Agreement without Peace? International Mediation and Time Inconsistency Problems


  • I am indebted to Dan Reiter, Cliff Carrubba, Jeff Staton, Eric Reinhardt, Kristian Gleditsch, David Lake, Branislav Slantchev, Barbara Walter, David Meyer, Holger Schmidt, participants of the PIPES workshop at the University of Chicago, and three anonymous reviewers for helpful comments and discussion. I am especially appreciative of Nigel Lo's assistance in coding the data.

Kyle Beardsley is assistant professor of political science, Emory University, 317 Tarbutton Hall, Atlanta, GA 30322 (


Mediation has competing short- and long-term effects. In the short run, the actors are better able to identify and settle on a mutually satisfying outcome. In the long run, mediation can create artificial incentives that, as the mediator's influence wanes and the combatants' demands change, leave the actors with an agreement less durable than one that would have been achieved without mediation. This article tests the observable implications from this logic using a set of international crises from 1918 to 2001. The results reconcile findings in the previous literature that inconsistently portray the effectiveness of mediation.

A cursory glance at the diplomatic realm of international politics would suggest that mediation is an important element in the resolution of international conflict. Nobel Peace Prizes are awarded for successful efforts, and national leaders offer their services with considerable fanfare. Seasoned diplomats have contributed numerous essays and books on the subject. Yet mediation is rarely given much attention in the general literatures on international conflict. How, if at all, does mediation influence the powerful forces of international politics that lead states to clash?

This article addresses competing views and findings of international mediation as a useful conflict management vehicle. Through a novel theoretical framework that specifies how mediation interacts with the crisis bargaining environment, it argues that mediation does well to secure short-term peace but leaves much to be desired in the long run. Mediators can facilitate information exchange, use leverage to increase the perceived costs of conflict, and promise enforcement. Since mediators generally do not remain involved after a crisis, such temporary effects, however, tend to weaken as the actors' bargaining positions change and the mediator's involvement wanes over time. More importantly, mediators often promote settlements that are not self-enforcing. Mediators thus not only struggle to curtail time inconsistency problems—when today's agreements become unsatisfactory tomorrow—they may actually exacerbate them by increasing the probability that short-term arrangements will not be viable in the future.

Empirical findings using dyadic data of all international crises from 1918 to 2001 confirm that intermediaries can create artificial incentives which shape short-term arrangements that are not viable in the absence of constant third-party influence. Mediation is positively associated with short-term peace agreements but not with the prevention of crisis recurrence. Event history models that account for nonproportional hazard rates and self-selection of mediation reveal that the ability for mediation to assuage postconflict relations diminishes over time until it actually makes crises more likely to recur.1

Mediating Previous Work on Mediation

Studies of international mediation lack consensus in whether it helps secure peace. Smith and Stam (2003) have formally argued that, absent the manipulation of conflict costs and benefits, mediators have little potential to affect the resolution process. Werner and Yuen (2005) empirically find that cease-fires after third-party pressure to end war are much shorter lived than uninterrupted conflict bargaining. Gartner and Bercovitch (2006) and Bercovitch and Gartner (2006) have also found that mediation attempts in international conflicts do poorly when compared to negotiation attempts. Werner (1999b) shows, inter alia, that mediation has no impact on the duration of peace after war. These findings appear to contrast with other empirical efforts that portray mediation as a quite effective peacemaking strategy. Notable studies that have found a positive effect of mediation include Dixon (1994, 1996), Regan and Stam (2000), Walter (2002), Beardsley et al. (2006), Wilkenfeld et al. (2003), Wilkenfeld et al. (2005), Regan and Aydin (2006), Frazier and Dixon (2006), Rauchhaus (2006), and Svensson (2007b).

While the extant literature appears to be inconsistent, there are important patterns in the findings. The aforementioned empirical studies that find that mediation has a positive influence on conflict management primarily focus on short-term outcomes such as whether mediation led to an agreement or suppressed escalation patterns. This is also true of much of the vast qualitative and historical mediation literature. The findings of Werner and Yuen (2005), Werner (1999b), and Gartner and Bercovitch (2006), on the other hand, are more relevant to the stability of peace after conflict has subsided. It appears that both optimistic and pessimistic views can correctly describe mediation's effectiveness, depending on the scope of the evaluation period. The core argument here thus reconciles these disparate findings and asserts that mediation can do well in the short term but not in the long term.

To be clear, Werner and Yuen (2005), Gartner and Bercovitch (2006), and Bercovitch and Gartner (2006) do not dismiss the practice of mediation based on its record. Werner and Yuen argue, without testing, that third parties may be able to shorten conflicts and thus offset their long-term shortcomings. Gartner and Bercovitch argue, also without directly testing, that selection effects obscure the true impact of mediation, as mediation attempts are likely in more difficult situations than negotiation attempts. These propositions need to be tested. This article asserts and empirically finds that mediation does have short-term positive effects, but that its long-term ineffectiveness is the product of something much more fundamental than selection effects. Mediators simply lack the capacity to have much impact on peace after hostilities end and actually promote agreements that are more likely in the long run to falter.

On a theoretical plane, most of the mediation literature focuses solely on the mechanisms of mediation most useful for short-term crisis abatement.2 Some scholars, however, have considered how mediators might increase postconflict stability.3 That is, mediation has been seen as relevant to both conflict resolution and conflict management. While the approach here agrees that mediation's benefits should primarily be associated with short-term behavior, we would actually miss an important dynamic by not considering the long-term impact of mediation. Not only should mediation generally not be expected ex ante to contribute to long-term peace stability, much more, its short-term effects often disrupt more permanent conflict resolution processes.

