The Effectiveness of Peacekeeping Missions in Civil Wars: Disaggregating Interventions, 1948-2000 Tobias Böhmelt, Nils W. Metternich, Ulrich H. Pilster, and Andrea Ruggeri* University of Essex, United Kingdom First Draft (10,199 words) December 8, 2009 * The authors are listed in alphabetical order.
2 Abstract The recent literature on the effectiveness of peacekeeping missions is largely driven by two streams. While some scholars provide evidence for a significant positive contribution, others claim that the United Nations and other peacekeeping organizations tend to negatively influence peace processes or have no impact at all. Two shortcomings are characteristic for both rationales: peacekeeping interventions are either operationalized as a simple binary variable or categorized along the distinct purposes of the missions in question. This, however, ignores the composition of peacekeeping troops, since peacekeeping interventions are treated as a black box, thereby leaving out valuable information. We intend to fill this gap by disaggregating state coalitions in peacekeeping operations. More specifically, this paper analyzes in detail the size and leverage of an intervening group, peacekeepers regime types, and their supply-side incentives. We elaborate a theory and, utilizing newly compiled data on peacekeeping operations in civil wars from 1948 to 2000, test our hypotheses with a selection-duration estimator. Keywords: peacekeeping; effectiveness; group size; regime type; supply-side incentives; selection-duration analysis
3 Introduction The recent research on the effectiveness of peacekeeping missions by the United Nations (UN), other international organizations, ad-hoc coalitions, or a single state is driven by a heated debate. While scholars such as Doyle and Sambanis (2000; 2006), Fortna (2003; 2004), and Gilligan and Sergenti (2006) provide evidence for a significant positive contribution, authors such as Dubey (2003) claim instead that peacekeeping missions tend to have no impact at all. Furthermore, Greig and Diehl (2005) even find evidence that peacekeeping is likely to have a negative impact on the long-term effectiveness of a peace process. Two shortcomings pertaining to the examination of peacekeeping per se are characteristic for these streams in the literature. First, peacekeeping missions are either treated as a simple binary variable or, second, scholars merely distinguish along the categories of involvement, e.g., enforcement, monitoring, or state building. In other words, when studying the effectiveness of these interventions, the previous work has largely ignored the characteristics of the nation states that form a peacekeeping coalition, and how these countries coordinate their mission. This, consequently, disregards the composition of peacekeeping troops, and leaves out valuable information that may be able to contribute significantly toward our understanding of the effectiveness or failure of peacekeeping missions. We intend to fill this gap by disaggregating state coalitions in peacekeeping operations, thereby opening the black box of such interventions. The rationale behind our research is grounded in the fact that states contributing to such operations are driven by their own incentives and are tied to each other through various kinds of relationships. They share information and resources, and pursue other means of synchronizing and/or integrating activities. Militaries operating within a peacekeeping coalition vary in terms of their doctrines, technology and their operational readiness, training, intelligence and communication, mobility and force protection, or
4 logistics. Ultimately, this may make interventions more synergetic or create conflicts among the peacekeepers that actually hamper successful operations. In addressing these issues in detail, this paper contributes to the existing literature by focusing on a set of crucial aspects of the interveners that form peacekeeping missions. First, we examine the size of an intervening group, arguing that size and peacekeeping effectiveness are characterized by an inverted U-shaped relationship. More interveners are generally able to draw upon a vast amount of leverage and resources, and create synergy due to combined efforts, making them more effective. On the other hand, a very large number of interveners also implies higher organizational costs and greater heterogeneity of interests, leading to less effective outcomes (Buchanan and Tullock 1962; Snidal 1994; see also Olson 1965). Second, we examine the regime type of peacekeepers and how this affects their ability to coordinate their efforts. A coalition of states that is largely democratic should be more effective due to inclusivity, a shared culture of peaceful conflict resolution, and increased communication flows (see Dixon 1994; Raymond 1994; Bercovitch and Houston 2000; Mitchell 2002). Finally, we focus on the supplyside incentives of peacekeepers. Countries generally choose to participate in peacekeeping missions due to rational calculations and, hence, this does not occur at random. As a result, only if countries have a thorough interest in peace in a conflict country, then we would expect them to invest more in order to prevent further fighting, ultimately making those peacekeeping operations more effective. In the remainder of this paper, we first review the relevant literature on the effectiveness of UN and non-un peacekeeping missions, emphasizing their weaknesses in light of the possibility of disaggregating interventions. Building upon existing studies that examine peacekeeping missions from a commitment-problem perspective, we then develop a theory that combines group size, regime type, as well as supply-side incentives with peacekeeping effectiveness. Afterwards,
5 we detail the research design and, utilizing newly compiled data on peacekeeping operations in civil wars from 1948 to 2000, test our hypotheses with a selection-duration estimator (Boehmke, Morey, and Shannon 2006). The final section concludes. Literature Review Do peacekeepers promote peace? In particular against the background that the post Cold War era has seen an impressive increase in peacekeeping interventions, this is a crucial question and answering it is not only of academic interest, but should also significantly influence the rationales of policymakers (Fortna and Howard 2008). However, the previous literature has not yet been able to agree if peacekeepers help sustaining peace after conflicts or if they are actually counterproductive due to the failure of their interventions. Nevertheless, multiple studies tried to answer whether peacekeepers can make a positive contribution to peace, but as stated above obtained mixed findings. First, Doyle and Sambanis (2000; see also 2005; 2006) examine the interventions of UN peacekeepers and -builders in civil wars after the end of World War II. The authors argue that peacekeeping missions have to address the roots of hostility, local capacities for change toward a peaceful settlement, and a sufficiently large international commitment to the peace process for reaching effective outcomes. In order to test their claims, Doyle and Sambanis (2000: 783f) employ a binary item on the success/failure of a peacekeeping effort and, in consistence with their empirical findings, conclude that the UN has a significantly positive impact toward resolving civil wars and keeping peace. Second, Fortna (2003; 2004) pursues to contribute to the same research question, albeit with a different research design. Instead of relying on the rather subjective binary variable from Doyle and Sambanis (2000; see also 2006), she uses the duration of peace after the warring parties
