Democratic Third Parties, Conflict Intensity, and International Mediation Tracking

International mediation is commonly defined as “a process of conflict management, related to but distinct from the parties’ own negotiations, where those in conflict seek the assistance of, or accept an offer of help from, an outsider (whether an individual, an organization, a group, or a state) to change their perceptions or behavior, and to do so without resorting to physical force or invoking the authority of law” (Bercovitch 1997: 130). Its onset is driven by demand and supply‐side incentives in that both the belligerents and the third party must agree on an intervention (e.g., Greig 2005; Beardsley and Greig 2009; Beardsley 2011; Crescenzi, Kadera, and Mitchell 2011). It is well established that mediation is more likely to occur in conflicts that are especially difficult (e.g., Bercovitch and Jackson 2001; Greig 2005; Greig and Regan 2008; Beardsley and Greig 2009; Melin and Svensson 2009; Beardsley 2010; DeRouen Jr, Bercovitch, and Pospieszna 2011; Hellman 2012; Wallensteen and Svensson 2014; Bakaki, Böhmelt, and Bove 2016; Böhmelt 2016; Keels and Greig 2019; Lundgren and Svensson 2020). A key reason for this is that it is precisely these cases that need third‐party assistance to reach a peaceful settlement (see Mehrl and Böhmelt 2021). The “easier,”—i.e., less intense—disputes are generally less likely to attract mediation as belligerents can solve their differences more often by themselves and the involvement of the international community is neither desired nor offered.

Another key finding in the literature is that democratic third parties are more likely than nondemocracies to mediate (e.g., Crescenzi et al. 2011; Hellman 2012; Beardsley and Lo 2013; Wallensteen and Svensson 2014; Yazici 2020). Democracies are both more likely to offer their services and to be accepted as interveners by the antagonists. As governments familiar with the norms of nonviolent conflict resolution and compromise from political processes at the domestic level, democracies spread these ideas internationally when mediating conflicts (e.g., Dixon 1993, 1994; Mitchell 2002; Andersson 2006). Similarly, democracies are more credible than nondemocracies when serving as mediators due to the transparency of their domestic institutions (Crescenzi et al. 2011). It is plausible that a synergetic effect of these two factors—high‐intensity conflict and a democratic third party—is that democracies will be even more likely to mediate when encountering a more difficult conflict.

Counterintuitively, however, this article argues that it is precisely the combination of third parties’ democratic regime type and high‐conflict intensity that lowers the likelihood of international mediation. Although the more challenging disputes are, in fact, the ones most in need of third‐party assistance, they are also more likely to fail. This imposes costs not only on the fighting parties, but also on the mediator. In the words of Beardsley and Lo:

Hence, there can be significant costs stemming from mediation and domestic audiences may punish their leaders if peace interventions are ineffective. Democracies suffer more from audience costs than do autocratic leaders (Bueno de Mesquita et al. 2005; Tomz 2007, 2008, 2012; Tomz and Weeks 2013, 2020; Chiba, Johnson, and Leeds 2015) and thus are more reluctant to mediate the most difficult disputes. As a result, I contend that democracies will strategically choose to mediate easier disputes, with their higher likelihood of success and lower risk of failure.

To illustrate this argument and especially the potential costs stemming from mediation, consider the involvement of the US in the Middle East peace process, which eventually proved to be quite costly politically. Beardsley and Lo (2013: 80) report that President Carter “felt that his involvement in the Middle East peace process actually hurt him politically because of opposition to how strongly he pushed Israel and because the implementation of the Camp David Accords was more difficult and incomplete than originally hoped.” Likewise, James Baker, the former US Secretary of State, had to conduct four rounds of shuttle diplomacy in 1991 only to get actors to agree to the Madrid Conference—which was likely hard to sell as “value for money” to the domestic audience (Beardsley 2011: 22). Or consider the civil war in Burundi (1993–2005). While other aspects certainly played a role, it is striking that Western democracies refrained from offering mediation at all in this rather intense dispute (Greig and Regan 2008). Beardsley (2011) showed how democratic governments may be quite reluctant to pursue mediation if an agreement is unpopular “at home.” And Norway became increasingly concerned about its involvement in the Sri Lankan peace process in the post‐2006 phase. After initial mediation efforts failed, a full‐scale war broke out in 2006, and the prospects of achieving any solution became rather low (Sørbø et al. 2011). The Norwegian government was at least partly taking into account potential damage to its reputation as a “persistent and patient mediator” (Sørbø et al. 2011: 62) if there was continued intervention, but conflict worsened. My argument helps explain cases like these and, generally, democracies’ mediation of less challenging disputes instead of the more difficult conflicts.

