Genocide Watch

by Matthew Krain
International Studies Quarterly, September 2005

This study examines the effectiveness of overt military intervention in slowing or stopping the killing during ongoing instances of genocide or politicide. Six alternative hypotheses regarding the potential effects of intervention on genocide/politicide severity are tested in a cross-national longitudinal analysis of all ongoing genodices or politicides from 1995 to 1997. The results suggest that interventions that directly challenge the perpetrator or aid the target of the brutal policy are the only effective type of military responses, increasing the probability that the magnitude of the slaughter can be slowed or stopped. Impartial interventions seem to be ineffective at reducing severity, and interventions to challenge the perpetrator do not make matters worse for the targets of genocide or politicide. The findings are consistent with recent arguments that attempts to prevent or alleviate mass killings should focus on opposing, restraining, or disarming perpetrators and/or removing them from power.

I long for the day when we can say with confidence that, confronted with a new Rwanda or a new Srebrenica, the world would respond effectively.

(Kofi Annan, Speech to the Stockholm International Forum on Preventing Genocide, January 26, 2004)

More than half a century after the adoption of the United Nations Genocide Convention, cases of states intentionally killing their own people continue to mount (Harff, 2003). Genocides and politicides have been a consistent part of the international political landscape since at least the end of World War II, and there is no reason to expect that state-sponsored mass murder will cease to be a problem (Harff, 1992, 2003; Gurr, 1994). The persistence of genocides and politicides, and the savagery of high-profile instances in Bangladesh, Uganda, Cambodia, Rwanda, and Bosnia, among other cases, prompted renewed interest in developing measures to halt the killing in cases of state-sponsored mass murder.

While intervention failures led many national-level policy makers to doubt the effectiveness of overt military interventions in stopping the killing during highly complex state failures (White House, 1994), international norms evolved toward an increasing acceptance of interventions at the expense of state sovereignty. The last three UN Secretaries General have weighed in on this subject. Javier Perez de Cuellar (1991) argued in his final report that "[T]he fact that in diverse situations the United Nations has not been able to prevent atrocities cannot be accepted as an argument, legal or moral, against the necessary corrective action, especially when peace is threatened." In his "mission statement"An Agenda for Peace, Boutros-Ghali (1995) echoed these sentiments. More recently, Kofi Annan (1998) observed that "state frontiers should no longer be seen as watertight protection for war criminals or mass murderers."

The failure of the international community to stop high-profile atrocities in Bosnia and Rwanda, the evolving consensus that atrocities need to be halted even at the expense of state sovereignty, and the debate over action in Kosovo and East Timor, and more recently Darfur, have all highlighted the need for some systematic evidence of intervention's effectiveness in stopping or reducing the severity of mass killings. Yet, despite its obvious importance to decision makers, we have no systematic evidence as yet as to what effects interventions have on the severity of state-sponsored mass murder. Policy makers faced with situations like those in Rwanda or Bosnia, Kosovo or Darfur, are forced to rely on past experience with interventions in other types of internal conflicts, often with disastrous results.1 This study is a step toward a better understanding of the effectiveness of potential responses by the international community to genocides and politicides. Below, I examine the effectiveness of overt military intervention in stopping the killing during ongoing instances of state-sponsored mass murder.2

Not all intrastate conflicts are the same. Unfortunately, the literature on intervention in internal conflict makes no such fine distinctions. Civil wars have different causal mechanisms than do instances of state-sponsored mass murder. As a result, we might expect intervention to have differential effects on different types of internal conflict. State-sponsored mass murder is a lethal policy carried out by the state against its own people (Krain, 1997).3 This includes the more widely used term genocides as well as the more recent term politicides. Genocides are mass killings in which the victims are defined by association with a particular communal group. Politicides are mass killings in which victims are defined primarily in terms of their hierarchical position or political opposition to the regime and dominant groups (Harff, 1992:28). In both cases, there is intent on the part of the aggressor to destroy the target group "in whole or in part."4

