The Short-Term Effects of Mediation on Low-Intensity Civil Wars

Civil wars are notoriously intractable. Scott Gartner (2012) has outlined several reasons for this intransigence. First, the asymmetrical nature of civil war makes it less likely that the government will negotiate because it expects to win outright. Second, negotiating with a rebel group can be anathema to a government that fears that doing so will grant legitimacy to the rebels. Third, many civil wars represent all‐or‐nothing affairs in which there is little or no room for compromise. For example, a revolution's stated goal of taking over a government or deposing a leader may leave little room for negotiation. Fourth, many nations do not welcome third‐party intervention into domestic conflicts, which closes an important avenue for negotiation. Fifth, many rebel groups enjoy support from another state or states. Finally, Gartner argues that rebels, because they are “un‐official” actors, will often not live up to agreements. Others (e.g., Fearon 2004) have argued that governments are also likely to fail to live up to agreements.

Despite these roadblocks to third‐party involvement, mediation to prevent or manage the outbreak of civil war, curb civil war violence, and/or prevent an escalation has become an increasingly popular option in the post–Cold War era (see DeRouen and Bercovitch 2012). Mediation is often perceived as requiring years of painstaking work to yield concrete positive results. Civil wars in Bosnia, El Salvador, Guatemala, Indonesia (Aceh), and Sudan, for example, were mediated for many years before final success became evident. We have found reasons to believe, however, that mediation can yield short‐term success in civil wars.

This article assesses the ability of mediation to de‐escalate low‐intensity civil wars in the short term. Using a short time perspective on conflict behavior is a novel way of measuring mediation effectiveness. Even short‐term success can open the door to humanitarian aid access and provide a foundation for a comprehensive peace. Karl DeRouen, Rob Kevlihan, and Glen Biglaiser (2013) have reported that civil war mediation has a positive effect on humanitarian aid. It is plausible that potential donors will see mediation as a sign of commitment to the peace process, and therefore aid has a greater chance of having a propitious effect on resolving the crisis. The U.N. Office for the Coordination of Humanitarian Affairs (OCHA) (2011) observes that humanitarian action can help build long‐term peace by building capacities. Stronger state capacity can be a springboard upon which development agencies can build. Second, humanitarian aid donors would do well to incorporate conflict experts in the planning stage.

Kyle Beardsley (2011) demonstrated that mediation can provide short‐term respite. In the short run, he wrote, mediators offer incentives to disputants that result in a wider range of acceptable solutions, commit to postconflict security guarantees, and provide political cover for leaders who can become vulnerable if they make concessions. He argues, however, that mediation often fails in the long term. Specifically, Beardsley (2008) found that, although mediation of international crises leads to short‐term settlements, in the long term the probability of recurrence is higher than if there had been no mediation. Our study differs in that we focus on the effects of mediation on whether civil wars are active (at least twenty‐five battle‐related deaths) in subsequent years rather than on international crises and formal agreements.

Our analysis takes several important conflict dimensions into account: type of war (territorial versus control of government), mediator identity (permanent member of the United Nations Security Council [P5], United Nations [U.N.], or neighboring state), mediator focus (disputant behavior or root incompatibility), and mediator strategy (direct or indirect talks). We also consider the potential differing impact of mediation in territorial wars and in democratic countries. Finally, we test whether bilateral talks have a positive effect on the probability of a civil war being mediated.

After controlling for polity score (level of democracy), conflict duration, per capita gross domestic product (GDP), and whether or not a peace agreement was signed, we found that direct forms of mediation in which all parties meet face to face were the most likely to yield short‐term success in our sample of civil wars occurring from 1993 to 2004. That is, mediation using direct approaches worked when the focus was on either behavior or incompatibility whether the mediation was conducted by a permanent member of the United Nations Security Council (hereafter referred to as “P5”), by the U.N., or by a neighboring state. (The five permanent members of the United Nations Security Council are China, France, Russia, the United Kingdom, and the U.S.) Second, P5 and neighbor mediation prove themselves the most versatile because they seem to be effective when the sample is limited to only democracies or only to territorial wars. We found some evidence that P5 indirect mediation (not all parties meet together) works in the democratic and territorial war samples, and neighbor indirect mediation works in the democratic subsample. Bilateral talks (in which the third party does not pass information between disputants) increase the probability of subsequent mediation. Finally, P5 mediation (any focus or style) was effective no matter the nature of the war.

All too often, talks between disputants offer little hope of resolving civil war. Negotiated settlements are fraught with pitfalls, such as credible commitment and implementation problems. In these cases, peaceful third‐party diplomacy offers great promise. We define a third party as an actor involved in either helping to regulate the conflict incompatibility, to change conflict behavior, or to regulate other conflict issues (see Wallensteen 2007).