6 agreed on a ceasefire for operationalizing peacekeeping effectiveness. Among other results, Fortna (2004: 283f) finds support that peacekeeping interventions of both the UN and non-un organizations lengthen the peace and increase post-conflict stability. The author also naïvely disaggregates peacekeeping missions along 1) observer, 2) traditional, 3) multidimensional, and 4) enforcement interventions, and apart from the latter category obtains a significant as well as positive contribution toward peace for all of these mission types. However, King and Zeng (2007a; 2007b) challenge Fortna s findings from a methodological perspective. More precisely, conflicts with peacekeeping interventions are unlikely to be a randomly selected set and ignoring this can either under- or overestimate the effectiveness of third-party interventions. Because of this self-selective character of peacekeeping missions, a research design that does not consider selection effects can give rise to misleading and biased findings (Heckman 1979). As a suggestion to circumvent this issue, Gilligan and Sergenti (2006) employ matching methods. Matching corrects for the non-random assignment and controls for the existence of confounding factors by identifying cases in a control group that match as closely as possible treated observations, i.e., cases where we actually observe peacekeeping. The differences between the treated and the untreated cases are then reduced to a minimum, and based upon the comparison of these most similar cases, Gilligan and Sergenti (2006) can make inferences about the causal effect of UN peacekeeping, where any difference between the two sets is then solely attributed to the treatment. Given a sophisticated research design, the authors are able to rule out influences of several covariates and find that UN peacekeeping is indeed likely to make a positive contribution to peace. A shortcoming of matching as such is again, however, that it requires treating peacekeeping as a simple dichotomous item. Next to the studies of these previous authors, another stream in the literature takes a different perspective. Dubuy (2003) does not find that UN peacekeeping either shortens or lengthens the
7 duration of peace after a civil war has ended. Further, although Greig and Diehl (2005) are able to unveil some evidence that UN peacekeeping helps decreasing the chances of renewed warfare, this is rather a short-term effect. In the long run, peacekeeping interventions make it actually less likely that belligerents will engage in negotiations to completely resolve the conflict, and, hence, this increases the chances that a new dispute will break out due to the same issues that caused that conflict in the first place. Finally, Diehl, Reifschneider, and Hensel (1996; see also Beardsley and Schmidt 2008) examine the impact of UN interventions on the likelihood and severity of repeated conflict between belligerents. These authors categorize UN interventions along their underlying purposes, i.e., passive interventions, active missions, operational deployment, and a residual category, but Diehl et al. (1996: 693ff) do not find any significant effect of either mission type on the chances of repeated dispute or conflict severity. In sum, despite the fact that all these studies addressed the same research question, the findings are mixed. One stream finds a positive contribution of peacekeeping interventions, while the other is highly skeptical of that. We argue that two key theoretical shortcomings, which have major empirical implications, are likely to be the main reason for that. In all of these examinations, peacekeeping missions are either treated as a simple binary variable, i.e., did peacekeeping occur or not, or just characterized along the mission type or purpose of an intervention in question. This leaves out a significant amount of valuable information, however, since these previous models do not look beyond the dichotomous distinction between peacekeeping missions and cases that did not see those interventions. One notable exception is Mariani (2008), who argues in consistence with our project that the existing research has failed to produce a sound theory of peacekeeping because it treated peacekeeping operations as a black box. The author attempts to disaggregate UN interventions along the functional specifications of a mission in question. Put differently, Mariani (2008) uses the peacekeeping functions framework
8 as developed by Diehl, Druckman, and Wall (1998) in order to categorize peacekeeping missions as 1) traditional peacekeeping, 2) observation missions, 3) collective enforcement, 4) election supervision, 5) humanitarian assistance, 6) state building, 7) pacification, 8) preventive deployment, 9) arms controls, 10) protective services, 11) interventions in support of democracy, and 12) sanctions enforcement. Employing these categories, the author then examines their impact on the duration of peace after a ceasefire agreement. However, Mariani s (2008) findings are largely inconclusive, lacking substantial or statistical significance at conventional levels. Although we appreciate this first step toward moving beyond the simple dichotomy, we believe that Mariani s (2008) findings emphasize that other mechanisms are at work when disaggregating peacekeeping missions. More specifically, in the following we focus on the actors that form a peacekeeping mission, i.e., the nation states. These countries differ in terms of their interests, preferences, regime types, or capabilities, which in turn is likely to influence whether a peacekeeping intervention will be effective or not. Theory: Disaggregating Peacekeeping Missions - Size, Democracy, and Supply- Side Incentives General Approach. Commitment problems are the focus of numerous studies that seek to explain the onset (e.g., Powell 2006), duration (e.g., Fearon 2004), or recurrence (e.g., Werner and Yuen 2005) of interstate disputes and civil wars (see also Fearon 1995; Walter 1997; 2009). 1 Commitment problems arise if belligerents perceive a high risk that opponents will renege on agreements in the future (Powell 2006). Reneging on settlements is usually caused by future power shifts, which can occur due to the following reasons. First, warring parties typically have to demobilize after the end of a conflict. However, demobilization increases the vulnerability of 1 Due to the focus of our study, we elaborate our arguments solely in a civil-war context.