The Nexus of Bargaining and Mediation

This article adopts the widely cited definition presented by Bercovitch and Houston, which states that mediation is “a reactive process of conflict management whereby parties seek the assistance of, or accept an offer of help from, an individual, group, or organization to change their behavior, settle their conflict, or resolve their problem without resorting to physical force or invoking the authority of the law” (1996, 13). This conception of mediation incorporates numerous styles ranging from mere facilitation of negotiations to more involved positive and negative inducements. Framing how mediation interacts with the negotiating environment in similar terms of the often-used bargaining model explicated by Fearon (1995) and Powell (1999) will help clarify what mediation does. The following discussion will first present the information and commitment problems that plague conflict resolution. It will then demonstrate how three mediation strategies—information provision, leveraging of conflict costs, and promises of enforcement—can ameliorate the bargaining environment, but only in the short run. Through these models, we can see why each of the three mediation strategies typically fails to secure long-term peace and can actually make durable resolution more difficult because they promote terms of agreement that are dissatisfying over time in the absence of the third party.4

Figure 1a gives a spatial representation of the overlapping bargaining space.5 In this situation, there are two actors, A and B, who are in dispute over the distribution of some good. Each gets a value of 1 at its ideal point. The actors can agree on some alternative in the issue space, or they can fight. If they fight, A's expected payoff is at point p, which is just his probability of winning multiplied by the value of winning. But A pays a cost of ca while fighting, so his expected net value of fighting is at the point pca. Similarly, B's expected net value of fighting is at p+cb, as the expected benefit of fighting is 1 −p and the cost is cb. The space between pca and p+cb represents the set of alternatives mutually preferable to conflict because they give each actor more than the expected value of fighting. Point q is A's value for the status quo, so that 1 −q is B's value. When q falls outside of the overlapping bargaining range, there is a very real risk of serious bargaining breakdown, as one of the actors is dissatisfied. When q is inside the overlapping bargaining range, then the threat of complete breakdown is much less because neither state is truly dissatisfied.

Figure 1.

Bargaining under Information and Commitment Barriers

Scholars have maintained a crucial distinction between problems of information barriers and credible commitment problems (Fearon 1995; Powell 2004a, 2004b). Information barriers are generally caused by each side's incentive to appear stronger than it really is and thus exploit the private information of its own capabilities and resolve. Commitment problems, when time inconsistency problems are expected, are situations in which the actors might have incentives to renege on an agreement in the future even though it is acceptable in the present.

Figure 1b illustrates a typical information story to bargaining failure. To capture the effect of uncertainty, let us assume that B's estimate of A's costs of conflict falls on a uniform distribution with upper bound cua and lower bound cla. It is possible for B to estimate that A is actually not dissatisfied and is only bluffing when it demands a new settlement.6 That is, if B thinks that A's costs are cua, then it will assume that A is bluffing since the status quo is actually worth more than pcua. When B holds these beliefs and rejects any movement from the status quo, A will prefer to fight and bargaining failure will occur.

Turning to commitment problems, Figure 1c depicts a generic time inconsistency problem in which the actors' expected benefits from fighting change as power and costs change over time. A's expected evaluation of conflict changes from period 1 to period 2 with some uncertainty. In period 1, A's probability of victory is p1, and the overlapping bargaining space is in between p1ca and p1+cb. In period 2, both A's and B's expected payoffs from conflict lie on a uniform distribution between (p2ca)u and (p2ca)l, and between (p2+cb)l and (p2+cb)u, respectively. The space between (p2ca)l and (p2+cb)l represents a safe area, in that any agreements reached there in period 1 will also always be acceptable in period 2. The spaces between p1ca and (p2ca)l and between p1+cb and (p2+cb)l are potential danger areas where time inconsistencies could be a problem. Here, it is possible for the actors to reach an agreement in period 1 that is not satisfactory in period 2. If this were the case, then the newly dissatisfied state will make a challenge in the second period. The stability of peace will thus be threatened in both the next period and the present, as actors will be less willing to settle when the issue is likely to remain unresolved.7

In what follows, I present three different mechanisms by which mediation interacts with the bargaining environment to build the case for why mediators are frequently able to overcome problems in the negotiation phase but can also make the problems in the implementation phase more difficult. Information provision, leveraging costs, and promises of monitoring and enforcement can all enable combatants to more quickly recognize and agree to settlements that are mutually preferable to fighting. Each of these mechanisms, especially the latter two, will also have the tendency to advance peace arrangements that are not tenable in the absence of substantial third-party influence.

Information Provision

One of the ways a mediator can help the actors bargain more efficiently is through reducing barriers to information. Recent scholarship formally considers how mediators enable the combatants' prior expectations of costs and outcomes to converge on appropriate offers and concessions (Kydd 2003; Rauchhaus 2006). As an example of informational mediation, in a 1990 crisis between India and Pakistan, the United States sent Deputy Director of the CIA, Robert Gates, along with Richard Haass (National Security Council aide for the region), to the region as mediators. These intelligence experts attempted to reduce tensions by providing Pakistan a realistic estimate of the military imbalance and warning India about Pakistan's potential for nuclear use.

Figure 2a illustrates how mediation might facilitate smoother bargaining via an informational role. The top graphic, identical to Figure 1b, presents a hypothetical situation without mediation, and the bottom one presents a situation with mediation. In both situations, A is dissatisfied because q is less than pca. With mediation, the new upper bound for B's estimation of A's costs is dua, and the new lower bound is dla. If mediation can effectively reduce the amount of uncertainty over A's costs—reducing the potential range in which it might fall—then B will always perceive that A is dissatisfied. While this does not guarantee an agreement, since sometimes A might reject any demands between pdua and pca, it makes bargaining success more likely.

Figure 2.

Mediation's Influence on the Bargaining Process

While much attention has been paid to mediation's informational role, a closer examination reveals some limitations of this mechanism. Simply, it is not clear how mediators are able to access information that the combatants themselves are not privy to. A combatant will hesitate to reveal private information to a mediator because the mediator has an incentive to make the revealed information public, thereby reducing the misinformation. In addition, few intermediaries will have superior intelligence than the combatants themselves and also be willing to share such sensitive intelligence for another state's use. Moreover, some scholars have argued that mediators will often manipulate the flow of information to maximize the probability of settlement (Princen 1992; Smith and Stam 2003; Young 1967). Recognizing the incentive of mediators to do this, the combatants might discount any information that the mediators do provide.

We should consequently expect that mediators play more of an indirect conduit of information rather than a direct conduit. Mediators can reduce the transaction costs of information exchange by being the medium of the exchange or providing a neutral forum for negotiations. Alternatively, mediators might be a witness to negotiations, preventing manipulation of what was said. The third parties can also serve as a second opinion when the state of the world is uncertain and the actors need confirmation of their own perceptions. Finally, mediators might provide political cover to actors facing audience costs for concessions (Beardsley 2007; Pruitt 1981) or who are caught in the bargainer's dilemma (Greig 2005; Ott 1972). That is, actors are often reluctant to reveal their willingness to make concessions for fear that they will be punished at home or that the other side might exploit signals of weakness. Third parties can take responsibility for any concessionary proposals or affirm their prudence and thereby facilitate the exchange of information without either side looking overly eager for peace.