Third‐party diplomacy is a set of nonviolent measures taken by third parties in armed conflict situations. Mediation is an especially useful form of diplomacy in which a third party assists disputants by acting as a go‐between. Third parties can be states, intergovernmental organizations, individuals (e.g., former President Jimmy Carter), nongovernmental organizations, or religious organizations (see Bercovitch and Schneider 2000). Third‐party mediation helps settle the “conflict or resolve … differences without resorting to physical force or invoking the authority of law” (Bercovitch, Anagnoson, and Wille 1991: 8).

Third‐party diplomatic talks can be broken down into four types: indirect, direct, bilateral, and unclear (see Möller et al. 2007). In indirect talks, the disputants talk indirectly through the third party, which relays information from one party to the other — thus, it is a form of mediation. Bilateral talks include talks between the third party and only one of the warring parties. Bilateral talks do not necessarily imply that information is passed between disputants, and as such do not comprise mediation as it is conventionally defined. Often, the third party is only talking to the government. Direct talks are a form of mediation in which the combatants meet face to face with a third party present (these are further described below.) Unclear talks may be either direct or indirect or both, but it is not clear which (these unclear cases are included in our sample of all mediations.)

Traditionally, research on conflict prevention and management has focused on case or comparative studies. The first wave of studies of preventive diplomacy identified the importance of third‐party timing, the effectiveness of mediation by influential international diplomats and major powers that can provide credible guarantees, and the importance of combining diplomatic tools (Lund 2008). In recent years, research has shifted toward large‐sample quantitative studies, a process that has been facilitated by the increase in quality data. Some of the concepts being explored recently by researchers include mediator qualities and identity (see Bercovitch and Gartner 2006), mediation timing (see Beardsley 2008), and mediator bias (see Svensson 2007).

One general finding has been that if an armed conflict avoids escalation in the short term, for reasons not fully understood, the long‐term risk of escalation drops sharply (Heldt 2008). Thus, it is important for third parties to focus their attention on averting an escalation in the early phases of a conflict to avoid a large‐scale escalation later.

Previous studies have pointed to the effects of a number of international measures on the risk of escalation or negotiated settlements (e.g., Melander, Möller, and Öberg 2009; Öberg, Möller, and Wallensteen 2009). For instance, Erik Melander, Frida Möller, and Magnus Öberg (2009) found that arbitration and “good offices” (diplomacy in which the third party does not actively take part in direct talks with the disputants but rather facilitates talks by providing a venue and facilities, for example) are associated with a decreased risk of escalation to war, whereas permanent observers are associated with an increased likelihood of escalation to war. Birger Heldt (2008) found that third party‐aided direct talks as well as mediated indirect talks increase the probability of a negotiated settlement.

Not only do the diplomatic initiatives themselves matter, previous studies have shown that the identity or type of third party also affects the outcome (Heldt 2008; Melander, Möller, and Öberg 2009). In fact, in some instances, the type of third party seems to be more important than the type of measure. For instance, they found that mediation by the U.N. and neighboring states is associated with an increased risk of escalation to war, and that only mediation by a P5 member is significantly associated with a lower risk of escalation to war.

Several previous studies have reported that mediation can reduce the probability that a conflict will escalate to a civil war of greater than one thousand deaths (see, e.g., Möller et al. 2007; Heldt 2008; Melander, Möller, and Öberg 2009; Öberg, Möller, and Wallensteen 2009). But we expect more of diplomacy: if it is truly effective, we argue, it should also lead to the de‐escalation of low‐intensity civil wars. Put simply, we expect that, in general, armed conflicts will de‐escalate if third parties engage the belligerents in mediation activities. We expect mediation to help parties overcome the many difficulties they face and shift the focus from fighting to constructive problem solving.

We have tested this hypothesis using a new research design. Specifically, the models we have tested focus on whether a low‐intensity (25–999 deaths) civil war was active in the year following the mediation. In other words, instead of looking at the effects of mediation on conflict onset or settlement, we explore the ability of mediation to de‐escalate a low‐intensity civil war.

Our first hypothesis is general. We look at the effect of any mediation on the probability of a low‐intensity war being active in the next year. In general, a cease‐fire is a minimum starting point for mediators and conflict managers. Mediation is designed to help the parties bridge the information gap, increase trust, reduce uncertainty and misperceptions, and overcome many of the difficulties the parties face (Weiss 2003). The well‐respected Armed Conflict Dataset of Uppsala Conflict Data Program Peace Research Institute Oslo (UCDP‐PRIO) (see Harbom and Wallensteen 2010) uses twenty five as the threshold for a dispute to be called an armed conflict; thus, a decrease in the number of casualties to less than twenty five is a reasonable yardstick for success. The first hypothesis is simply:

Mediation sequencing refers to the different approaches mediators take in a mediation process. The “gradualism” approach is based on the belief that mediators should create trust and a positive atmosphere between the belligerents by moving from simpler to more difficult and complex issues (Weiss 2003). When mutual trust has been gained, the mediator can move to more difficult issues (Gilady and Russett 2002). One way to reach this point is for the third party to meet the belligerents separately in bilateral talks. Taking time and getting to know the parties and their positions, the third party gains the trust and can get the parties to agree to mediation.