9 ex-belligerents. Especially in the context of civil wars, the government is generally the only party left with forces, weapons, and sufficient military power to fight possible internal threats after demobilization. This, however, implies that once the rebel organization(s) have decreased their military capacity, the government has an opportunity to renege on former agreements (Walter 2009). Second, even if demobilization is not implemented, former opponents are likely to have different recovery rates, which may highly determine their prospective fighting capabilities or bargaining leverage afterwards. Finally, the incumbent government might exploit post-war institutional weaknesses to politically marginalize former opponents. For example, leaders in democratic systems can take advantage of different mechanisms to manipulate political power. One of the most prominent ones is systematic electoral fraud (Collier 2009). Government leaders might also form new coalitions, thereby marginalizing groups after the conflict, although they were promised government posts in peace negotiations. Additionally, leaders might manipulate the political balance by redrawing election districts or distributing private goods to regions that are decisive for staying in power. In autocratic systems these kinds of commitment problems might be even worse. Ministers can be expelled, opposition leaders may be killed, and constitutions rewritten, ultimately paving the way for the absolute control of the government leader. Against this background, the question arises how peacekeeping interventions can help to solve these commitment problems. We argue that third parties seeking to intervene and to mitigate commitment problems have to focus on two different, albeit related issues. A peacekeeping force either has to reduce the probability of a prospective power shift or, second, to increase the costs of exploiting it. However, third parties are only able to credibly address these issues, if the benefits of mitigating possible commitment problems outweigh the costs of doing
10 so. More formally, peacekeeping has to fulfill the following basic condition in order to yield credibility (see also Lake and Rothchild 1996): EU (~ Enforce) < EU ( Enforce) (1) Put differently, the expected utility of renewed conflict, EU(~Enforce), has to be smaller than the expected utility of peace, given the peacekeeping effort of some third parties to punish the government or to effectively control rebel organizations. Furthermore, we treat EU(Enforce), i.e., the expected utility of enforcing the terms of a settlement, as being determined by a probability (p) of keeping peace (Ω) minus any direct or indirect costs (c) that are associated with an intervention: EU ( Enforce) = pω c (2) In a third step, we further assume that p is a contest success function (see Skarperdas 1996), which is determined by the forces (F P ) invested by peacekeepers as well as the resistance of peace spoilers (F S ): e FP p = (3) F + F Here, e is the effectiveness parameter of the peacekeepers. When combining (3) with (2) and (1), we see that, ceteris paribus, higher levels of the peacekeeping effectiveness e induce that EU(Enforce) increases as well. Hence, it is then also more likely that peacekeepers come across as more credible, since EU(Enforce) is larger than EU(~Enforce). However, not all peacekeeping forces are created equal, but it is likely that these vary in terms of their credibility and, hence, effectiveness. Put differently, we need to answer what determines e, i.e., the effectiveness of a peacekeeping mission? In what follows, we suggest that it is necessary to focus on the underlying factors that drive a peacekeeping force and that e e P S
11 depends on the number of actors involved in a peacekeeping intervention, their regime types, and supply-side incentives. Size. The literature generally argues that a large coalition of states is more likely than smaller peacekeeping forces to secure a peace agreement and to increase effectiveness (see Walter 2002; Wright and Greig 2008). First, Touval and Zartman (1985: 40), for example, emphasize that third parties only gain the warring parties consent to intervene if they are likely help the antagonists out of a predicament, and for this they usually need leverage. Ward (2006: 153) also argues that states must have the ability to threaten or to bribe other countries, i.e., using leverage, in order to ensure cooperation. The combined leverage of a large peacekeeping force can then create obligations and expectations that help enforce compliance with a peaceful agreement. A small peacekeeping force does often not have such an amount of leverage that may be necessary for effectively securing peace after conflicts. Second, since non-cooperating actors, i.e., antagonists that are reluctant to comply with a ceasefire and the states of the intervening coalition, can be excluded from other forms of reciprocity, larger peacekeeping forces increase the likelihood that both interveners and the warring parties will pursue shared goals through complementary means. Third, states as part of a peacekeeping coalition are likely to have connected issues and interests. This establishes trust, which is a crucial element for overcoming collective action problems and being ultimately more effective in keeping peace. Forth, third parties can signal their commitment through mission size (Walter 1997; 2002). Larger missions serve as a signal of credibility due to of sunken costs: only parties that are credibly committed to a peace process would be willing to pay high costs. Finally, larger peacekeeping forces usually have a higher rate of communication, which facilitates coordination and decreases uncertainty (see Frazier and Dixon 2006: 391ff; Dorussen and Ward 2008).