An informational role, when effective, will realistically only have a short-term positive impact. The only way for information provision alone to diminish potential time inconsistency problems, as the actors' capabilities and conflict costs shift, would be to increase the probability that an agreement will be reached near the most stable region of the bargaining range, i.e., well above each actor's minimum acceptable offer. There is little reason to suspect this occurs often. Given that third parties are limited in their abilities to facilitate information exchange, as discussed above, it would ask a lot of a mediator to accurately discover the most stable agreements and get the actors to agree to them. Instead, the goal of information provision is typically to help the actors recognize some agreement that is mutually preferable to conflict in the present. The resulting agreements may or may not be mutually preferable to conflict in the future, as many mediators can, if at all, realistically only provide information about the present state of affairs. Moreover, if the mediator is primarily serving as an indirect vehicle of information or providing political cover, then the mediator can only be effective to the extent that it is engaged in the negotiations, which is likely to be temporary.

Third-party information provision could conceivably also make agreements less durable, although we should expect this negative effect to be weaker than that of the other two mediation strategies considered below. If the focus of the informative mediation is on the opponent's “red lines”—the minimum acceptable offers—then this could encourage agreements near the edge of the overlapping bargaining range. In particular, if mediation reveals information that suggests an actor had actually underestimated how much an opponent is willing to concede, then that actor will tend to revise its demands toward the opponent's red line. Returning to Figure 1c, the alternatives close to the edge of the initial overlapping bargaining space are most prone to time inconsistency problems as preferences change over time. Additionally, if mediation substitutes for direct communication, then arrangements reached without mediation might prove more durable because the actors would have been forced to develop bilateral mechanisms of learning and communication that enhance future renegotiation.

Leveraging Costs

Mediators can also use a cost-maximizing, or manipulative, strategy to help achieve an agreement when the actors cannot find a mutually acceptable agreement (Bercovitch, Anagnoson, and Wille 1991; Bercovitch and Houston 1996; Carnevale 1986; Princen 1992; Schrodt and Gerner 2004). To maximize the costs of conflict and enhance the benefits of peace, mediators with substantial resources can use tangible carrots and sticks, or they can use more intangible leverage, such as offering the promise of closer long-term ties and prestige. Carter's 1978 Camp David mediation exemplifies both tangible leverage, in the form of aid to Egypt and Israel, and intangible leverage, in the form of not abandoning the region to the Soviet sphere of influence (Princen 1992). Even mediators without great wealth or power can exert leverage on a bargaining situation, especially through both international and domestic audience costs. When Roosevelt mediated between Russia and Japan at Portsmouth in 1905, he had relatively little leverage compared to the other great powers. By securing the endorsements of Germany and France prior to making a proposal, he placed pressure on each side to not alienate the two great powers and move them into closer ties with the opposition (Princen 1992). Weak mediators can also increase the domestic audience costs for failing to reach an agreement. In 1994, Jimmy Carter, as a private citizen mediating between the United States and the DPRK, announced a deal that had not been approved by the United States on international television, thereby placing the Clinton administration in an awkward situation if it refused the deal (Wit, Poneman, and Galucci 2004).

By manipulating the opportunity costs of conflict, mediators can expand the overlapping bargaining space so that the actors have a greater chance of finding a proposal in that area. To illustrate, in Figure 2b, the bargaining space without mediation does not contain the status quo q, so A will be dissatisfied. If mediation is able to increase the opportunity costs of conflict by some amount m, then the overlapping bargaining space with mediation is between pcam and p+cb+m. With mediation, q is greater than pcam, so A would actually be satisfied in this situation and bargaining should be easier. Having mediation also increases the range of alternatives within the overlapping bargaining space.8

Leveraging costs will have the effect of making more potential agreements preferable to the immediate costs of conflict, but it will also make long-term peace less stable.9 Positive and negative inducements can actually generate a time inconsistency problem even when one did not previously exist before mediation. Werner and Yuen (2005) show through similar models that when third parties promote an artificial settlement—one that would not be mutually beneficial in the absence of a third party manipulating the conflict costs—through expanding the perceived overlapping bargaining space, that settlement may not prove stable when the third party is no longer in the picture. If the actors in Figure 2b reach an agreement between pcam and pca or between p+cb and p+cb+m, such agreements will be unstable in the future. As the mediator loses interest over time, m will effectively retreat toward 0, and there will be a higher probability that a state will become dissatisfied with the initial agreement. So, leveraging costs can generate its own sources of instability.

Promises of Monitoring and Enforcement

Mediators can try to directly confront time inconsistency problems through promises of monitoring and enforcement. For example, UN mediation efforts often naturally evolve into monitoring or peacekeeping once the conflict is over, as was the case when the UN mission to the Democratic Republic of Congo (MONUC) was involved in mediation prior to conflict termination and then in peacekeeping afterward. While monitoring and enforcement are qualitatively different from mediation, a mediator can promise to stay involved after settlement as a vehicle to encourage the settlement in the first place (Bercovitch 1997; Lake and Rothchild 1996; Touval and Zartman 1985; Young 1972; Zartman and Touval 1985). A third party that can maximize costs of renewed conflict in the future may decrease the incentive to cheat or renegotiate by either increasing the probability of cheating detection or increasing the costly consequences of cheating (Walter 2002). Figure 2c illustrates how mediation might, in theory, reduce a credible commitment problem. The top graphic is identical to that of Figure 1c. In the bottom graphic, mediation expands the period 2 costs of conflict by m, representing the higher costs of breaking a period 1 agreement when monitoring and enforcement is in place. By raising the costs of conflict in the future, the mediator widens the safe area of possible agreements that are mutually satisfying in both periods and thereby decreases the potential areas with time inconsistency problems. In the depicted example, the mediator eliminates all possible time inconsistency areas: all alternatives acceptable in period 1 are also acceptable in period 2.