The empirical analysis focuses on international mediation in civil conflicts from 1946 to 2011 using the Civil Wars Mediation data set (DeRouen Jr, Bercovitch, and Pospieszna 2011). While I find evidence for a positive effect individually of either third‐party democracy or conflict intensity on the probability of mediation, democracies are significantly more likely to mediate civil conflicts only when casualties are low. When battle‐related deaths exceed 1,000 or so fatalities, democracies are no more or less likely to mediate a dispute than their nondemocratic counterparts. Hence, the positive effect stemming from third parties’ democratic regime type disappears. Several robustness checks further increase confidence in the main result, including the replication of an analysis of interstate conflicts by Crescenzi et al. (2011).

This research sheds new light on the persistent selection effects surrounding international mediation (see, e.g., Gartner and Bercovitch 2006; Beber 2012; Melin, Gartner, and Bercovitch 2013) and provides important insights into the very effectiveness of mediation (see, e.g., Beardsley 2008; Böhmelt ). Democracies are quite strategic about the conflicts in which they choose to involve themselves, avoiding the more difficult ones. This mirrors democracies’ foreign policy behavior more generally (see, e.g., Reiter and Stam 2002; Beardsley and Lo 2013; Chiba, Johnson, and Leeds 2015; Böhmelt and Butkute 2018). Therefore, it is probably an overstatement to say that democracies are the most effective mediators (Dixon 1993, 1994)—they merely select the “easier” cases. My findings also inform the work on reputation in conflict management (e.g., Keels and Greig 2019) and, potentially, peacekeeping (e.g., Andersson 2006). The article concludes with a discussion of policy implications and avenues for future research.

When does mediation occur? The general framework for explaining the onset of mediation focuses on a demand by the belligerents and the supply by at least one third party (e.g., Greig 2005; Beardsley and Greig 2009; Beardsley 2011; Crescenzi et al. 2011). In addition, for a mediation process to begin, the parties must have an interest in intervention and the benefits of the mediation must outweigh the costs of continued fighting. Indeed, mediation is costly—both for the belligerents who must agree on some outside intervention that cuts into their sovereignty and for the mediator. For the latter, third parties face physical, political, and reputational costs, particularly if the mediation is ultimately not successful (Greig 2005; Beardsley 2010; Beardsley and Lo 2013; Chatagnier 2019). However, mediation also has benefits, which increases the chances that third parties will get involved. As discussed in Beardsley and Lo (2013), third parties directly benefit from mediating when they are able to reduce negative externalities and obtain reputational gains, when there are humanitarian benefits, and when they have a direct stake in a conflict. Greig (2005) argues that in light of this, the most important influences behind demand and supply are the likelihood of mediation success, the characteristics of the disputants and the conflict, the previous conflict management history, and outside threats. This article focuses on two factors related to the likelihood of mediation success and the conflict characteristics: the regime type of the potential mediator and the intensity of a dispute.

First, it is well established that when conflicts are mediated, they are more likely to be mediated by democracies (e.g., Crescenzi et al. 2011; Hellman 2012; Beardsley and Lo 2013; Wallensteen and Svensson 2014; Yazici 2020). One reason for this is that democracies follow certain practices and norms in their domestic‐level political decision‐making processes and are imbued with the values of compromise and nonviolent conflict resolution (Dixon 1993, 1994). When democracies engage in international mediation, they apply these norms and values to the intervention, increasing the potential for success (Dixon 1993, 1994; Andersson 2006; Bakaki, Böhmelt, and Bove 2016). Consider, for example, Böhmelt (2011: 868), who argues that democratic interveners are likely to be more effective as “they operate under a norm of bounded competition that favors the use of compromise” (see also Dixon 1993). A second reason is that democracies are seen as more credible as they “face greater audience costs for deception in the conflict management process because they face greater scrutiny in the free press and because they pay domestic costs for foreign policy failure” (Crescenzi et al. 2011: 1,089). More credibility is associated with higher effectiveness, which makes the fighting parties more likely to demand democratic third parties over other types of interveners (Hellman 2012).