12 Figure 1. The Impact of Leverage on Peacekeeping Effectiveness. Peacekeeping Effectiveness Leverage However, it is likely that there are decreasing returns with respect to higher levels of peacekeeping effectiveness arising from more leverage and, hence, larger forces. In other words, we argue that a very large coalition of peacekeepers is only to a certain extent more effective than smaller groups. 2 Intervening states vary in terms of their interests, intelligence, capabilities, communication flows, etc. The larger a coalition of peacekeepers, the higher the likelihood of unanticipated problems of coordination, which ultimately may cause that a sufficiently large coalition does not contribute significantly more than a smaller set of interveners to effective outcomes. Figure 1 depicts this relationship. Related to, albeit different from the last claim is the mechanism arguing that larger groups providing a public good in this case peace are more difficult and more costly to organize (see Buchanan and Tullock 1962; Olson 1965: 46ff; Chamberlain 1974). Organizational costs increase with group size in two different ways. First, the larger a group of states forming a peacekeeping force, the more likely it is that the intervening coalition is characterized by greater heterogeneity of interests (Snidal 1994). This in turn makes it more difficult for the peacekeepers to reach any agreement at all or something beyond a minimal consensus with regards to the strategy of their 2 Note that this argument may well explain why Walter (2002) does not find empirical support for her claim on the linear relationship between capabilities and peacekeeping effectiveness.
13 operation. Second, states can participate in peacekeeping efforts without contributing much or anything at all, but obtaining heightened international prestige after an intervention ended successfully (see Walter 1997; 2000). Beardsley (2010) argues in a similar way when suggesting a collective action problem in which the presence of multiple third parties with the potential to offer mediation reduces the likelihood that any mediation will take place at all. Under these circumstances, enforcement through monitoring and/or sanctioning is essential in order to prevent states from free-riding on other states peacekeeping efforts (Fearon 1998: 270). Monitoring procedures ensure that contributions become perceptible for the entire peacekeeping coalition. In addition, only the prospect to detect non-contributions to an effort of peacekeeping allows states to sanction defectors accordingly (see Axelrod and Keohane 1985: 235; North 1990: 57; Olson 1965: 45f). However, with an increasing number of peacekeepers, the more costly it is to organize monitoring devices that provide perceptibility of individual contributions and enforcing actors compliance also becomes less likely to succeed (see Axelrod and Keohane 1985: 234ff; Frohlich and Oppenheimer 1974; North 1990: 57; Olson 1965: 36). Figure 2. The Impact of Organizational Costs on Peacekeeping Effectiveness. Peacekeeping Effectiveness Organizational Costs Figure 2 illustrates this mechanism. Organizational costs hardly matter when a peacekeeping force comprises a few states only. At some point, however, the impact of organizational costs is
14 likely to decrease the chances of cooperative success or the level of peacekeeping effectiveness as in our study rapidly. When combing this curve with Figure 1 above, we obtain an inverted U- shaped relationship between the size of an intervening group and peacekeeping effectiveness as a result. We therefore postulate: Hypothesis 1. The size of an intervening group and peacekeeping effectiveness are characterized by an inverted U-shaped relationship. Both small and very large peacekeeping forces are less likely to intervene effectively. Peacekeepers Regime Types. Drawing upon the democratic peace literature (e.g., Oneal and Russett 1999), democracies have a common structural background and institutional procedures. This gives these kinds of regimes more inclusivity, a higher level of transparency, as well as promotes a common understanding that induces shared norms and perceptions. One of the most important norms of democracies is the peaceful resolution of conflicts (e.g., Dixon 1994; Raymond 1994; Bercovitch and Houston 2000; Mitchell 2002). This should make democratic belligerents not only more likely to settle their disputes peacefully, but also have an impact on the efforts of peacekeepers that are (largely) democratic. Democratic interveners all share, practice, and understand concepts such as limited governments, respect for judicial practices, as well as constitutional constraints (Simmons 1999). Due to their previous, long-lasting practice and experience with these issues they will be more likely to be effective in intervening disputes than (largely) autocratic coalitions, which are not familiar with them. Two related and similar arguments are worth noting here. First, Dixon (1994) claims that democracies will be more effective when intervening, since they operate under a norm of bounded competition that favors the use of compromise. When participating in peacekeeping