More practically, a mediator's ability to have a positive long-run impact through promises of monitoring and enforcement is also limited, and potentially counterproductive, even though the target is time inconsistency. Only the most resource-abundant states and international organizations can do an effective job keeping the sides honest after a conflict. However, states typically have little incentive to pay the costs for others' security, and all third parties face budget constraints and multiple commitments. Moreover, the types of agreements that are difficult to reach without mediation but that result with third-party promises of monitoring and enforcement are precisely those facing time inconsistency problems—those arrangements outside of the safe range without mediation in Figure 2c. Once the inflated costs of future conflict, m, decrease to zero, such agreements are less tenable. So, many mediators are prone to encourage agreements that ultimately fail in the long run because of the third parties' inability to indefinitely sustain their needed security guarantees.

If a mediator's influence is prone to wane, then how do promises of monitoring and enforcement increase the likelihood of an initial agreement? Although monitoring and enforcement are not often able to decrease the probability of cheating in the relatively distant future for reasons elaborated above, states will still feel less vulnerable to cheating because they know that former mediators in the international community have an interest in intervening when one side is in peril of severe losses. Expecting third parties to assist in the most severe violations requires much less of the mediators than expecting them to credibly deter any type of future violation. By decreasing the vulnerability to cheating, but not necessarily the probability, mediators can allow the actors to seek short-term peace even if the long-term prospects are not as great. Additionally, mediators may be able to encourage outcomes with promises that are credible in the first few years after a crisis, but not much longer. Given that actors typically discount the future, they will often be willing to make the trade-off of gaining short-term peace at the expense of potential long-term instability. Perhaps counterintuitively, promises of monitoring and enforcement, like the other two mediation mechanisms, will thus have a more positive impact in the short term.

Observable Implications

Mediators should do well in securing crisis abatement and short-term agreements. The presentation of the theoretical models reveals that through facilitating information revelation, leveraging the costs of conflict, and promising future involvement, third parties can enable the actors to find an alternative mutually preferable to fighting. Even though mediation is not expected to have much of a long-term positive influence, mediators can temporarily stop the killing and escalating hostilities. I expect this effect to be manifested in an increased probability of the combatants reaching a voluntary formal agreement.

Formal agreements are costly to make and entail costs to reneging, so actors will generally only sign when they are strongly satisfied, at least in the immediate future, with the terms (Fortna 2004c). While actors can resolve their crises satisfactorily in other ways besides a formal agreement, from a purely probabilistic standpoint, the likelihood of crisis actors being willing to sign an agreement should increase when the bargaining process is more efficient. Other crisis outcomes such as perpetual stalemate or complete capitulation will become less likely when the actors more easily recognize a set of mutually acceptable alternatives. If the theoretical framework is valid, we should observe crisis actors as better able to identify and commit to a negotiated settlement when mediators are present to decrease misinformation, increase the costs of fighting, or decrease the vulnerability to cheating.

  • H1: Crises with mediation will have a higher probability of achieving a formal agreement than crises without mediation.

In terms of long-term stability, however, the discussion above posits that mediators are not expected to be sufficiently good monitors, enforcers, or information providers to have much positive impact on time inconsistency problems. Third-party influence can also create new time inconsistency problems or make preexisting ones worse. Expanding conflict costs—in the present through leverage or in the future through promises of future involvement—encourages agreements that are not consistent with the actors' long-term interests; moreover, information provision may only highlight agreements near the boundaries of what is momentarily acceptable and prevent development of bilateral mechanisms for renegotiation. The benefit of mediation to crisis recurrence should thus diminish over time. Mediated crises should on average be less likely to immediately recur because it takes time for a mediator's influence to wane and for the preferences of the combatants to shift. Over time, the actors will be increasingly likely to prefer reneging on the previous agreement that becomes dissatisfying because it is not reflective of their future preferences in the absence of third-party influence. In the long run, the involvement of the mediator will actually have the unintended but pernicious effect of increasing the likelihood of another crisis.

  • H2: Mediators will be more likely to prevent immediate crisis recurrence but less likely to prevent crisis recurrence further in the future.


The universe of cases in this analysis consists of international crises from the International Crisis Behavior (ICB) data project. Brecher and Wilkenfeld (2000) define crises as having three necessary conditions that together are sufficient. First, an international crisis requires that a state perceives another state as a threat to its basic values. Second, there needs to be a heightened perception of the threat of military hostilities. Third, a crisis requires that the states face a finite time horizon for decision making. The advantage of using crises over other measures of conflict is that crises do not imply either use of force, or even military mobilization. As a result, the mediation efforts considered in this article occur in a broad array of contexts in which a potential exists for violence, but violence does not necessarily occur.

The ICB version 6 data include 443 crises from 1918 to 2003, but data are only complete for the other variables in this analysis until 2001. The unit of analysis is the nondirected dyad, as I am interested in assessing if a formal agreement ends the crisis for a dyad, as well as how soon, if ever, the same dyad returns to a crisis situation. Hewitt (2003) defines the ICB crisis dyads used here. I aggregate the intrawar crises in the ICB data so that the time period begins on the first day of the first crisis and ends on the last day of the last crisis. This is done since, except for the last one, each intrawar crisis by definition must recur.10


In the models of whether a crisis ends in a formal agreement, probit and ordered probit specifications are used with the crisis dyad as the unit of analysis. For the analyses of peace spells until the next crisis, a time-varying survival time, or event history, analysis is used. The dyads enter the data on the first day of their first crisis and exit the last day possible for the analysis—either the last day that the dyad exists or on 31 December 2001. The dyads are observed at the start of each year and at the onset of a new crisis, with time since the previous crisis measured in days. This leaves 378 crisis dyads and 349 recurrences. Goertz, Jones, and Diehl (2005) note that event history analyses of international conflicts often use too long a time frame in which recurrence can occur. For example, in 1919, Britain and Afghanistan engaged in a crisis during the Third Afghan War. These two countries did not participate in opposite ends of a crisis again until 2001, after an 82-year pause. Conventional event history models of recurring conflict would treat the 2001 crisis as a repeat of the 1919 one, even though the two crises are so far removed and have nothing to do with each other. As a result, I follow Diehl, Reifschneider, and Hensel (1996) and use a 10-year temporal domain in which failure can occur. That is, I omit from the analysis all observations between the 10-year cutoff and the next crisis, at which point the dyad reenters the data.