Thus, our second hypothesis is:

We now turn attention to mediator identity. Previous studies have shown that powerful nations are more successful in preventing an escalation of a low‐intensity armed conflict (see Melander, Möller, and Öberg 2009). An explanation for this finding is that major powers can provide credible guarantees and diplomatic efforts such as mediation may be more effective when — at least implicitly — they are backed by a credible threat of force or commitment of resources. Thus, powerful nations, such as the permanent members of the United Nations Security Council (P5), may be more effective mediators because of their relative military and economic clout (Melander, Möller, and Öberg 2009).

The U.N. is the most frequent civil war mediator (DeRouen and Bercovitch 2012). United Nations involvement in peacekeeping and peacebuilding has been productive because the organization has a strong capacity (see Doyle and Sambanis 2000). But the U.N. “selects in” to (chooses to participate in) more complicated and intractable conflicts, which could lower its observed involvement success rate (DeRouen 2003). Because mediation may be more effective if the measures are backed by a credible threat of force or commitment of resources, mediators with power and resources, such as the P5, may be better suited to steer a conflict toward peace. The U.N. — an organization dependent on its members and without any unilateral power — would consequently be a less effective mediator.

In some regions, mediation by neighbors occurs frequently (Möller and Svensson 2007). Neighboring countries often mediate if they have a vested interest in the outcome of the war. Neighbors often fear that rebels will cross into their country and cause unrest, or will use the country as a base of operations (e.g., Karen National Union troops fighting in Myanmar could cross into Thailand). Another concern is that the country will become a destination for refugees from a nearby civil war. Finally, neighbors may prefer one party over others because of historical, ethnic, or economic ties. Any combination of these factors can bias a neighbor toward one side in a civil war. Biased neighbors sometimes mediate, and Isak Svensson (2007, 2009) has reported that biased mediators can increase the likelihood of negotiated settlements. Biased mediators can help shore up credible commitment problems: if the first side agrees to mediation knowing that the mediator is biased toward the second side, then the first side will be seen as credibly committed to the peace process.

Our next two hypotheses are:

Our next hypothesis combines mediator identity, mediation strategy (direct versus indirect), mediator focus (behavior versus incompatibility), war type, and the probability of active war in the subsequent year.

As before, we expect P5 mediators to be the most effective, and we expect territorial wars to be more amenable to mediation that uses an incompatibility focus. We expect this unique P5 effectiveness to carry over and be apparent regardless of style used, focus, war type, or regime type. In other words, P5 mediators should be noticeably more successful in any of these situations.

Öberg and Möller (2010) explored the relevance of the sequencing of mediation activities. They reported that mediations that initially focus on the central incompatibility of the parties are significantly less likely to reach a negotiated agreement. The implication is that dealing with “easier” topics early in a mediation process, rather than starting with tough core issues, is more effective. Dealing with disputant behavior is, thus, conceivably “easier” regardless of the identity of the mediator. In such cases, success is often measured by the signing of cease‐fires and process agreements as opposed to full agreements. Thus, we posit that mediation will be relatively more likely to reduce the probability of violence in the following year when it deals with disputant behavior. Mediation that deals with the root cause of the conflict, however, should also be expected to lead to de‐escalation, but is likely not to be manifested in the short term as it takes time to resolve deep‐seated incompatibilities.

As stated above, mediation sequencing theory refers to the different approaches mediators take in a mediation process. The order in which a mediator addresses issues in a mediation process is one potential point of mediator influence. The literature explores two main sequencing styles: the aforementioned gradualism approach and the so‐called “boulder in the road” approach (Weiss 2003). Mediators who employ the first approach try to move from simpler to more difficult and complex issues. Their rationale is that trust between the belligerents is low, and taking small steps will create initial trust and a positive atmosphere (Weiss 2003). By the time the parties reach the more difficult topics, according to this theory, they will have built mutual trust as well as momentum (Gilady and Russett 2002).

Mediators who use the latter approach recommend addressing the more difficult issues first, thereby removing the “boulder,” the most difficult issue or stumbling block, and thus facilitating resolution of the remaining issues (Weiss 2003).

According to the gradualist, mediation efforts function as steps to gain trust with the goal of reaching an understanding. Before parties are ready to meet face to face, indirect talks might be necessary to develop a framework for negotiations, set boundaries, create momentum, and start building trust. But to reach deals, hammer out difficult provisions, and sign agreements, direct talks are likely to be necessary. By preceding direct talks, indirect talks can allow the disputants to gain a better understanding of the adversary that might dispel the rumors, distrust, and innuendo that build during war.

Researchers have examined the impact of mediation sequencing. Heldt (2009) found that the order of various third‐party actions can influence whether an emerging conflict ends through a cease‐fire agreement or a peace agreement. He also found that the relative timing of good offices vis à vis mediation matters as does the relative timing of direct talks vis à vis indirect talks (Heldt 2009).