15 operations, democracies seek to follow this principle. Second, a peacekeeping coalition of states that is (largely) democratic will be more effective, as it credibly commits not only to these principles of intervention, but also to actively enforcing peace after an agreement and is therefore more effective in convincingly signaling belligerents to settle their conflict even in the absence of strong incentives for doing so (Gilady and Russett 2002: 404; see Werner and Yuen 2005). Finally, the inclusivity and transparency of democratic systems allows enhanced information flows between these sorts of regimes, making them more effective in keeping and securing peace due to two reasons. First, increased information about each other ultimately leads to less uncertainty, more trust, and higher reliability (Mitchell, Kadera and Crescenzi 2005: 8). This is crucial against the background that conflict management, and hence peacekeeping, depends on the credible and truthful transmission of information (Dorussen and Ward 2008: 193). On the opposite, promises of autocracies are less likely to be credible, since there is an insecure hold of power and the absence of an heir (Olson 1993: 571). Second, increased flows of information allow democracies to coordinate their efforts and resources more effectively (Choi 2003: 145). Unlike autocracies, democracies tend to have transparent, open systems, which increases the likelihood that democracies know about each others internal evaluations, their intentions, the intensity of their preferences, and their willingness to adhere to an agreement even in adverse future circumstances (Keohane 1983: 162f). To recap, democratic peacekeeping coalitions are characterized by high levels of communication, shared norms, and inclusivity, which should facilitate coordinative efforts between these kinds of states. This in turn is likely to increase the rate of success of peacekeeping interventions. Our second hypothesis therefore claims:
16 Hypothesis 2. The more democratic a peacekeeping force, the more effective the intervention attempt. Supply-Side Incentives. Countries that comprise a peacekeeping force must have sufficient and credible interests at stake when intervening in order to successfully keep peace (Touval and Zartman 1985: 258f; Lake and Rothchild 1996; Walter 1997: 340; 2000). States have interests, if they share perceptions and norms, or have a common understanding with the belligerents. The colonial ties between Belgium and the Democratic Republic of the Congo, for instance, can explain to a significant extent the involvement of the former in most of the peacekeeping missions there. Most importantly, however, countries seeking to intervene are driven by strategic and security-related interests. Gleditsch, Salehyan, and Schultz (2008) theoretically argue and empirically demonstrate that civil wars increase the likelihood of interstate wars in/with the neighborhood of a conflict state. Among other causal mechanisms presented in their research, Gleditsch et al. (2008: 486f) show that civil wars induce negative externalities that affect the decision-making calculus of geographically close states. More specifically, civil war countries may initiate military action against neighboring states in order to undertake cross-border counterinsurgency actions (see also Salehyan 2007a) or to engage in retaliatory attacks against interventionist neighbors in the hope of coercing those into withdrawing support (Schultz 2007). This in turn increases the likelihood of conflict between the civil war country and one or more neighboring states significantly. As a result, the latter do have strong incentives for preventing not only the spread of conflict, since this is always an inefficient activity (Fearon 1995), but also to address the source of this potential negative externality, i.e., the civil war as such. In other words, civil wars increase the likelihood that surrounding states have a valid security interest in
17 the conflict, which ultimately should increase their incentives to effectively contain civil war and to promote peace. In a similar, albeit different manner, Gleditsch et al. (2008) highlight spillover effects (see also Sandler 2004). Civil wars can create new tensions because of spillovers from the fighting, such as mass refugee migration, or damage of cross-border infrastructure (Gleditsch et al. 2008: 487). Especially the former consequence of civil wars poses severe challenges to neighboring states, since refugee flows are likely to be an economic burden and/or cause political instability (Sandler 2004: 201; Salehyan and Gleditsch 2006). Likewise, Murdoch and Sandler (2002a; 2002b; 2004) show that both the income per capita and economic growth severely drop in countries that are in proximity to those states experiencing a civil war. Eventually, those spillovers can become militarized when states close their borders or take other measures to prevent conflict spillover transmissions. In sum, peacekeepers must have incentives to effectively keep peace in a conflict state (Walter 1997; 2000). If these incentives are not present, then we would expect that the chances of a peacekeeping force to ensure a stable and sustaining peace drop crucially. However, supply-side incentives are most likely to be given if states have a relevant security interest in a conflict most prominently due to negative externalities and spillover effects, since these would directly harm a country neighboring a civil war state (Salehyan 2007b). In turn, supply-side incentives should make parties more willing to bear the costs and risks and remain committed to peacekeeping. Hence, we test the following claim in our last hypothesis: Hypothesis 3. The stronger a peacekeeping force s supply-side incentives, the more effective the intervention attempt.