A third reason is that mediation is more likely to occur in the more difficult conflicts (e.g., Bercovitch and Jackson 2001; Greig 2005; Greig and Regan 2008; Beardsley and Greig 2009; Melin and Svensson 2009; Beardsley 2010; Hellman 2012; Wallensteen and Svensson 2014; Bakaki, Böhmelt, and Bove 2016; Böhmelt 2016; Keels and Greig 2019; Lundgren and Svensson 2020). In general, these cases are the ones requiring assistance from third parties (see Mehrl and Böhmelt 2021), since belligerents can solve their differences more often by themselves in the “easier,” i.e., less intense disputes. Violence, in particular, shapes conflict intensity (Bercovitch and Langley 1993; Böhmelt 2016), driving the “immediate, current costs of conflict experienced by the warring parties” (Greig and Regan 2008: 772). It may even be suggested (see, e.g., Bercovitch and Houston 2000) that violence is the only conflict characteristic sufficient to persuade the antagonists and (potential) mediators that a conflict is “ripe” for intervention (see also Böhmelt 2016). In the words of Bercovitch (2004), “a ripe moment describes a phase in the life cycle of the conflict where the parties feel exhausted and hurt, or where they may not wish to countenance any further losses and are prepared to commit to a settlement, or at least believe one to be possible” (see also Greig and Regan 2008). Hence, conflict intensity is strongly linked to conflict ripeness and the likelihood of mediation (see also Böhmelt 2013).

Against this background, I develop the counterintuitive argument that international mediation is only more likely when conflict intensity is low. The starting point for this claim is that mediation, and especially mediation failure, is costly for the third party (Greig 2005; Beardsley 2010, 2011; Beardsley and Lo 2013; Chatagnier 2019), particularly as to reputation—the domestic audience can be “concerned with their state’s reputation” (Chiba, Johnson, and Leeds 2015: 968). Conflict intensity and, in turn, the (anticipated) outcome of an intervention determine the amount of these costs. The more difficult cases are precisely those that need outside assistance. Moreover, these disputes are also more difficult to solve, making mediation failure more likely and an intervention more costly (Bercovitch and Schneider 2000; Beardsley and Lo 2013). Note that this mechanism affects all potential mediators in the same way, independent of their form of government, which raises the question what role democracy plays.

I assume that political leaders want to retain power and, to this end, implement policies and take actions that they believe will extend their stay in office (Bueno de Mesquita et al. 2005)—policies and actions that benefit those citizens whose support they need for gaining and staying in office. Successful mediations in which political leaders produce real benefits for these citizens, and for the country as a whole, may help them to stay in power. Bueno de Mesquita et al. (2005) defined the terms of the “selectorate” and the “winning coalition” in this context. The former pertains to those who participate in the selection process for a political leader. The latter stands for that selectorate share, which keeps a leader in power. When comparing democracies with other forms of government, authoritarian regimes tend to have smaller winning coalitions relative to the selectorate. Democracies, in fact, are generally characterized by larger groups that must be pleased in order to prolong political survival. What is more, democratic citizens can participate more directly in the leader‐selection process and leaders in democratic countries can be removed from office more easily than nondemocratic ones if the domestic audience does not favor their policies (Bueno de Mesquita et al. 2005).

This creates a “democratic self‐selection mechanism.” Political leaders, when considering whether or not to mediate a conflict, anticipate that some problems will occur over the course of an intervention. In the worst case, mediation fails, which raises the costs of engagement significantly and could make the domestic audience less supportive of the leadership. While domestic issues are more visible to an electorate than international mediation efforts, there is ample research demonstrating that voters also care about their countries’ foreign policy and international reputation, and that they will punish their leadership if they are dissatisfied with its performance in the international arena (Tomz 2007, 2008, 2012; Tomz and Weeks, 2013, 2020; Chiba, Johnson, and Leeds 2015; Tingley and Tomz 2020). McGillivray and Smith (2008) have demonstrated that domestic audiences benefit from punishing their leaders for reputational damage. If a leader’s reputation is damaged by a failed mediation attempt of the leader or members of the government and state bureaucracy (Beardsley 2011; Beardsley and Lo 2013), the public could become less supportive of the executive (see also Chiba, Johnson, and Leeds 2015).

As indicated above, removing a leader from office is less costly and can be done more quickly in democracies, while democratic state executives are more strongly dependent on larger parts of the population for their political survival (Bueno de Mesquita et al. 2005). Thus, democracies face higher costs stemming from (primarily) domestic audiences when mediation fails, which makes it likely that democratic leaders will be more careful in terms of selecting the conflicts they offer to mediate. In turn, democratic third parties should seek to avoid challenging cases that make mediation failure more likely, but instead opt for intervening in those disputes where tangible benefits can be more easily achieved at a lower risk. As a result, I expect democracies to mediate only those cases that they anticipate are easy to solve, and they should be less likely to commit to mediating the more difficult disputes.^1.