The event history models primarily rely on the Cox specification because it does not require strong assumptions about the shape of the hazard function. Hypothesis 2 expects that mediation's impact on peace spells will start strongly but diminish over time. In essence, the hypothesis suggests that the hazard rate is not proportional over time and in fact depends on the interaction of time and mediation. Box-Steffensmeier, Reiter, and Zorn (2003) demonstrate that by including interaction terms of time and the independent variables, analysts can easily relax the proportional hazards assumption and directly model such expectations. In this regard, I interact all the variables with the time since the previous crisis and include the interaction terms as independent variables.

When data like these have units (dyads) that can fail repeatedly, Box-Steffensmeier and Jones (2004) advocate using a stratified event history model in which the repeated events are treated as conditional on there being a previous failure. I thus define the strata as how many crises a dyad has been in since entering the data set and run the Cox models using the conditional gap-time specification.11

Selection Effects Selection effects are likely to confound accurate analysis because mediation is not randomly chosen. Studies such as those of Young (1967), Zartman (1985), and Rubin (1991) have focused on “ripe” moments of conflict as crucial in explaining when mediation is likely to succeed. That is, there are crucial moments during the life span of a conflict in which the actors are no longer able to bear the costs of conflict, as during a mutually hurting stalemate, and are able to recognize the feasibility of settlement. Such moments provide the third parties with a critical opportunity to contribute to conflict abatement. Rubin (1991) stresses that mediation is more likely to occur in these ripe moments. Terris and Maoz (2005) similarly find that mediation is more likely to occur in more versatile conflicts that can be easily transformed into games of cooperation. Greig and Diehl (2006), building on Greig (2005), have discovered that mediation is much more likely between enduring rivals after the relationship has “softened up.” Yet other notable work has posited and found confirming empirical evidence that mediation actually occurs in more intractable conflicts. Hensel (2001) finds that nonbinding third-party management attempts of North American territorial claims tend to occur when there is a history of unsuccessful negotiations and recent conflict. More recently, Gartner and Bercovitch (2006) and Bercovitch and Gartner (2006) contend that mediators tend to intervene in the conflicts that are most prone to elude long-term resolution.

Both lines of argument posit that the combatants and third parties have preferences for mediation that are tied to the expected outcomes. If this is the case, then variables likely to be unobservable may be driving both the occurrence of mediation and its outcomes. For example, imagine that both views of mediation selection are valid, such that mediation occurs when it is ripe for short-term resolution but also more difficult to achieve a stable peace.12 Analyses that do not take into account the selection process will tend to exaggerate mediation's short-run impact, but understate mediation's role in stabilizing long-term peace.

A bivariate probit model can be used to account for the endogenous choice of mediation. The specification of the model, from Greene (2003, 710) and Maddala (1983, 122), involves two processes with dichotomous outcomes. In the first equation, the outcome is whether mediation occurs. The second equation of crisis outcomes is then estimated simultaneously while taking into account the correlation in the equations' error processes. Greene (2003) reports that no complications will arise due to mediation appearing as an independent variable in the second equation, making it recursive. After converting the data to dyad-year, or discrete-time, format, the bivariate probit model can be modified for event history analysis by including functions of time as covariates. In the outcome equation, the dependent variable is simply whether a recurrence in time t occurred. Including the peace-spell length, its square and cube, and the interactions between peace-spell length and the independent variables allows the baseline hazard rates to vary over time, thereby enabling the model to account for duration dependence.13


The principal dependent variable used to test Hypothesis 1 is a dichotomous measure of whether or not a voluntary formal agreement ended the crisis. The ICB data code this as true if the crisis ended with a treaty, armistice, or cease-fire agreement. These formal agreements are distinct from formal terms that are imposed on a defeated state, which is important because the dependent variable should be a measure of short-term bargaining success, and imposed agreements would indicate complete bargaining failure. Using the ICB case summaries, as well as data from Fortna (2004c) and Lo, Hashimoto, and Reiter (forthcoming), I coded this variable at the dyadic level.14

I also disaggregated formal agreements for further analysis, as it is important to know if mediation only leads to weak cease-fires instead of more robust agreements, thereby presenting an alternative explanation for mediation's record in sustaining long-term peace. Four types of short-term outcomes are created: no formal agreement, nonrobust agreements, robust agreements, and strongly robust agreements. In forming this typology, I consider whether the formal agreements contain provisions for peacekeeping, confidence-building measures, arms control, or further conflict resolution of the most contentious issues.15 Nonrobust formal agreements are those that do not contain these additional protocols and are merely formalized intents to withdraw from combat areas. At the other extreme, I coded for those agreements that contained multiple such provisions for peacebuilding initiatives. For example, the 2000 agreement that terminated the war between Ethiopia and Eritrea was signed by UN Secretary General Kofi Annan and U.S. Secretary of State Madeleine Albright, and it called for 4,200 UN peacekeepers and postconflict progress toward such arrangements as a demilitarized border, prisoner exchange, and compensation. Of the 171 formal agreements in the data, I code 19 as nonrobust agreements, 106 as robust, and 46 as strongly robust.

For Hypothesis 2, the duration between crises is assessed as the dependent variable in the event history analyses. Since low-level crises—ones without any use of force—might not reflect a fundamental return to conflict, I also include additional analyses to see if mediation increases the spells of peace from the end of a crisis to the beginning of a crisis with violent clashes. To test this, an event history analysis is used where a recurrence event only occurs when a crisis starts in the next ten years and involves acts of violence as recorded in the ICB data.

The key independent variable is whether mediation occurred. While the ICB data have a variable that measures the occurrence of mediation, this variable is only measured at the crisis level. Since multiple crisis dyads can be part of a crisis with mediation, but mediation does not necessarily occur in every dyad, the existing mediation variable had to be recoded. Using the same definition of mediation used in the ICB data, a new dyadic mediation variable was coded based on the ICB case summaries and historical accounts. Mediation only occurred in a dyad if there is documented evidence that a third party participated in negotiations with both sides of the dyad during the course of the crisis. Of the 689 crisis dyads in the data, 161, just over 23%, experienced mediation.