In conclusion, we expect direct talks to be more effective once trust has been gained and momentum has been created along the way through a process of indirect talks.

Next, we divided the full sample into two subsamples: democratic states and territorial (secessionist wars) war types. The international attention that often accompanies mediation might make the disputants less likely to kill so that they are not perceived badly. This effect will be intensified in a democratic state with a free press. Democratic states might be more willing to enter into good faith mediation efforts because they have institutionalized conflict management mechanisms, transparency, and credibility. Some democracies, however, may prefer to handle situations without outside intervention because of these very same mechanisms. In other words, democracies prefer to use their own conflict management institutions and norms to handle their civil wars. Further, democracies are, by and large, the wealthiest and most successful states, and their heightened status may make them reluctant to accept outside help. Karl DeRouen, Jacob Bercovitch, and Peter Wallensteen (2012) reported that secessionist wars are more likely to recur than are wars over governments. This has also been reported by James Fearon (2004), who attributes this to credible commitment problems.

Mediators often favor the implementation of democratic reforms as part of the conflict management process presumably because these states are more transparent and as a norm more likely to favor peaceful conflict resolution. As Wallensteen and his colleagues (2009: 254) noted, “contentious issues which arise in the implementation of a peace agreement are more likely to be dealt with peacefully within a democratic system, rather than becoming a trigger for conflict recurrence.” Further, they noted that democracy is fully consistent with power sharing, and several scholars have reported the positive benefits of peace agreements that divide power (see, e.g., Hartzell and Hoddie 2003).

Casualty rates are often lower in territorial/secessionist (the terms are used interchangeably) than in other kinds of civil wars (see Fearon 2004), and diplomacy may have less of an impact on the death rate of these wars because they are already low. In territorial wars, the sides are entrenched, and it may be harder to change their behavior. Addressing the root incompatibility may be the only way diplomacy can work in territorial wars. But initial discussions about conflict behavior may be necessary before undertaking the more difficult talks over core issues that eventually will lead to de‐escalation. In other words, the mediator could initiate the process with indirect talks, shuttling between the parties, discussing topics that are easier to approach, such as conflict behavior, to improve the atmosphere and make the parties feel more secure. When the time is ripe, the parties begin discussing more difficult issues, such as their main incompatibility, face to face in direct talks with the mediator present. Using this type of sequencing of formats (indirect/direct talks) and topics (behavior/incompatibility), the mediator can enhance the probability of a positive outcome.

It is also possible that the order in which the topic of the conflict is addressed could be important. In other words, a mediator could address either the root cause of the conflict or the behaviors of the disputants first. Öberg and Möller (2010) examined whether the sequencing of issues in the mediation process is associated with the likelihood of a negotiated agreement (peace or cease‐fire agreement). They found that mediation that initially deals with “easier” topics, that is, issues other than the incompatibility and conflict behavior, has a significantly higher likelihood of reaching a negotiated agreement that terminates the conflict. In contrast, mediation that starts with such difficult issues as core incompatibilities is less likely to produce an end to the conflict. Furthermore, they also found that beginning with issues related to conflict behavior does not significantly affect the likelihood of reaching an agreement.

In sum, evidence indicates that the issues discussed have an effect on the outcome of a mediation process. Although Öberg and Möller's (2010) findings suggest that discussing conflict behavior is a less effective strategy for reaching an agreement, we believe these topics can have a short‐term effect on conflict activity because provisions expected to have a quick impact on casualty rates, such as cease‐fires, demilitarized zones, and the withdrawal of troops, are often discussed in the talks.

Our fourth group of hypotheses is as follows:

The conflict and mediation data we have examined in this study are from the Managing Intrastate Low‐Intensity Conflict (MILC) data set compiled by the UCDP in Uppsala, Sweden, and the Folke Bernadotte Academy in Stockholm. The MILC data comprise 3,018 third‐party events that occurred in 127 intrastate, low‐intensity dyadic conflicts over the period 1993 to 2004. The data set categorizes nine different types of third‐party measures directed at low‐intensity armed dyads during this period: indirect, direct, unclear, and bilateral talks; good offices; arbitration; fact‐finding missions; permanent observers; and peacekeeping operations (Melander, Möller, and Öberg 2009).

For our study, we included years in which the conflict exhibited minor intensity (i.e., 25–999 battle‐related deaths) from 1993 to 2004. We used the UCDP‐PRIO database (Harbom and Wallensteen 2010) to create the dyad year array. Our observations are based on the MILC data set, which in turn is based on UCDP database version 2005. The UCDP has since revised the data, however. Accordingly, the dyad years have been adjusted to match the latest update of the UCDP database. We have made the following revisions:

1. In instances in which a dyad year has been “upgraded” to war, years following this “war‐year” have been excluded. For example, the Taliban dyad in Afghanistan has been “upgraded” from minor intensity in 1995 to a level of war. Therefore, we have excluded all years from 1995 forward from our data.