18 Research Design Data. We employ a modified version of Fortna s (2004) data on peace spells in or after civil wars in order to empirically test our hypotheses. These data are based upon the collection efforts of Doyle and Sambanis (2000) and cover civil wars in the post World War II era until the year 2000. A civil war is defined as as an armed conflict that caused more than 1,000 total battle deaths; that represented a challenge to the sovereignty of an internationally recognized state; and occurred within the recognized boundary of that state; that involved the state as one of the principal combatants; and in which the rebels were able to mount an organized military opposition to the state and to inflict significant casualties on the state (Fortna 2004: 276; see also Doyle and Sambanis 2000). Furthermore, we follow the consensus in the literature and define peacekeeping as the deployment of international personnel under the coordination of the UN or any other organization/coalition/actor to help maintain peace and security (see Fortna and Howard 2008: 285). Fortna s (2004) data suffer from a few shortcomings, such as omitting some short-lived ceasefires, but note that this should not affect our results negatively. On the contrary, the exclusion of these ceasefires is likely to underestimate the impact of our covariates, since more comprehensive data would find the stabilizing effects of peacekeeping to be even greater (Fortna 2004: 276). Using Fortna (2004) as our basic framework, we then modified the data by collecting additional information on our core variables of interest, which are described in detail below. The unit of analysis, i.e., a ceasefire or the peace spell of a civil war, does not alter, and we believe that relying on these data provides two crucial advantages over other kinds of civil war data. First, our data sample has been drawn from the population of civil wars since 1948 and not from the population of peacekeeping efforts. This gives us leverage for comparing peacekeeping
19 interventions and those cases that did not see such a conflict management attempt. This also mitigates the impact of selection bias (Regan and Stam 2000: 240). Second, Fortna s (2004) data include all kinds of peacekeeping interventions in civil wars, i.e., enforcement, monitoring, traditional peacekeeping, and state building. This ensures that we cover the comprehensive spectrum of peacekeeping operations, increasing the generalizability of our findings. Dependent Variable. In order to measure the effectiveness of peacekeeping interventions, we follow Fortna (2004) and employ the duration of peace, i.e., the time elapsed between the termination of fighting and the start of another war, if any, between the same parties. In total, we obtain 389 observations that cover 130 civil wars in 1948-2000. Out of these 389 observations, 47 experienced a failure. Our dependent variable ranges between 1 and 19,131 days, while its mean is at 4,668 days. As our dependent variable measures the length of peace, a regular duration-type estimator seems the appropriate choice for testing our hypotheses. However, recall that assessing the effectiveness of peacekeeping efforts that we observe must also consider the prior selection stage of intervention, or the question of what influences the willingness of third parties to intervene in civil wars. Both stages are intertwined and the civil wars that actually see peacekeeping are unlikely to be a random sample (Gilligan and Sergenti 2006). On one hand, we may surmise that peacekeepers are more likely to intervene in civil wars that are easier. However, it may also be the case that peacekeepers primarily intervene in the more difficult wars, as these cases pose higher risks of instability to the international order. If the first scenario applies, we may exaggerate the effectiveness of peacekeeping, while the latter may lead us to underestimate it.
20 More generally, previous research ignoring the role of selection may have yielded misleading results on the effectiveness of peacekeeping (King and Zeng 2007a; 2007b). To control for this non-random assignment, we employ a duration estimator with sample selection (Boehmke et al. 2006). The correction of this estimator mirrors those for other selection models (see Heckman 1979) by developing a maximum likelihood estimator that simultaneously explains the selection and duration processes. Furthermore, this model estimates the correlation between the error terms in the two equations, allowing us to test for the presence of non-random sample selection. 3 We use a Weibull distribution for the duration stage, while the first stage follows a probit setup and employs the onset of peacekeeping in civil wars as the treatment item. Explanatory Variables. The core variables of interest follow the lines of our theoretical argumentation. 4 First, we run models where we include a count item measuring the number of states that formed a peacekeeping intervention. We collected the information on the size of peacekeeping operations using the Military Balance statistics from the International Institute for Strategic Studies (1965-2000), and information from the United Nations Peacekeeping Department as well as the Secretary-General s reports to the Security Council. 5 This variable ranges in the interval [1; 43]. We include its squared term in the models to test our hypothesis on the inverted U-shaped relationship between coalition size and peacekeeping effectiveness. 3 Boehmke et al. (2006) use a bivariate exponential distribution to implement their Weibull estimator. One limitation of that is that the ρ coefficient estimating the error correlation is restricted to ρ smaller or equal to 0.25. However Boehmke et al. (2006) argue that even if ρ is outside this range, their estimator may provide a better approximation to the unbiased relationships than completely ignoring selection problems. We also run their estimator with a lognormal instead of a Weibull distribution, as ρ is not restricted for this former case. Generally, our results stay robust, but ρ becomes fairly large (0.8) indeed. 4 Note that we also replicate two models from Fortna (2004) in the next section. These models essentially incorporate the same set of variables as described here with one exception. Fortna (2004: 285) employs a naïve approach in order to disaggregate peacekeeping interventions. More precisely, she dichotomously categorizes missions as 1) observer, 2) traditional peacekeeping, 3) multidimensional peacekeeping, and 4) enforcement efforts, using no peacekeeping as a baseline category. In some of the following models, hence, we use the same set of disaggregation variables in order to demonstrate that this categorization does not seem to be sufficient. 5 Related documents can be found at http://www.un.org/en/peacekeeping/ and http://www.un.org/documents/ repsc.htm