The models include as controls contextual variables similar to those that Fortna (2004c) finds to be related to the strength of formal agreements and the duration of peace after formal agreements. Each of these variables is also plausibly related to mediation incidence, as they shape the bargaining environment and thus the expectations over the costs and benefits of mediation. Starting with the models of formal agreements, the number of previous crises in the past 10 years, as a measure of the conflict history of a crisis dyad, is a crucial control variable. This provides information about the rivalries, which Goertz, Jones, and Diehl (2005) stress as important to conflict behavior, and Greig (2005) finds that rivalry history matters to mediation occurrence. Second, violence level captures the severity and costs of conflict, and it is coded from the existing ICB data as a 4-point variable from no violence to minor clashes to major clashes to full-scale war. Third, the crisis duration can increase the opportunity for mediation to occur as well as indicate the long-term prospects for peace. The natural log of the length of each crisis (in days) is taken. Fourth, a joint democracy variable captures the role of domestic institutions and is coded as true if both sides of the conflict dyad are a democracy—both states have at least a 6 value on the 20-point Polity IV aggregate index variable.16 Fifth, the nature of the crisis outcome—victory for one side or a tie—should provide a key indicator of the willingness of the actors to bargain and the extent to which the conflict relationship is clarified by fighting. Victory is coded in the ICB data as whether a state achieved all of its crisis goals.

In addition to these control variables, three more are used for the models of peace duration, continuing to draw from Fortna (2004c). The distance between the combatants shapes the opportunity to have contested issues in the future. The natural log of the distance between the capital cities is used for this variable.17 The S-score indicator of the similarity of the combatants' alliance portfolios additionally controls for affinity among the actors' foreign policies that could lead to both stable peace and willingness to engage in multilateral conflict resolution. This variable is taken directly from the Correlates of War data, using the weighted global indicator. Finally, the change in capability share from the previous crisis between the same members of the dyad can influence both credible commitment problems and uncertainty about the current distribution of power (Powell 1999, 2004b; Werner 1999a). The National Material Capabilities data include a composite index of capabilities (CINC) for each state-year, calculated from six measures: military expenditures, total population, urban population, iron and steel consumption, primary energy consumption, and military personnel (Singer, Bremer, and Stuckey 1972). The highest CINC score in the dyad is divided by the sum of the two CINC scores to produce an indicator of the power share that the strongest actor has. Change is then measured as the difference between the present capability share and the capability share at the termination of the previous crisis.18

When using the bivariate probit models, I draw from Beardsley (2006) to specify the independent variables in the mediation equation. I include three variables that are used in the outcome equations: violence level, crisis duration, and the presence of a democratic dyad because these are key variables that capture the opportunity and need for mediation. Independent variables should also be included in the selection equation that are not in the outcome equation (Maddala 1983, 122). Capability share, which was not found to influence the outcome processes, is thus included because power relationships are frequently considered as important to mediation dynamics.19 In addition, the presence of an ethnic component in the crisis, as coded in the ICB data, accounts for the salience of the crisis domestically and thus captures the pressure on the states to reach more favorable terms. I also include three variables that are related to the international community's incentives to push for mediation. The presence of a defense pact within a crisis dyad, as coded using the Correlates of War alliance data, entails that other alliance members will have an incentive to press for peace in order to maintain the integrity of the alliance. Geostrategic salience also captures the interest that third parties might have in shaping a crisis. It is measured from an existing ICB variable that is a 4-point measure of the extent to which the crisis is systemically relevant, with lower values indicating that the crisis only has ramifications for the local level. Finally, Beardsley et al. (2006) find that the percentage of mediated crises has fluctuated over time and is especially increasing after the Cold War, so I include the ICB 5-point period variable that corresponds to shifts in the international system polarity.


Simple bivariate assessments of mediation's association with formal agreements and recurrence confirms that mediation has a positive short-term impact but is ineffective in the long run. While only 14% of the unmediated crisis dyads experienced a formal agreement, 45% of the mediated ones did. When looking at mediation's role in preventing crisis recurrence, it is striking how similar the mediated and unmediated cases are. Forty-nine percent of the unmediated crisis dyads recur, and 49% of the mediated crises recur.

Like the bivariate analysis, the multiple regression models reveal that mediation has a strong relationship with achieving a formal agreement. Model 1 in Table 1 shows the impact of mediation in a probit model, while Model 2 presents the bivariate probit results.20 Since the estimate for ρ is statistically insignificant—there is no evidence of correlation in the disturbances between the selection and outcome equations—regular probit should be unbiased. The positive relationship between mediation and a formal termination of the crisis remains robust anyway. Substantively, using Model 1 with all the other variables set to their medians, the expected probability of a formal agreement, generated using Clarify (King, Tomz, and Wittenberg 2000), is 12% without mediation and 38% with mediation.

Table 1.  Models of Formal Agreements
IVFormal AgreementRobust AgreementAgreement Strength
(1: Probit)(2: Bivariate Probit)(3: Probit)(4: Ordered Probit)
  1. *p ≤ 0.05 in a one-tail test. Standard errors in parentheses.

Outcome Equation    
  Num. Prev. Crises last 10 yrs.0.0220.0250.057−0.004
  Violence Level0.0160.0140.0270.067
  ln(Crisis Duration)0.304*0.291*0.359*0.380*
  Joint Democracy−0.579*−0.586*−0.712*−0.439
  Cut 1   2.663
  Cut 2   2.778
  Cut 3   3.753
Mediation Selection Equation
  Violence Level 0.127  
  ln(Crisis Duration) 0.186*  
  Joint Democracy 0.054  
  Capability Share −1.200*  
  Defense Pact 0.602*  
  Ethnic Component 0.956*  
  Geostrategic Salience −0.232*  
  Period 0.267*  
  Constant −1.780*  
 ρ −0.201  

Models 3 and 4 unpack the formal agreements. Model 3 uses robust agreements, described above, as the dependent variable. Even with the higher standard for agreement, mediation has a positive and statistically significant effect. Model 4 uses ordered probit, where the dependent variable is a 4-point variable of agreement strength from no agreement to nonrobust agreement to robust agreement to highly robust agreement. The results demonstrate that mediated crises have a higher likelihood of ending in a strong agreement. This further supports the argument that mediation does well in securing short-term agreements and does not lend credence to the suspicion that mediation merely contributes to weak agreements that are by nature prone to break down.

Table 2 presents a sequence of event history models. The coefficients are given as listed, such that one would exponentiate them to recover hazard ratios. Values above zero indicate that the corresponding variable shortens the peace spells between crises. Models 5 through 7 are event- history models with the proportional hazards assumption relaxed by including the interaction terms of the independent variables and elapsed time. The interaction between joint democracy and elapsed time was found to be statistically insignificant and is not included in the displayed model.

