 threshold) would have continued in the absence of conflict management, equating to 63 additional conflict-years. The counterfactual analysis draws attention to the role of mediation but also observer UN peacekeeping missions in managing conflicts between India and Pakistan, Turkey and Cyprus, and Israel and Syria. Mediation and a small observer mission (MOMEP), with troops from Argentina, Brazil, Chile, and the USA, provided effective intervention in the short war between Ecuador and Peru in 1995. In the majority of cases, however, mediation primarily contributed to ending hostilities and no peacekeepers were deployed. For interstate wars, we find a greater positive influence of conflict management during the Cold War period.
Combining inter-and intrastate conflicts, the counterfactual analysis shows that conflict management already had an impact from the early 1960s to the late 1980s, largely because of the conflict termination efforts of mediators. This effect increases significantly with the end of the Cold War, as the advent of more robust forms of peacekeeping helped mediators to terminate conflicts that would otherwise likely have continued. Rather than observing a steep decline in violence throughout the early 1990s, our analysis suggests that without conflict management the number of conflicts would have at best stayed constant for another decade, and then merely returned to the level of conflict observed throughout the 1970s and 1980s.

Conclusions
Especially since the end of the Cold War, the use of peaceful forms of conflict management has increased significantly with a key role for the UN, making mediation and peacekeeping the primary means through which the international community now attempts to manage and resolve violent conflict. We have examined the extent to which mediation and peacekeeping are associated with the termination of hostilities. Since this is a shared objective of peacekeeping and mediation, it is arguably the best yardstick to compare their relative contributions. The first key insight is that both mediation and peacekeeping have a positive and complementary effect on conflict termination. The counterfactual approach moreover illustrates the substantive impact of mediation and peacekeeping, evaluated collectively as forms of conflict management, on the global conflict trends. However, it is important to account for important differences between type of conflict (interstate vs. intrastate conflict), period (Cold War and post-Cold War), and peacekeeping mandate (traditional vs. transformative).
A further notable finding is that any effect of peacekeeping is shown to be conditional on mediation, which underscores the recent call of UN Secretary-General Guterres to view peacekeeping as part of 'inclusive peace processes'. In line with a number of recent studies, in particular, Beardsley, Cunningham & White (2019), DeRouen & Chowdhury (2018), and Kathman & Benson (2019), we find that peacekeeping can matter for mediation. Specifically, transformative peacekeeping boosted the effectiveness of mediation to halt civil wars in the post-Cold War period. However, we found no evidence that peacekeeping mattered for ending hostilities in the absence of mediation. This underscores the importance of studying peacekeeping and mediation in tandem and provides evidence for the primacy of mediation.
We recognize that peacekeeping could work primarily by reducing the likelihood of conflict recurrence (Sambanis & Doyle, 2000;Fortna, 2008). Figure 1 shows that mediation occasionally continues post-conflict, and we find some indications that peacekeeping and mediation also complement each other in reducing the chances of conflict recurrence. It is, however, very difficult to disentangle the effects of mediation and peacekeeping on conflict recurrence. Mediation that continues post-conflict is only rarely recorded and often remains largely informal. It is commonly recognized that the UN mediates as part of the deployment of a peacekeeping mission, which further interweaves mediation and peacekeeping.

Replication data
The dataset and do-files for the empirical analysis in this article, along with the Online appendix, are available at https://www.prio.org/jpr/datasets/. All analyses were conducted using STATA 15. numerous colleagues who commented on prior versions of this article.

Funding
We gratefully acknowledge invaluable financial support from the Folke Bernadotte Academy.