21 Second, the overall level of democracy of an intervening coalition is measured by the average value of all intervening countries polity2 value from the Polity IV data (Marshall and Jaggers 2002), which ranges from -6.38 (almost full autocracy) to +10 (full democracy) in our sample. The average polity score is calculated by weighting each peacekeeping country s individual polity2 value using their number of troops contributed to a mission. If only one state forms a peacekeeping force, the polity2 score is employed without any further modifications. We expect this item to be positively related to the duration of peace. Third, for capturing the supply-side incentives of peacekeepers, we rely on the notion that in particular externalities and spillover effects may drive security concerns (Gleditsch et al. 2008). Based upon that, we generated variables that measure the capital-capital distance between an intervener and the civil war country (see Pearson 1974), and then calculated an average score in order to obtain the final variable for each peacekeeping operation. This item is again weighted using a country s specific troop contribution and we originally obtained the data via EUGene (Bennett and Stam 2000). Thus, we claim that the smaller the geographical distance between peacekeepers and belligerents as measured by this variable, the higher the propensity for longer peace spells. Control Covariates. Although we are primarily interested in disaggregating peacekeeping interventions and what impact this may have on the duration of peace, we also have to control for other factors that influence the dependent variable and our core independent variables in order to avoid potential omitted variable bias. Also, including determinants of peacekeeping effectiveness that other studies have identified may further help to control for selection effects. Finally, we need the control items for identifying the influences of peacekeeping onset, i.e., the dependent variable in the first stage of our empirical models. We closely follow Doyle and Sambanis s
22 (2000) and Fortna s (2004: 277) rationale as well as research design in order to decide on the controls. More specifically, we consider for the selection stage, first, a dummy variable coding the outcome of a war. Victory captures whether the fighting ended with a decisive victory by one side. Second, we include an item measuring whether a peace treaty was signed (Treaty). Third, another dichotomous variable (Wartype) distinguishes ethnic, religious, and identity conflicts from ideological, revolutionary, or other wars. Forth, we also include a variable pertaining to the costs of a war. Intensity is measured by the natural log of the number of battle as well as civilian deaths in the context of a civil war. Fifth, including a duration item corrects in addition to our other strategies for possible temporal dependencies and selection effects (see Regan and Stam 2000). Duration measures the length of a war in months. Sixth, a larger number of parties involved in a civil war increases uncertainty and, hence, we incorporate a binary variable (Factions) that marks whether the war involved more than two factions. Seventh, the level of development of a civil war state is coded with a proxy based on per capita electricity consumption (Development), and a country s natural resource dependence is measured via Exports, i.e., the primary commodity exports as a percent of GDP. Eighth, as elaborated above, democracies have a common structural background and institutional procedures, which increases the chances that these states are more favorable towards a stable peace. We therefore consider the prior history of democracy in a civil war country as measured by the average polity2 score over the five years before the war (Prior Democracy). Finally, we include a variable on the size of the government s army in a civil war (Army). Since Fortna (2004) uses the same set of variables not only when explaining which civil wars eventually see peacekeeping efforts, but also for examining the influences of peace duration, we
23 additionally run models in which we include these control variables in the outcome equation of the following empirical models. Empirical Findings We start with replicating some of Fortna s (2004) models that we employ as baseline estimations. Taking the existing selection effects into account, we then expand her analysis by trying to replicate her findings via the selection-duration model (Boehmke et al. 2006). This allows us to compare Fortna s (2004) and our disaggregation approach more precisely. Finally, we substitute Fortna s (2004) mission categories for our variables on the disaggregated peacekeepers characteristics, showing that our theoretical arguments about the effect of size, regime type, and supply-side incentives are largely supported by the empirical findings. In more detail, Fortna (2004: 279) estimates one model for the post World War II period for determining where peacekeepers intervene. The logit estimation uses a dummy variable for peacekeeping onset as the dependent variable and the controls we outlined above as explanatory items. In order to unveil the impact of peacekeeping, Fortna (2004: 285) continues with a Cox model, where the duration of peace after a ceasefire has been established signifies the dependent variable. The same set of controls is used again, but peacekeeping per se is disaggregated via a naïve approach, i.e., along the categories of observer, traditional, multinational, and enforcement missions. The replication estimates are listed in Table 1.
24 Table 1. Replication Models of Fortna (2004). PK Onset Peace Duration (Logit Model 1) (Cox Model 2) Observer 0.20 (0.13)** Traditional Peacekeeping 0.46 (0.29)* Multidimensional Peacekeeping 0.72 (0.46)* Enforcement Mission 1.88 (1.26) Victory -3.53 0.09 (1.01)*** (0.07)*** Treaty -1.04 0.35 (1.06) (0.25) Wartype 0.48 1.32 (0.42) (0.53) Intensity 0.07 1.24 (0.16) (0.12)** Duration -0.002 0.996 (0.003) (0.002) Factions 0.48 0.64 (0.55) (0.25) Exports 1.45 20.77 (3.77) (36.63)* Development 0.0006 0.999 (0.0003)* (0.0004)** Prior Democracy -0.04 0.96 (0.06) (0.03) Government Army Size -0.003** 0.999 (0.001) (0.0004)* Constant 0.73 (1.95) N 110 357 Log Pseudo Likelihood -44.05-165.94 Wald χ 2 34.38*** 113.97*** Pseudo R 2 0.39 ----- Note: table entries are coefficients (logit model) or odds ratios (Cox model); clustered robust standard errors in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1% (two-tailed). The logit model suggests that peacekeepers are more likely to intervene in more developed countries with an internal conflict and in civil war states where the government is weak. However, the most robust impact stems from Victory: peacekeepers tend to stay out of countries where conflict has ended in a decisive victory for either side. 6 As Table 1 further shows, 6 Recall that we use the logit s exact model specifications for the selection equation in the selection-duration estimator. Hence, a precise identification of that stage is a relatively strong requirement for doing so. This is given
25 we were also able to replicate Fortna s (2004: 285) duration model, where she distinguishes between different categories of peacekeeping and finds that peacekeeping along mission types does prolong peace. Note, however, that this model emphasizes at the same time that it is difficult to differentiate between the impact of each specific mission type: regardless if a peacekeeping force acts as an observer, executes traditional peacekeeping, is engaged in peacebuilding, or seeks to enforce peace, the effect on the duration of peace seems to be the same, as the size of the hazard ratios only marginally differs from each other. For the remainder of the analysis, we would like to stress two points accordingly. First and as outlined above, we appreciate Fortna s (2004) insights on the importance of distinguishing between different types of peacekeeping missions, although the replicated models demonstrate that her approach seems not entirely sufficient. Second, and following our theory, we need to go further by a) taking selection effects more directly into account and b) approaching the disaggregation of peacekeeping missions from a different perspective. As a first step toward addressing these issues, we continue with Table 2 that replicates the previous models using the selection-duration estimator (Boehmke et al. 2006). As indicated before, the selection stage corresponds to Fortna s (2004) logit estimations, while the duration stage firstly considers the different peacekeeping mission types only (Model 3) and then proceeds with incorporating the other determinants of peace duration (Model 4). Both model estimations leave out observer missions, using it as the reference category. here, however. The replicated logit model has a satisfactory fit and correctly classifies ca. 83% of the cases (69% in the null model).