2. In instances in which a dyad year has been “downgraded” to inactivity, that year has been excluded from the data set. For example, the Movement for Democracy and Development (MDD) dyad in Chad 1998 has been “downgraded” from minor intensity to inactive. Therefore, for the MDD conflict, we excluded 1998 from the data.

3. Some dyads have been excluded altogether by the UCDP for various reasons. For example, the Harakat Amal dyad in Israel has been removed from the UCDP data set as it did not fulfill the incompatibility criterion when examined in greater detail.

4. The conflict type of some dyads has, after scrutiny by the UCDP, been changed from territorial to war over control of government. For example, rebel groups fighting over the territory of Southern Sudan have now been recoded as participating in a conflict over government power. Also, the Union of Armed Resistance Forces dyad in Niger has been moved from the territorial conflict to government power conflict category.

We did not transfer some UCDP revisions to our data set, however. For instance, some newly found active years (a previously inactive year that has been revised to active) were not transferred because we found no data on diplomatic activities for these added years — we simply left them out.

Our dependent variable is whether the subsequent year following the mediation was an active one (at least twenty‐five battle‐related deaths). Thus, it is a dichotomous variable.

Our key independent variables are mediator identity, focus, and strategy. For all types of talks, the MILC data set includes the focus of the mediation, differentiating between mediation that focuses on root incompatibility, conflict behavior, and other conflict issues. Having a focus on incompatibility means the parties would be discussing the incompatible positions that are at the core of the conflict, such as transition of power, division of territory, integration of former rebels into the army, and electoral reform. Mediation focused on conflict behavior would seek to transform the behavior of the parties, by, for example, encouraging cease‐fires, the creation of demilitarized zones, and the withdrawal of troops. Examples of “other conflict issues” that could be discussed would include how to deal with refugees, the implementation of future talks, and general security concerns.

In terms of strategy, we have focused on direct talks, indirect talks, and bilateral talks, as defined in the MILC. Direct talks are face‐to‐face talks between belligerents with a third party present. Between 2000 and 2003, for example, mediator Henry Dunant Centre brought the Indonesian government and the GAM (Free Aceh Movement) together for direct talks in Geneva, which led to a cease‐fire and agreements on further meetings.

A similar type of mediated talks are indirect talks in which the warring parties communicate via a third party but do not meet in person. A typical example of this is the so‐called shuttle diplomacy, in which the mediator goes back and forth from one party to the other. In Burundi in the early 2000s, South African politician Jacob Zuma and mediator Nelson Mandela met the National Council for the Defense of Democracy – Forces for the Defense of Democracy rebels and the Burundi government separately in an intense effort to get the rebels to come to the negotiating table and to agree to a cease‐fire. Their shuttle diplomacy succeeded in getting the government of Burundi and the rebels to the negotiation table for direct talks. The negotiations, although marred by problems, yielded a cease‐fire agreement in October and a peace agreement in December of 2002.

A third strategy is bilateral talks. These talks occur between one or both of the belligerents and the third party, and do not comprise mediation according to the standard definition because the third party meets with the party(ies) separately and does not share information between the parties. This type of talk might occur when a party(ies) does not desire mediation. Bilateral talks can often include situations in which the third party simply explores the positions of the parties (Melander, Möller, and Öberg 2009).

In addition to coding for the type of diplomatic activity, we have also included information on the identity of the third party. We focus on three categories: P5 country, U.N., or neighboring country.

We have also controlled for whether a peace agreement was signed; income as measured by per capita GDP; the polity score, which is largely a measure of the level of constraints on the executive, with greater restraint being associated with stronger democracy; the duration of the war; and whether the war was territorial in nature.

We would expect the presence of a peace agreement to be an important determinant of whether the conflict is active in the subsequent year. Research indicates that peace agreements loaded with power‐sharing provisions lead to stable peace (e.g., Hartzell and Hoddie 2003). We have also controlled for peace agreements because most of them contain cease‐fire provisions (see DeRouen, Lea, and Wallensteen 2009), and therefore would be expected to have short‐term impact on subsequent casualties. These agreement data are from the UCDP Peace Agreement Dataset (Harbom, Högbladh, and Wallensteen 2006).

We expect states with higher incomes to have greater capacity to implement peace (see DeRouen et al. 2010). The implementation of agreements costs money and requires an adequate bureaucratic infrastructure. We use GDP per capita as a control for state capacity. Higher levels of development also increase the opportunity costs for rebels. These data are from the Penn World Tables (Heston, Summers, and Aten 2006).

We have calculated war duration from UCDP data. Our expectation is that, as a war goes on for longer, the probability it will end in a given year decreases. For example, the probability of very long low‐intensity wars in India or Myanmar ending in a given year is low. Some wars, such as those in Colombia, Myanmar, and India, have endured so long that the odds of mediation having a short‐term success seem long, whereas relatively new wars may have a chance of being successfully mediated before hostility levels have locked in.