The main data source for my empirical analysis is the Civil Wars Mediation (CWM) data set (DeRouen Jr, Bercovitch, and Pospieszna 2011). Using the definition of civil conflict as set forth in the Uppsala Conflict Data Program (UCDP) Armed Conflict data set (Gleditsch et al. 2002; Pettersson and Eck 2018), the latest version of the CWM data set has information on mediations in all intrastate disputes between 1946 and 2011. Using these data, I first disaggregate the information into conflict years, creating yearly observations for each dispute. A conflict drops off the data set when it is terminated either militarily or peacefully. In turn, each conflict year is paired with any state in the international system (except for the country at war) as these constitute the universe of cases of potential mediators. The unit of analysis I employ is then the conflict‐potential‐ mediator year and, without missing values, there are initially 294,807 observations or data points in total since 1946. According to the CWM data, 0.14 percent of those data cases ultimately saw mediation. This information on instances of third‐party mediation is used as my dependent variable and, due to its binary nature, I rely on logistic regression models. To account for temporal dependencies in mediation, I include the cubic‐polynomial approximation (Carter and Signorino 2010) of the time since the last mediation attempt (if any) in all models. In essence, the cubic polynomials capture how much the time since the last intervention (if any) shapes the likelihood of mediation. Including these variables also allows me to control for the argument that prior mediation is a strong predictor of mediation: third parties that have previously mediated a conflict are more likely to do so in the future.

The core explanatory variables of interest pertain to the potential mediator’s regime type, conflict intensity, and an interaction of the two items: a multiplicative specification where I use the product of the democracy variable and the intensity item (Brambor, Clark, and Golder 2006; Berry, Golder, and Milton 2012). First, for each potential intervener, I code a variable receiving a value of 1 if it is democratic in a given year (0 otherwise). For defining democracies, I employ the Polity IV data (Marshall and Jaggers 2018) and use the commonly used cut‐off point of 7 (Jaggers and Gurr 1995; Goldsmith, Chalup, and Quinlan 2008; Bogaards 2012). Second, conflict intensity is usually defined by fatalities (DeRouen Jr, Bercovitch, and Pospieszna 2011; Hellman 2012; Bakaki, Böhmelt, and Bove 2016; Böhmelt 2016; Mehrl and Böhmelt, 2021) and I draw on the Peace Research Institute Oslo (PRIO) Battle Deaths data set (Lacina and Gleditsch 2005; Lacina, Gleditsch, and Russett 2006) to this end. The most current version was released in 2017 and covers the conflict‐years between 1946 and 2008. I employ a variable capturing the number of battle‐related deaths in a given conflict‐year as these are likely more visible in the international arena than a cumulative measure. The variable is also temporally lagged and log‐transformed. Both the democracy variable and the battle‐deaths item, introduced to the models and examined individually, should be positively signed, i.e., thus raising the probability to see mediation. Jointly considered via the muliplicative interaction, however, I expect to see a negative effect.

Coming to these confounding factors, I follow existing scholarship and model mediation onset as a function of supply and demand (e.g., Greig 2005; Greig and Regan 2008; Beardsley and Greig 2009; Beardsley 2011; Crescenzi et al. 2011; Keels and Greig 2019; Lundgren and Svensson 2020). In light of this, the controls pertain to other (potential) mediator characteristics besides regime type, features of the belligerents (mainly the state‐representing governmental actor in a civil war), and aspects of the conflict next to intensity. First, I control for third parties’ power as measured by the Correlates of War’s Composite Index of National Capability (CINC) score (Singer 1988) as well as population and GDP per capita. The latter two are log‐transformed and originally taken from the World Bank Development Indicators.^2.

Second, demand and supply are also captured via the ties connecting a conflict party and a potential mediator. I include the logged capital‐to‐capital distance between the third party and the civil‐war state (see also Crescenzi et al. 2011). Smaller distances are likely tied to greater willingness and opportunity to intervene, making mediation more likely. In addition, I created two indicators on states’ joint memberships in international organizations (Melin 2011). The variable Allies is binary and receives a value of 1 if the civil‐war country and a third party share at least one alliance membership, the latter being defined by the Alliance Treaty Obligations and Provisions (ATOP) data (Leeds, Ritter, and Mitchell 2002). The item IGO Memberships is the number of co‐memberships in international organizations as coded by the Correlates of War Project (Pevehouse, Nordstrom, and McManus 2020). More joint memberships between two states signal common norms and interests in foreign affairs, which may be positively associated with mediation demand and supply. In the Appendix, I also consider colonial and ethnic ties.

The variables I employ to characterize belligerents’ characteristics largely mirror the ones for the third party. Hence, there are items for the civil‐war state’s regime type (operationalized the same way as the binary regime‐type indicator introduced above), its CINC score, its population, and its income. Finally, I include a variable on the duration of a conflict until a given year, which I calculated using the information in the CWM data set (DeRouen Jr, Bercovitch, and Pospieszna 2011). Table One summarizes the variables’ descriptive statistics.