Moreover, much of the recent literature on intervention has focused on its effects on civil war duration. Empirical findings seem to indicate that external interventions tend to result in civil wars of longer duration, but that under particular circumstances, they can shorten the duration of the conflict (Regan, 1996, 2000, 2002; Balch-Lindsay and Enterline, 2000). There is a belief among some policy makers that shortening an internal conflict's duration reduces the severity of that conflict (Holbrooke, 1998). Yet, this may not hold true for all types of internal conflicts. An examination of a few high-profile cases of state-sponsored mass murder reveals the reason why the duration and severity of genocides or politicides are not always related. In a matter of 100 days, almost 800,000 people were slaughtered in Rwanda, a rate of approximately 8,000 per day. Here, the duration of the slaughter was one of the shortest on record; yet, the rate of the killings was nearly unprecedented. Another example is the short but astonishingly brutal killings of somewhere between 1,250,000 and 3,000,000 Bengalis by the Pakistani military in 1971 (Harff and Gurr, 1988). Indeed, duration is not strongly correlated with severity, although it has been found to be a significant factor increasing it (Krain, 1997).

There is also an important theoretical difference between intervention's effects on duration and severity. Interventions can shorten a conflict, but might hasten perpetrators to ramp up their genocidal policy within that period of time. On the other hand, interventions could lead to a hurting stalemate, but also force perpetrators to change policies and curtail mass killings of civilians and non-combatants. Finally, it is possible that interventions might not end a genocide or politicide, but might significantly reduce the severity of the killings. While obviously a less than optimal outcome, surely this is a better result than the status quo during a genocide or politicide. To fully understand the effects of intervention on the lethality of ongoing instances of state-sponsored mass murder, we must examine severity rather than duration.

Moreover, any approach to studying the effects of intervention on the severity of ongoing genocides or politicides must focus on how the intervention affects the behavior of the perpetrator. The study of interventions in ethnic conflict and civil wars focuses on the effects of interventions on the decision-making processes and actions of each side in the conflict because such conflicts are somewhat symmetrical. Scholars of these types of conflicts begin with the assumption that each side has some influence over whether the conflict will continue or abate (Licklider, 1995; Rothchild and Lake, 1998; Regan, 2000, 2002). Genocides and politicides, while also instances of large-scale political violence within state borders, are by definition policy choices made by the perpetrator about murderous action against a given target population. These "final solutions" are strategies devised and implemented by the perpetrator to accomplish their most important objectives, counterthreats to power, and solve their most difficult problems (Valentino, 2004). No action on the part of the target will likely change the decision-making calculus of the perpetrator. Even in cases where civilians are targeted to counter, deter, or drain support from guerrilla opposition (Valentino, Huth, and Balch-Lindsay, 2004), perpetrators are likely to change their murderous policy only in reaction to changes they perceive in the political or military dynamics of the conflict they face (Midlarsky, 2002). Thus, any attempt to understand how intervention might affect the severity of genocides or politicides must focus on the intervention's effect on the perpetrator rather than on multiple sides in a conflict.

If, when evaluating the effects of intervention in an ongoing genocide or politicide, we should focus on the effect on the perpetrator, then the reasons behind the intervention are less relevant. Therefore, I also make the counterintuitive and simplifying assumption that the motivation of the intervener is irrelevant to the outcome of interest here changes in the severity of state-sponsored mass murder. Some actors may intervene specifically to end an ongoing genocide or politicide, but as recent UN history shows, such "noble" or "humanitarian" aims do not always translate into effectiveness (Feil, 1998; Des Forges, 1999; United Nations, 1999; Power, 2002a). Similarly, interventions that successfully curtail mass killing may not have been intended as humanitarian missions by the intervener. Tanzanian intervention in Uganda against the regime of Idi Amin was viewed by the Tanzanian leadership through a lens more akin to realpolitik than to liberal humanitarianism (Stohl, 1987:157). Vietnam's intervention in Cambodia "had a humanitarian consequence but was not motivated by humanitarian concerns" (Power, 2002a:141). Indian intervention in what is now Bangladesh was primarily an effort to counter Pakistani power and to stem the tide of refugees, and only secondarily an altruistic mission to stop state-sponsored mass murder (Kuper, 1985; Charney, 1999:836). Nevertheless, these interventions were successful in reducing the severity of the killings, and eventually ending them, despite less than humanitarian motivations. Thus, I consider intervener motivations less relevant to the study of intervention's effects than how these interventions affect the perpetrators.