Table 2.  Event History Models of Crisis Recurrence
IVTime Until CrisisTime Until Major Clashes
(5: Cox)(6: Bivariate Probit)(7: Cox)
  1. *p ≤ 0.05 in a one-tail test. Standard errors in parentheses.

Outcome Equation   
  Mediation * Time0.00035*0.00018*0.00038*
  Num. Prev. Crises last 10 yrs.0.981*0.180*0.550*
  Violence Level0.327*0.0240.374*
  ln(Crisis Duration)0.213*−0.0430.243*
  Joint Democracy−0.402−0.033−1.026*
  Change in Capability Share−2.305−0.247−2.685
  Prev. Crises * Time−0.0011*−0.00005-0.00026*
  Violence * Time−0.00036*−0.00003−0.00040*
  ln(Crisis Duration) * Time−0.00030*−0.00001−0.00028*
  ln(Distance) * Time−0.00009*1.91 × 10−6−0.00011*
(0.00003)(1.31 × 10−5)(0.00003)
  S-score * Time−0.00082*0.00014−0.0010*
  Change in Capability * Time0.0043*0.0018*0.0042*
  Victory * Time−0.00078*−8.61 × 10−7−0.00076*
(0.00018)(7.75 × 10−5)(0.00020)
  Time since Previous Crisis 0.0013* 
  Time2 −9.30 × 10−7* 
(2.03 × 10−7) 
  Time3 1.54 × 10−10* 
(4.03 × 10−11) 
  Constant −0.993* 
Mediation Selection Equation
  Violence Level 0.049* 
  ln(Crisis Duration) 0.182* 
  Joint Democracy 0.274* 
  Capability Share −1.233* 
  Defense Pact 0.571* 
  Ethnic Component 0.828* 
  Geostrategic Salience −0.291* 
  Period 0.241* 
  Constant −1.364* 
 ρ 0.074 

Model 5, using a Cox specification, reveals that the impact of mediation on peace spells diminishes with time. At low levels of time since the previous crisis, mediation has a statistically significant and negative effect on the hazard rate—it prevents immediate crisis recurrence. At high levels of elapsed time, mediation does less well, even to the point in which mediated crises have a higher likelihood of recurrence than unmediated ones. Accounting for self-selection into mediation, the bivariate probit in Model 6 tells a similar story, and the statistically insignificant ρ estimate, which captures the correlation between the two equations, decreases concern for endogeneity bias in Model 5.

Figure 3 illustrates this effect more clearly. This figure results from a probit model on the dyad-year data that includes the elapsed time and its interaction with mediation.21 To adequately account for duration dependence, I follow Carter and Signorino (2007) and include the squared and cubed elapsed time as covariates as well.22 From the figure, dyads less than four years removed from a crisis are less likely to return to crisis when they had mediation. After about four years, those dyads that had mediation become more prone to recurrence than dyads enjoying an unmediated peace. It should be noted that the difference between the trends is statistically significant (p < 0.05, one-tail test) before about two years and after about six years after a crisis.

Figure 3.

The Impact of Mediation and Time on Recurrence

Model 7 tests to see if the same relationships hold if we consider peace spells to end only after violence erupts between the actors. Low-level crisis recurrence may not indicate much about the durability of peace if those crises do not escalate. Even with a higher threshold of conflict, namely the outbreak of military clashes, the results still reveal that mediation only does well to prevent future violence in the short term and actually increases the prospects of military clashes in the long term.


The results confirm both hypotheses. Mediation has a strong short-term impact but can often inhibit long-term peace. More specifically, crisis dyads with mediation are much more likely to reach a formal agreement that terminates the crisis, and they are less likely to experience a recurrence of crisis within the first few years after a crisis. Yet mediators tend to only produce a pause before the dyad eventually becomes even more prone to recurrence than if it had not had mediation, as expected in Hypothesis 2. Even when accounting for selection effects, mediation increases the long-term probability of crisis recurrence. Each of the mediation strategies, especially the leveraging of costs and the promising of monitoring and enforcement, creates incentives for the actors to arrive at solutions that are not durable. The recipients of mediation will tend to become dissatisfied with the settlements as their capabilities and resolve change and as the artificial incentives from the mediator diminish. In this vein, Zartman and Touval write, “Left alone to their own devices, parties may fall out of an agreement just as it is being made or implemented” (1985, 44).

Despite the unintended negative long-term effects, mediation still has many virtues. Mediation does reduce the chance that a peace will immediately fail, and that effect on average lasts a few years after a crisis. Coupled with the findings that mediation does well to enable the actors to reach a formal agreement, there are clear benefits to mediation. Through revealing alternatives that are mutually preferable to fighting and decreasing the actors' perceived vulnerability to future cheating, mediators allow the actors to identify settlements with which they are comfortable enough to formalize. Many of these crises are devastating and any lulls can be a positive development to the reduction of suffering and instability (Bercovitch and Gartner 2006). In light of the long-term risks of mediation, it is these short-term benefits, and the tendency for actors to discount future streams of utility when there is greater short-term need, that explain why actors ever turn to mediation. Moreover, the results should not be interpreted as suggesting that mediated crises are unconditionally more likely to recur. Recall that unmediated peace arrangements are much more likely to fail in the first few years after a crisis. The key point is that mediation does very well in sustaining short-term peace at the expense of some potential for extremely durable peace. Uncovering the nature of the trade-off between short-term peace and long-term stability and understanding why some actors are more willing to make the trade-off than others is a fruitful avenue for future theoretical and empirical research.

There are at least three implications that this article's findings suggest for enhancing the impact of mediation on conflict resolution. First, third parties need to be more aware that the incentives they create while mediating may fuel a time inconsistency problem. While the incentives can be very direct, as were Holbrook's threats at Dayton, it is likely that third parties might create additional artificial incentives beyond mere offers of aid or threats of sanctions. As Princen (1992) argues, principal mediators can alter the costs and benefits to agreement just by being part of the negotiations and making bilateral exchanges multilateral. When powerful states like the United States mediate, they need not make explicit threats or promises of action. The mere involvement of the third party changes, if only temporarily, the incentives of the combatants, who might have interests in wooing the allegiance and protection of a great power. Powerful third parties that may not intentionally alter the costs of conflict thus must be cognizant of how their involvement may create temporary dynamics that, if used to inform the bargained outcome, will lead to an untenable peace after the third party is out of the picture.