The main models of the analysis are summarized in Table Two: Model 1 only incorporates the core variables of interest (next to the temporal controls), but not the interaction term. The control covariates are then introduced in Model 2, albeit the multiplicative specification is still missing. Finally, Models 3 and 4 comprise Democratic Mediator × Battle‐Related Deaths (log), while the latter is my full specification with the control variables included as well. The table entries allow only for a direct interpretation of the signs and significance levels, but not the strength or substance of an effect. Thus, I show average marginal effects and first difference estimates in Figures One and Two, respectively.

I begin by discussing Models 1–2, which do not incorporate the interaction term Democratic Mediator × Battle‐Related Deaths (log), but introduce the variables for third parties’ regime type and conflict intensity individually. The inclusion of the control variables does not alter the effect direction of either Democratic Mediator or Battle‐Related Deaths (log), but it changes the statistical significance: Democratic Mediator is no longer significant in Model 2, but Battle‐Related Deaths (log) now is. Generally, though, and in line with previous research (e.g., Bercovitch and Jackson 2001; Greig 2005; Greig and Regan 2008; Beardsley and Greig 2009; Melin and Svensson 2009; Beardsley 2010; Crescenzi et al. 2011; Hellman 2012; Beardsley and Lo 2013; Wallensteen and Svensson 2014; Bakaki, Böhmelt, and Bove 2016; Böhmelt 2016; Keels and Greig 2019; Lundgren and Svensson 2020; Yazici 2020), the first two models suggest that democratic states can be more likely to mediate, and that more intense disputes are more likely to attract mediation.

The positive effects of the individual items pertaining to third‐party democracy and conflict intensity remain when including the multiplicative term in Models 3–4. Given the latter component that is included now, however, they can only be interpreted with the other term set to 0. For instance, keeping Battle‐Related Deaths (log) constant at 0, Democratic Mediator has a positive and significant effect on the likelihood of mediation—the coefficient estimate is 1.464 in Model 4. Nonetheless, the interesting aspect of Models 3–4—and the actual test of the theoretical argument—is Democratic Mediator × Battle‐Related Deaths (log). As expected, the coefficient is negatively signed and significant. That is, democratic third parties are likely to be associated with a lower likelihood of mediation when dispute intensity is high. A direct, substantively meaningful interpretation of this coefficient estimate is difficult with the table entries alone, but I have calculated average marginal effects for the interaction term that are displayed in Figure One. As shown there, the positive and significant effect of Democratic Mediator persists—but only for low to medium levels of Battle‐Related Deaths (log). The likelihood to mediate is higher by about 0.1 percentage points for democratic third parties when a conflict has seen about twenty deaths. While this may seem like a small effect, recall the relatively large sample size and that mediation is a rare phenomenon in general; given these two characteristics, any effect is likely to be rather small to begin with. Returning to Figure One, the impact of Democratic Mediator decreases and, eventually, becomes indistinguishable from 0 with a larger number of fatalities. According to Figure One, democratic third parties are not more or less likely to mediate than nondemocracies when fatalities are higher than about 500–1,000.

Connecting these findings back to the theoretical argument and existing literature, I do find support for earlier results reporting that it is especially the more difficult cases that attract mediation and that democratic third parties are particularly more likely to intervene (e.g., Bercovitch and Jackson 2001; Greig 2005; Greig and Regan 2008; Beardsley and Greig 2009; Melin and Svensson 2009; Beardsley 2010; Crescenzi et al. 2011; Hellman 2012; Beardsley and Lo 2013; Wallensteen and Svensson 2014; Bakaki, Böhmelt, and Bove 2016; Böhmelt 2016; Keels and Greig 2019; Lundgren and Svensson 2020; Yazici 2020). However, the joint, interactive effect is negative, thus shedding light on quite a few instances of difficult conflict circumstances, where democracies did not mediate. Specifically, my results are consistent with the claim that democratic leaders will anticipate the costs of mediation, which are especially high when interventions fail. As democracies suffer more than authoritarian regimes from audience costs, particularly domestic audience costs, they select themselves only into those conflicts where the chances of success are a priori higher—namely, the low‐intensity, low‐fatality disputes (see also Greig 2015). The more difficult cases are avoided, as these might be more challenging to solve, hence raising the costs and, ultimately, threatening a leader’s ability to stay in power. This result is also in line with an implication in Beardsley and Lo, namely, that “democratic neighbors will be especially more likely to become involved as conflict managers as salience increases and the possibility for settlement remains feasible” (2013: 80).