The polity score is used as an indicator of regime type. These data are from Monty Marshall and Keith Jaggers (2006), and are measured on a twenty‐one‐point scale. In general, we expect that higher democracy scores would decrease the probability that a conflict will be active in the subsequent year because democracies are believed to be more amenable to peaceful resolution of disputes (see Dixon 1994; Hegre et al. 2001; DeRouen and Goldfinch 2005; Greig 2005). The relationship between civil war and regime type may not be linear, however. Havard Hegre and his colleagues (2001) reported that civil war is more likely in intermediate (anocratic) regimes, and found no difference between autocracies and democracies.

We have coded territorial war category as 1 if the war concerns autonomy or secessionist claims. Wallensteen and his colleagues (2009) reported on the intractability of territorial wars, noting that these conflicts are long and generally more difficult to end than wars over control of government, ideology, and national power. One view is that territorial civil wars are less amenable to mediation because, as Fearon (2004) argued, the rebels do not view the government as a credible negotiating partner in these wars. Another view is that because of credible commitment issues, mediation is necessary to obtain an agreement.

Because the observations are dyad years (i.e., each observation is defined by an active year of fighting between a rebel group and a state) and the dependent variable is dichotomous (i.e., two possible outcomes: active or not active), we specify cross‐sectional time‐series logit models. This method allows us to examine whether mediation has an impact on conflict activity in the subsequent year. We conducted diagnostics using Hausman (1978) and Wooldridge (2010) tests, which indicated a need for fixed‐effects corrections in a few cases or AR(1) corrections (meaning first‐order autoregression; in other words, the successive error terms are correlated), as indicated in the tables.

The reference (comparison) category in the statistical models (presented below) is nonmediation. In other words, we are able to ascertain the impact of mediation by comparing mediated cases to ones that were not mediated (the reference category). These additional conflict management initiatives could include the following: bilateral talks, unclear talks (with the exception of those identified in Table One), arbitration, good offices, fact‐finding, and U.N. Permanent Observer Missions. The category in the models (see Table Two) with bilateral talks on the right‐hand side includes the same list of alternatives except for bilateral talks.

Below we report our findings and, where applicable, we refer to corresponding or supporting findings in the literature on diplomatic initiatives. This study is the first to assess the impact of mediation on short‐term conflict behavior. Previous research (e.g., Beardsley 2008) has tended to focus on whether mediation ends conflict via formal agreement, such as a cease‐fire or treaty, or whether mediation can reduce the probability of civil war recurrence (e.g., DeRouen and Wallensteen 2013).

First, our findings run somewhat contrary to Gartner's (2012) argument that mediation takes time before concrete results can be attained. Our findings strongly suggest that mediation can rather quickly help defuse a civil war. Of course, this is not the same as achieving a comprehensive agreement, but even a short‐term period of peace can provide a respite that makes it safe for aid workers to enter unstable regions (DeRouen, Kevlihan, and Biglaiser 2013) and create a foundation on which mediators can build more lasting agreements.

In their report on the effects of the sequencing of mediation on negotiated agreements, Öberg and Möller (2010) found that mediation that begins with less contentious topics than the core incompatibility and conflict behavior has a greater chance of leading to a negotiated agreement. On the other hand, mediation that opens with the more difficult issues is relatively less likely to achieve success. Our findings differ in that we found evidence that, at least for direct mediation, a focus on both behavior and incompatibility seems to work. We note that ours is not a sequencing study, and our research design does not differentiate early mediation from later.

The results pertaining to the first hypothesis can be seen in Table One. They do not support our first hypothesis (Hypothesis One) that any form of mediation will increase the probability of a low‐intensity civil war becoming inactive in the year following mediation. Peace agreements, as expected, increase this probability. Also, as we expected, conflict duration correlates positively with the probability of the conflict staying active in the next year. In other words, the longer a war endures, the greater the likelihood it will be active in any subsequent year.

Surprisingly, we found a positive effect for the presence of democratic institutions on the dependent variable (Hypothesis Four [c]) — in other words, the probability of the conflict being active in the subsequent years increases as the level of democracy increases. This challenges our expectation in Hypothesis Four (c) that democracies will have greater success with mediation. Several democracies — Sri Lanka, Spain, and India, for example — have been plagued by long‐running civil wars. Perhaps these countries are reluctant to allow third‐party help because of the idea that a democracy should be able to peacefully resolve its own conflicts. Also, this could be related to the curvilinear effect of democracy that Hegre and his colleagues (2001) identified, in which repressive states and full democracies are more peaceful, but states in the mid‐range are more unstable.

Perhaps the failure to substantiate Hypothesis One is not surprising. Mediation does not always work. Whereas many studies (e.g., Walter 2002; Svensson 2007; Melander, Möller, and Öberg 2009) point to mediation's effectiveness, others doubt that it provides long‐term success (see Beardsley 2011), or argue that mediations often fail because mediators choose to mediate the most difficult cases (see Gartner 2012). History is replete with cases of failed mediation — for example, in Israel, Sri Lanka, Angola, the Philippines, and Aceh — that underscore that mediation comes with no guarantee of success.