26 Table 2. Replication Models of Fortna (2004) using a Selection-Duration Estimator. Model 3 Model 4 Outcome Eq. Selection Eq. Outcome Eq. Selection Eq. Observer - baseline - - baseline - Traditional Peacekeeping 0.14 0.04 (0.43) (0.43) Multidimensional Peacekeeping -0.34-0.15 (0.37) (0.38) Enforcement Mission 0.01-0.05 (0.31) (0.53) Victory -2.88-0.79-2.99 (0.65)*** (0.37)** (0.67)*** Treaty -1.11-0.81-1.25 (0.67)* (0.35)** (0.69)* Wartype 0.42 0.05 0.42 (0.27) (0.29) (0.28) Intensity -0.004 0.08 0.002 (0.07) (0.08) (0.07) Duration -0.001-0.002-0.0007 (0.001) (0.002) (0.002) Factions 0.73-0.35 0.70 (0.31)** (0.33) (0.31)** Exports 0.66-3.57 0.46 (1.43) (2.08)* (1.45) Development 0.0003 0.00004 0.0003 (0.0002) (0.0001) (0.0002) Prior Democracy -0.04-0.06-0.04 (0.03) (0.02)*** (0.04) Government Army Size -0.001 0.0009-0.001 (0.001) (0.0008) (0.001) Constant 7.08 1.26 7.97 1.36 (0.28)*** (0.94) (0.90)*** (0.94) N (Censored) 360 (85) 360 (85) Log Pseudo Likelihood -687.56-675.40 Wald χ 2 54.88*** 56.50*** ρ 0.25*** 0.25*** Note: table entries are probit coefficients (selection equation) or coefficients (outcome equation) pertaining to the accelerated failure time, i.e., larger coefficients correspond to smaller hazards and longer durations; clustered robust standard errors in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1% (two-tailed). Table 2 clearly demonstrates that civil wars with peacekeeping interventions cannot be considered a random sample. The significant estimate of ρ indicates that the selection model fits the data better than independent estimates of logit/probit and duration equations, respectively.
27 The ρ estimates the correlation of the error terms in the two stages. This parameter can be highly sensitive to model specifications, however, and should be expected to be negative if unobserved features that increase the likelihood of selection (i.e., peacekeeping onset) actually decrease the duration of peace after ceasefires. The fact that ρ is positive in all models (including Table 3 below) implies that unobserved features that make peacekeeping missions more likely also induce longer spells of peace. This may perhaps be likened to a positive impact of unobserved forms of resolve among the interveners, as the literature suggests multiple other forms of unobserved influences that could have these effects arising from the characteristics of the disputants and the conflict itself (see Gartzke 1999). Note that our results here are mostly consistent with Table 1, but emphasize again that Fortna s (2004) disaggregation approach does not tell us much when trying to determine which peacekeeping/peacekeepers characteristics are crucial for effectiveness. More specifically and like Fortna (2004), we cannot distinguish between each specific type of peacekeeping. In addition, however, once we directly take into account selection effects, we do not find any significant impact of those peacekeeping categories on the duration of peace after a ceasefire. Model 4 demonstrates that adding controls does not alter this result. Although this is not the focus of this paper, note that with regards to the other determinants of peacekeeping onset and peacekeeping effectiveness we also see crucial differences as compared to the replicated models above, which do not simultaneously consider the two-stage process. The selection equation, for example, shows that Treaty and Factions become significant at conventional levels. The level of development of a civil war country and the size of a government s army play a far minor role now, though. Furthermore, the outcome equation unveils some major differences as well. Among others, although Victory stays robust, Treaty has now a significantly negative impact on the duration of peace. More importantly, Intensity changes its sign and is no longer significant, while