Coming to the control variables, their effects are consistent across model specifications and also mirror what earlier research has reported. First, more powerful third parties are more likely to mediate. This is evidenced by the positive and significant effects for Mediator CINC Score, Mediator GDP per capita (log), and Mediator Population (log) in Models 2 and 4. In general, these findings echo the rationale that belligerents tend to demand mediation from stronger actors as these might be better able to provide material incentives for settling a conflict (e.g., Bercovitch and Schneider 2000; Böhmelt ). At the same time, from a supply‐side perspective, more powerful actors have more means for intervention. Second, all variables pertaining to the state actor in a civil war are mostly statistically insignificant and not substantively meaningful. The only exception according to Table Two and Figure Two is that more populous belligerents are more likely to see mediation.

Third, ties binding third parties and antagonists are also found to be crucial. Smaller distances between and joint memberships in international organizations of state actor and third party increase the likelihood of mediation, although military alliances have a more substantive impact. Links via any IGO are also positively signed in Table Two and Figure Two, but not as substantively important as the Allies item. In fact, Allies exerts one of the strongest effects in the models. Finally, there is Duration, which is statistically significant in Table Two. This is mirrored by a first difference in Figure Two.

In the Appendix, I discuss additional analyses and a series of robustness checks that further increase the confidence in my core finding. Most importantly, I replicate one of the most influential studies on mediation in interstate conflicts (Crescenzi et al. 2011) and find a pattern that mirrors what I argue here: democratic third parties avoid difficult conflicts. Hence, my argument’s scope conditions are not limited to civil wars, but all sorts of disputes. Second, I summarize a model based on a non‐lagged intensity variable and plot the temporal dependence in mediation onset to show that the likelihood of third‐party intervention drops quickly over time as it approximates 0 after about five years. Third, I alter the cut‐off points for third parties to be defined as democracies and I use the original version of the polity2 scale (Marshall and Jaggers 2018). I control for the presence of other third parties and consider colonial as well as ethnic ties. Finally, I focus on politically relevant third parties only. The corresponding models and graphs are summarized in the Appendix.

Existing scholarship consistently reports that, all else being equal, democratic third parties are more likely than nondemocratic parties to intervene as mediators in conflicts. Research also shows that conflicts that are especially difficult are more likely than less difficult conflicts to attract the attention of mediators. I do not question these findings and, in fact, find support for these claims in my research. However, I advance the argument that a non‐positive impact is likely to be exerted on mediation onset once we consider third‐parties’ regime type and conflict intensity jointly. Specifically, democracies are more vulnerable to audience costs. After all, when dissatisfied with their government’s performance, a democratic audience will find it easier to remove a leader than an autocratic one. The crucial point here is that democratic leaders want to avoid policies or outcomes that may not please their primarily domestic audiences. Mediation failure could be one of those policies that leaders would want to circumvent as ineffective mediation interventions have the potential to impose high costs. While these costs do exist for autocratic leaders as well, they are more damaging in the case of democracies. The core of my theory states that democratic leaders likely anticipate all this and, thus, will try to avoid the more difficult cases. Instead, democratic leaders select themselves into the easier disputes, as these are more likely to be mediated successfully and, hence, are less costly.

I find strong and robust support for this argument. My main empirical analysis makes use of the CWM data set, covering civil wars and mediation attempts from 1946 until 2011. I also examine the validity of my theory with data on interstate conflicts in the supporting information set forth in the Appendix. The findings from these analyses point to the same pattern: democracies are more likely than nondemocracies to mediate low‐intensity disputes, but the positive impact on mediation onset vanishes once conflict intensity reaches higher levels. This result has important policy implications and may inform future scholarship on international mediation. This research sheds additional light on the persistent selection effects surrounding mediation (e.g., Gartner and Bercovitch 2006; Beardsley 2008; Böhmelt ; Beber 2012; Melin, Gartner, and Bercovitch 2013). It follows from there that belligerents may want to reconsider a mediation offer from a democracy as third‐party intervention—the antagonists could well be able to settle a conflict on their own and a third party may have merely offered their services as it is an “easy” case. It is particularly when democracies do not want to mediate that third‐party intervention is necessary. And given that democracies are also seen as more effective mediators than nondemocratice governments (Dixon 1993, 1994), the international community may want to provide extra incentives for democracies to intervene and try to peacefully settle disputes. Having said that, the high rate of effectiveness commonly associated with democratic mediation could simply be driven by their self‐selection into the less difficult cases. When accounting for this selection process, democratic third parties may be no more or less successful than their autocratic counterparts (see also Böhmelt 2011).