The logit models tested in Table Two indicate robust support for our second hypothesis, which states that bilateral talks will increase the probability that mediation will occur in subsequent years. We argue that bilateral talks enhance trust and make the sides more willing to undertake a mediation process. All bilateral talks — talks with government and talks with rebels — have a positive effect on the probability that mediation will take place in the subsequent year. This is evidence of a sequencing process.

We also found peace agreements to have a positive effect on the probability of subsequent mediation. Our results in Table Two column one indicate that the probability of mediation increases as state capacity increases. The Table Two results are interesting for several reasons. Previous studies have examined the sequencing of mediation approaches (i.e., indirect and direct talks) and found that it is much more likely that the parties will reach a negotiated end if indirect talks are held before third‐party moderated direct talks (e.g., Heldt 2009). No previous research, however, has looked at whether initial bilateral talks affect the chances that mediation will occur in the first place. This finding has policy implications because it offers new insights about the different ways that a mediator could effectively initiate a mediating process.

Hypothesis Three (a) involves mediator identity. Hypothesis Three (a) posits that mediation will be more likely to reduce the probability of violence in the subsequent year if it is backed by the credible threat of force or commitment of resources that a permanent member of the United Nations Security Council possesses. This would support previous studies that found that mediation by a P5 state significantly lowers the risk of escalation to full‐blown war (Melander, Möller, and Öberg 2009). Relatedly, it would also add new insights into mediation effectiveness by finding that the involvement of powerful mediators (such as a P5 country) can increase short‐term de‐escalation behavior.

Our results in Table Three confirm Hypothesis Three (a): P5 mediators appear to be more effective at de‐escalating low‐intensity wars. (The coefficient for neighbor states was also negative but not significant.) Major powers can often provide credible guarantees and can back up mediation efforts with credible promises of reward or threat of force. Again, we found the duration of the conflict and the country's polity score to correlate significantly with a likelihood that the conflict would be active in the following year (see Table Three.)

Hypothesis Three (b) posits that mediation is more effective when conducted by neighbors. This hypothesis was not supported. Melander, Möller, and Öberg (2009) also found mediation by neighbors to be ineffective. In fact, they found this type of mediation — as well as mediation by the U.N. — to be associated with an increased risk of escalation to war.

Our Hypothesis Four (a) is that direct mediation focusing on behavior should be more effective than indirect mediation in reducing the probability that the civil war will become inactive in the following year. Recall that Öberg and Möller (2010; see also Weiss 2003; Heldt 2009) have argued that leading with a focus on behavior is more likely to build trust over time. Heldt (2008) reported that direct and indirect mediation increase the chance of a negotiated settlement. The relative effectiveness of the mediator's focus on behavior should hold for all kinds of mediators. As indicated in Table Four A, we found this to hold for the three categories of mediators that we studied (P5, U.N., and neighbor mediators) when they used this approach. Table Four B, however, reveals the same results for all three mediator categories using direct approaches, but with a focus on incompatibility. Thus, we conclude that the approach the mediator takes trumps the focus of the mediation.

Hypothesis Four (b) posits that mediation focusing on behavior will be more likely to produce short‐term results. This should hold for all mediators and would support the gradualism argument (Weiss 2003). Again, we found that the focus of the mediation does not seem to matter. Tables Four (a) and Four (b) indicate that direct approaches work regardless of the focus.

Having established that a direct approach seems more fruitful at eliciting short‐term results, we divided the sample into democracies and territorial civil wars to ascertain whether the choice of mediation focus and direct versus indirect approach matters in these cases.

In Hypothesis Four (c), we expected that mediation would help de‐escalate civil war in a democracy because violence in a democracy is more transparent, citizens of a democracy have higher expectations of their government, and democratic governments have greater credibility than nondemocratic governments. In Hypothesis Four (d), we predicted that, to the extent diplomacy has an impact on de‐escalation in territorial wars, mediation that takes the core incompatibility of the combatants into account would be relatively more effective.

To save space, we report a summary of the results from Hypotheses Four (a) through Four (d) in Table Four (c) (note that this table also repeats findings reported above in Tables Four (a) and Four (b)). Hypothesis Four (c) is largely borne out: we found five of the six forms of direct mediation to be successful in democracies. Representatives from P5 countries and the U.N. have successfully mediated in democracies using both direct and indirect approaches, a finding that somewhat supports existing research (see Wallensteen et al. 2009, reporting on the successful mediations of Southeast Asian civil wars). Hypothesis Four (d), which states that an incompatibility‐focused mediation is the strategy more likely to work in territorial wars, is partially validated. Mediation by P5 members, both direct and indirect, and direct mediation by neighbors were effective in territorial wars when focused on the core incompatibility of the disputants. However, P5 and neighbor direct efforts were also effective when focused on behavior. Research has established the intractable nature of territorial civil wars. In particular, governments will often renege on agreements that grant autonomy or contain other territorial power‐sharing provisions when their relative success in the civil war improves (see Fearon 2004). Because of this, the rebels do not put much faith in the government's commitment to peace talks. Mediation can help surmount these credible commitment problems in which the rebels do not expect the government to live up to provisions in peace agreements.