A second insight is that the inability for mediation to stay conflict in the long run is not necessarily a failure of mediation per se, but a failure of the postmediation environment. If the third parties remain involved in encouraging cooperation within a dyad long after they mediate, the potential for time inconsistency problems should decrease. A case could be made for more delegation in peacekeeping activities. If more resolved crises could be handed over to well-funded global or regional peacekeepers with the sole task of increasing the costs for defection, the initial benefits of mediation would not wane so quickly. Such an approach speaks to the interdependence between the peacemakers and the peacekeepers. Recent scholarship that has demonstrated the utility of peacekeeping in sustaining postconflict peace confirms that conflicts which transition from mediation to peacekeeping would be better situated to not give up the gains toward peace accomplished during the mediation efforts (Doyle and Sambanis 2006; Fortna 2004a, 2004b; Walter 2002).

Finally, time inconsistency problems will not be so destabilizing if there is ample opportunity for renegotiation (Werner 1999b). More mediation during noncrisis situations is thus likely to improve the overall ability for peace to stabilize. An obvious example is the Camp David mediation, which occurred five years after the Yom Kippur War and during a period of relatively low hostilities between the Arabs and Israelis. Such mediation can be preventive, as it smooths the renegotiation process. Most mediation events, however, are reactive, as they wait for conflict to escalate.23 If third parties and combatants were more proactive, the short-term benefits of mediation could be better utilized and converted into more long-term stability.

One extension of the argument, left to future research, is how mediation performs in intrastate conflicts. While the analysis here is limited to interstate crises, the theoretical expectations of what mediators are and are not able to accomplish should transfer across bargaining contexts. One might expect that mediators would have even greater difficulty in producing durable agreements, as Walter (2002) argues that credible commitment problems—closely related to time inconsistency problems—are more pronounced in intrastate bargaining. The trade-off between short-term needs for spells of peace and long-term conflict resolution thus is likely to be even starker in mediated civil wars.


  • 1

    It might be argued that the changing contextual environment after crisis termination makes it difficult to observe mediation's downstream consequences. However, the use of time-varying covariates in the empirical analysis enables one to sort out the effect of mediation from the effects of changing contexts.

  • 2

    For example, the following studies are empirical analyses of short-term conflict or crisis management: Bercovitch and Gartner (2006); Bercovitch and Langley (1993); Bercovitch, Anagnoson, and Wille (1991); Dixon (1994, 1996); Regan and Aydin (2006); Regan and Stam (2000); Rauchhaus (2006); Svensson (2007a, 2007b); Wilkenfeld et al. (2003). See also Kleiboer (1996).

  • 3

    See especially Beardsley et al. (2006); Gartner and Bercovitch (2006); Greig (2001); Kriesberg (1996); Quinn et al. (2006); Touval and Zartman (1985); Werner and Yuen (2005); Wilkenfeld et al. (2005); Zartman and Touval (1985).

  • 4

    By considering how each mediation strategy can lead to agreements that are not durable over time, this article expands on the “unnatural agreement” argument presented by Werner and Yuen (2005), who only focus on how third parties can use leverage to force an agreement.

  • 5

    The specific placement of the points is arbitrary and a heuristic device. What is important is the positioning of the points relative to each other.

  • 6

    Note that the models here are not intended to illustrate situations in which bargaining failure always occurs but rather the conditions under which bargaining failure is possible.

  • 7

    Note that the time inconsistency problem would be even starker if p2 moved so far from p1 that there was no overlap between the initial and subsequent bargaining ranges, as in Powell (1999, 129). Bargaining in such a situation would be especially difficult.

  • 8

    Note that while it is possible for an actor to get less from mediation than what could be achieved without mediation, this does not mean that mediation of this sort is irrational for the actors to accept. Although it is possible for each actor to get an inferior bargain with mediation, it is also possible for each actor to get more than anything possible without mediation. Moreover, it is unlikely for leveraging costs to be the only tactic that a mediator employs, so combatants might seek mediation for other benefits, such as the information and monitoring roles.

  • 9

    Beardsley et al. (2006) present some empirical support for this expectation, as they find that manipulative mediation is a superior tactic in reaching short-term agreements but inferior in achieving long-term tension reduction.

  • 10

    I also omit six crisis dyads that begin before another crisis ends because this would lead to periods of time that would appear multiple times in the data.

  • 11

    Relatively few dyads experienced more than five repeated events, so, to ensure sufficient observations in each stratum, I set the stratification variable equal to five for the cases that have repeated more than five times. One might suspect additional unobserved heterogeneity from unique characteristics of each dyad. In such a situation, Box-Steffensmeier and Jones (2004) and Box-Steffensmeier, De Boef, and Joyce (2007) recommend using a frailty model. When a frailty model is estimated, however, the observed variance of the random effect, which reveals the potential for unobserved heterogeneity, is not statistically significant.

  • 12

    This argument is fairly consistent with that of Greig and Diehl (2006).

  • 13

    See also Carter and Signorino (2007), Beck, Katz, and Tucker (1998) and Box-Steffensmeier and Jones (2004, 89–93).

  • 14

    The formal agreements variable is not included in the event history models as an independent variable because it would be an intervening variable—not a confounding variable—in an analysis of the relationship between mediation and the duration of peace.

  • 15

    These categories simplify similar indicators of agreement strength that Fortna (2004c) uses. From her measures, I collapse the different types of peacekeeping into one category, I consider demilitarized zones and internal control of rogue activity as part of arms control, and I do not take into account the length of the formal agreements (longer agreements do not necessarily indicate stronger ones).

  • 16

    This variable does not appear in Fortna (2004c), but it is included because of the oft-studied role of domestic institutions in shaping conflict propensities and the bargaining environment.

  • 17

    Contiguous countries receive a distance of 0, and 1 is added to all distances before taking the natural log.

  • 18

    This variable is set to zero when there was not a previous crisis.

  • 19

    See Quinn et al. (2006).

  • 20

    The reported standard errors are robust with clustering on each crisis.

  • 21

    I use the probit model to facilitate manipulation of the interactive term for the graphical display. Box-Steffensmeier and Jones (2004) similarly advocate using models such as probit on grouped duration data when substantive interpretations of duration dependency are required.

  • 22

    This approach allows for more flexibility in how the baseline hazard rates vary and greater ability to track how recurrence proclivity changes with the elapsed time.

  • 23

    See Bercovitch (1996).