Future research on this topic could address several issues. First, new data collection efforts are necessary. Mediation remains a popular conflict‐management technique (but see also Lundgren and Svensson 2020), but data availability is limited and not updated for the most recent interstate and intrastate disputes. Second, the scope conditions and generalizability of the mechanism I postulate are quite broad, and it may be a worthwhile effort to examine whether peacekeeping is also affected by the identified selection problem (see Andersson 2006). Third, recent research intensified efforts to examine the reputation of actors involved in mediation attempts (Keels and Greig 2019). Given my results, it may be fruitful to examine the reputational concerns of (potential) mediators more thoroughly in future research. Fourth, if costs, accountability, and political‐survival considerations matter for mediation, sharing the burden of an intervention could be an option, raising the question of how self‐selection by democracies into easier disputes works under multiparty mediation (Böhmelt 2011, 2012). Finally, not all democracies are created equal. Disaggregating democratic regime type and reanalyzing the mechanism discussed here could further our understanding of conflict resolution and international mediation significantly.

In this Appendix, I present the following analyses and robustness checks that complement the main article’s argument and findings:

• A replication of a previous study on interstate conflicts implementing my interaction specification.

• I employ a non‐lagged intensity variable.

• Examining the temporal dependence of international mediation onset.

• I alter the cut‐off points defining the regime type of third parties.

• The original polity2 scale instead of a binary‐democracy specification for third parties and belligerent states.

• I control for the presence of other mediators.

• Models that incorporate variables on ethnic ties and colonial ties.

• I focus on politically relevant third parties.

The data I employ for the main analyses focus on domestic, i.e., intrastate conflict (DeRouen Jr, Bercovitch, and Pospieszna 2011). However, the proposed argument applies principally to interstate, that is, state‐to‐state, disputes as well. The reason for not including an analysis of such conflicts in the main text is essentially data availability: the data for interstate conflicts and mediation have either not been updated or lack sufficient information on the third party. For example, the most updated and arguably most comprehensive relevant data set is the International Crisis Behavior project (Brecher and Wilkenfeld 1997), which covers 482 international crises between 1918 and 2016 in its 13th edition. However, the identity of mediating third parties and especially their regime type is not always directly accessible. Moreover, the Issue Correlates of War Data Set (Frederick, Hensel, and Macaulay 2017; Hensel and Mitchell 2017) is another comprehensive data source on interstate disputes, but its coverage ends in 2001.

Having said that, despite data limitations, it still seems an effort worth making to examine whether the patterns I argue for also hold in the context of interstate conflict and, thus, I draw on Crescenzi et al. (2011), which is one of the most influential studies in the field on mediation demand and supply. In Table A1 of this Appendix, I replicate their main results (Crescenzi et al. 2011: 1084 (Models 3–4)) while including the interaction term for third‐party democracy (which is called Mediator’s Polity in Crescenzi et al. (2011)) and conflict intensity (which is called Issue Salience in Crescenzi et al. (2011)). The unit of analysis in the following models mirrors my setup of the main text, while I cover contentious issue claims between states until 2000. Model specifications and variables’ operationalization as well as their data sources are comprehensively described in Crescenzi et al. (2011: 1080ff).

Table A1 of this Appendix then summarizes two models. The first estimation is a probit model and I include the interaction Mediator’s Polity × Issue Salience. The second model is a Heckman‐type probit estimation comprising two stages: one outcome stage pertaining to whether mediation led to an agreement (or not) and a selection stage capturing mediation onset (i.e., whether it occurred or not). Regardless of the specification, though, the multiplicative term Mediator’s Polity × Issue Salience is negatively signed and significant at conventional levels. This mirrors my results discussed in the main text and emphasizes that my argument holds in the context of interstate disputes.

My intensity variable, Battle‐Related Deaths (log), captures the number of battle deaths of a given conflict‐year and is temporally lagged. As discussed in the main text, there are theoretical reasons and endogeneity concerns to do this. However, I also consider a variable that is not temporally lagged as dispute fatalities may well shape the prospects of mediation in the same year. Table A2 in this Appendix presents a model using a battle‐related deaths variable that is not temporally lagged; omitting the temporal lag for conflict intensity does thus not drive the results in important ways. All other robustness checks in the following are also based on present‐year values.

Mediation is time dependent: prior intervention is a strong predictor of mediation in that third parties that have previously mediated a conflict are more likely to do so in the future. I control for temporal autocorrelation like this with the standard approach suggested in Carter and Signorino (2010). Thus far, however, I have not interpreted the variables of this cubic‐polynomial approximation. Figure A1 in this Appendix addresses this as, based on Model 3 above, I have calculated the predicted probabilities of mediation onset for the years since the last mediation attempt (if any) while holding all other variables constant at their means. As shown in this graph, the likelihood of mediation is relatively high at an early stage of a conflict or immediately after a preceding intervention. This probability drops sharply, though, approximating 0 after about ten years since conflict outbreak/the last third‐party mediation.