The Xs in Table Four (c) represent cases in which the variable has a significant negative impact on the probability of the civil war being active in the next year; the +s represent positive effects. In some cases, success or failure was perfectly predicted so that variable is dropped. In other words, a particular type of mediation by a particular actor might always be associated with a civil war being active or inactive in the subsequent year. Table Four (c) results are from a generalized estimating equation regression using controls in Tables Four (a) and Four (b).

The results in Table Four (c) also clearly show the relative effectiveness of direct mediation approaches over indirect ones, and of a focus on disputant's conflict behavior over a focus on their core incompatibility. We also found clear evidence that mediation conducted by P5 members was consistently effective in direct, and to a lesser extent, indirect mediation. We also note that neighbors using indirect mediation approaches actually had positive effects, meaning this approach increased the likelihood that the conflict would become active in the year following mediation, for both territorial wars and in democracies. This suggests that neighbors may take self‐interested roles in brokering the outcome of civil wars and that they lack credibility with combatants and their supporters.

This study makes a valuable contribution to our knowledge and understanding of mediation and its impact on civil wars. Our results suggest the complexity of diplomatic initiatives and of their effects on the dynamics of an armed conflict. Perhaps the most interesting finding from a policy standpoint is the relative effectiveness of P5 mediation. This finding suggests that the P5 countries have an obligation to mediate low‐intensity civil wars because there is a high likelihood of successfully defusing the conflicts in the short term. In the absence of such mediation, many low‐intensity wars can become more intense conflicts resulting in many more deaths.

This study addressed the role of mediator identity, mediator focus, and mediator strategy, which are variables that have not previously been systematically assessed in toto for their impact on civil war mediation. Mediator identity appears more important than the approach. As mentioned, P5 mediation has a high likelihood of success regardless of war type or focus of mediation. Mediation by neighbors or the U.N. works when using a direct approach. More of the mediation identity/type/focus combinations led to successful de‐escalation in democratic states than in territorial wars. In democracies, mediation that focused on core incompatibility showed greater promise than mediation that focused on behavior. A policy implication here is that mediation is a very promising tool in democracies. Unfortunately, many of the civil wars in democracies such as India and Spain go unmediated.

Direct mediation (versus indirect) appears to have the higher likelihood of short‐term success (see Table Four C), although this is less pronounced in democratic states. Some might suggest this finding to be obvious and perhaps even trivial. In other words, if disputants are at the point where they are talking directly, then this is a strong sign the dispute will be de‐escalating imminently. Put differently, direct mediation may be epiphenomenal to the peace process (see Fortna 2004). A prominent mediator made this very comment to one of the authors (under Chatham House Rule) in response to this direct mediation finding.¹ Portraying direct mediation as a higher form, the mediator argued that once the disputants reach the level of direct mediation, we should, of course, be more likely to see positive results.

We believe, however, that the direct mediation finding is meaningful. For starters, direct mediation is usually done with public scrutiny, and as such the parties involved are under pressure from domestic and international interests to produce results. Second, direct mediation raises audience costs for the state. When the state agrees to sit at the table with the rebels, a certain level of legitimacy is conveyed to the rebels.

This finding has policy implications as well. Mediators should pursue direct mediation whenever possible. Relatedly, researchers might well analyze whether there is a sequential relationship between indirect and direct mediation. Scholars might also further explore the dynamics of direct mediation to understand why it is more effective than the indirect strategy.

We also report a focus on behavior to be generally more successful than a focus on core incompatibilities. This could be because a focus on the incompatibility takes longer to yield results than would a discussion of, for example, a cease‐fire, and thus the effect of an incompatibility focus may be missed when focusing on short‐term effects. A worthwhile scholarly pursuit would be to test the assertion that the incompatibility focus is more suited to long‐term success.

We found that bilateral talks increase the probability of mediation in the subsequent year. Thus, it can be said that bilateral talks potentially have an indirect effect on de‐escalation. One policy implication here is that third parties should not dismiss the possibility of mediating in cases in which the disputants are initially not open to the idea. In these cases, the disputants might be willing to accept the relatively lower political costs associated with bilateral talks before gradually consenting to mediation.

Conflict duration has a consistently positive impact on the probability of the conflict being active in the subsequent year. A policy implication here would be to attempt mediation earlier in the conflict.

In closing, we can extrapolate from these findings. One promising idea is that mediation leads to greater humanitarian access (see DeRouen, Kevlihan, and Biglaiser 2013). For example, mediation can lead to cease‐fires that would potentially allow aid workers to serve areas previously inaccessible during the fighting. In this sense, humanitarian aid and mediation can work together symbiotically toward long‐term peacebuilding (see U.N. OCHA 2011). This process, however, requires extensive coordination (see U.N. OCHA 2011).