I defined third‐party and belligerent democracies by via a binary version of the Polity IV project’s (Marshall and Jaggers 2018) polity2 score: in order to be categorized as democracies, states must score at least a value of 7 on the ordinal polity2 scale between −10 and +10. While this approach is commonly chosen in the literature (Jaggers and Gurr 1995; Goldsmith, Chalup, and Quinlan 2008; Bogaards 2012), Bogaards (2012) states that it is also quite conservative as some country‐years that may be characterized as “democratic” are left out. To this end, I lower the cut‐off point to +5: any country (third party or belligerent) scoring at least 5 on the polity2 scale is now defined as a democracy. I then re‐estimate Model 3 above with this new specification and present the results in Model 4 of this Appendix: the core finding is robust as Democratic Mediator × Battle‐Related Deaths (log) remains statistically significant at conventional levels and is negatively signed.

As with the previous robustness check, I use an alternative specification to define third parties’ and belligerents’ regime type. Instead of a binary classification, I now rely on the original polity2 score of the Polity IV data (Marshall and Jaggers 2018), which ranges between −10 (full autocracy) and +10 (full democracy). The revised model along those lines is presented in Table A4 of this Appendix: Mediator Polity2 × Battle‐Related Deaths (log) remains statistically significant.

The likelihood that a democracy mediates may be driven by the degree to which there is a need for mediation in the first place (e.g., Greig 2005; Greig and Regan 2008; Beardsley and Greig 2009; Beardsley 2011; Crescenzi et al. 2011; Keels and Greig 2019; Lundgren and Svensson 2020). I control for several confounding factors in this regard. However, the main models leave out whether another third party already intervenes. There is an extensive body of work on such instances of “multiparty mediation” (e.g., Böhmelt 2011, 2012). On one hand, if third parties already conduct a mediation before another state considers intervening, there might be a lower chance that mediation occurs as the “new” third party could see less of a need for its own involvement. At the same time, however, democracies in particular may find such a context attractive as other third parties’ involvement implies burden sharing in terms of material costs and reputational risks in case the mediation fails. I examine the influence of these two competing mechanisms by adding a control to this Appendix’s Model 3, which codes dichotomously whether at least one other third party is already involved in a conflict (or not, which is then coded as 0).

As Model 6 below shows, on one hand, the main result of a negative and significant interaction term is robust. On the other hand, the variable on multiparty mediation is positively signed and significant, which suggests that the presence of other mediators already involved increases the likelihood of mediation, including for democratic third parties. In the words of Böhmelt (2012: 704): “Unlike in the case of single‐party mediations, however, multiparty mediation signals to the domestic audience that the costs of intervention are being shared and, hence, should be lower than in the former case. Simultaneously, if the costs of a mediation attempt are shared in the sense that more than one party intervenes, this also increases the legitimacy of an intervention.”

In the main text, I control for the distance between conflict state and potential mediator, their ties via alliances, as well as their co‐memberships in international organizations to capture how state relationships affect the likelihood of mediation. In Table A6 of this Appendix, I also control for ethnic ties and links stemming from the colonial history of a country. Specifically, data on ethnic ties were assembled by Koga (2011) and the corresponding variable codes whether an ethnic link exists between the conflict state’s population and the potential intervener. Second, for the variable Colonial History, I code whether a potential mediator had a colony in a conflict state in the past (Rouvez 1994).

The main finding is robust and the inclusion or exclusion of the additional “ties” variables does not influence the overall pattern. However, while the ethnic‐ties item is statistically insignificant, Colonial History is positively signed and significant at 1 percent, which highlights that a shared colonial history between conflict state and third party very much influences the prospects of the third party's intervention. This mirrors earlier studies’ results (Rouvez 1994; Bove and Böhmelt 2019).

Not all third parties are equally likely to mediate conflicts. For example, the chances of Mongolia mediating the Colombian conflict are plausibly quite low to begin with. To account for this issue, I also present a model (Table A7 of this Appendix) that merely considers “politically relevant” third parties as potential mediators as defined in Lemke and Reed (2001). Politically relevant potential mediators are states neighboring a conflict country and/or major powers. Contiguity and major‐power status are defined in Stinnett et al. (2002).

When decreasing the sample size and focusing on “politically relevant” (Lemke and Reed 2001) potential mediators only (Model 9 of this Appendix), the chances of mediation are linked to a rise of about 1.2 percentage points for democratic third parties when a conflict has an intensity of about twenty fatalities. Hence, the main finding of conflict intensity moderating the likelihood of democratic third parties mediating is also robust in this last analysis.