Which intervention option is likely to be most effective in slowing or stopping the killing during an ongoing genocide or politicide? Doing nothing, as in the initial reactions to Bosnia and Rwanda, merely allows the killing to continue unabated, and may even escalate it by signaling apathy or consent (Gourevitch, 1998; Des Forges, 1999; Power, 2002a). Neutral interventions do not appear to have much of an ameliorative effect, and might also exacerbate the killing, as the establishment of "safe areas" in Bosnia and Rwanda demonstrated (Power, 2002a). Merely signaling that the world is watching is likely to do little to stop a regime bent on eliminating a domestic group. Aiding the perpetrator does not seem to be a route to reducing the severity of the killing either. Cases such as Uganda, Cambodia, and Bangladesh in the 1970s, and Bosnia and Kosovo in the 1990s, point to what I expect is the optimal solution. Interventions that directly challenge the perpetrator (or that directly provide support to the targets of genocide or politicide) should be most likely to reduce the severity of genocide or politicide.

Kofi Annan argues that "a deliberate and systematic attempt to terrorize, expel or murder an entire people must be met decisively with all necessary meansIn both [Bosnia and Kosovo] it required the use of force to bring a halt to the planned and systematic killing and expulsion of civilians" (United Nations, 1999:111). If state-sponsored mass murder has already begun, the perpetrators likely have already evaluated the international context and decided that there is a degree of permissiveness sufficient to allow them to commit genocide or politicide without consequence (Harff, 1986:168). One reason that they may not have been deterred is because they may view the credibility or resolve of potential interveners as low (Carment and Rowlands, 1998; Rothchild and Lake, 1998). Interventions that directly challenge the perpetrator by acting against them, or for the target, clearly signal the credibility and resolve of interveners.

Moreover, genocide or politicide is typically carried out by the perpetrator against unarmed civilians, even when these include supporters of guerrilla forces (Valentino, 2004; Valentino et al., 2004). If we assume that perpetrators choose to pursue policies of state-sponsored mass murder because their ability to overwhelm civilian targets makes it a "useful" choice (Valentino, 2004), then interventions that support the targets (or oppose the perpetrators) may force perpetrators to change their calculations. Challenging interventions force perpetrators to divert time and resources otherwise dedicated to a policy of genocide or politicide toward defense against the external challenge. Such interventions make state-sponsored mass murder a more difficult project fraught with even more serious military and/or political consequences than had existed previously. This, even if only temporarily, should stem the violence, especially if the genocidal policy is being carried out by "thugs" or "opportunistic bullies" who are unlikely to put up too much of a fight against an outside force (Mueller, 2000). The effect would be akin to throwing a wet blanket over an emerging fire it could prevent the spread of and perhaps even lead to a cessation of the killing by raising the costs to the perpetrator of continuing the policy (Rothchild and Lake, 1998).

This was precisely the reasoning behind General Dallaire's infamously ignored request to the UN for a more substantial military presence in Rwanda in the face of the coming genocide. Simply placing a well-equipped sizable force willing to oppose the perpetrator on the ground might have either prevented such killings or would have kept the killings from escalating (Feil, 1998; Gourevitch, 1998; Power, 2002a). Recent re-evaluations of the situation in Rwanda in 1994 have suggested that Dallaire may have been correct, and that the genocide in Rwanda may have been averted or made less severe by a timely intervention to check potential perpetrators (Feil, 1998; Des Forges, 1999; Power, 2002a).5

Note that this theoretical argument does not assume that interventions change the balance of power between the perpetrator and the target. Rather, the introduction of forces against the perpetrators (1) signals that the international context has changed from permissive to prohibitive, and that the genocidaires no longer remain unchallenged and (2) diverts valuable time and resources from policies of domestic group eradication toward defense against an external challenger. Because neither impartial interventions nor interventions that support the perpetrator address both of these simultaneously, neither are likely to reduce the severity of state-sponsored mass murder.

Despite the arguments and case evidence that challenging perpetrators is most likely to lead to a reduction in the killing, arguments in favor of other policy options persist. I detail these below, as alternative models to the Challenging Intervention Model, and then test which type of intervention best ameliorates the slaughter in ongoing genocides and politicides.

Impartial Intervention Model

Impartiality is the assumption that guides much of the thinking behind the role of international organizations in peacekeeping operations (Diehl, Reifschneider, Hensel, 1996:687 688). Some scholars and policy makers have argued that impartiality is necessary to insure the effectiveness of third-party interveners (Ratner, 1996:51 54). If so, then impartial interventions might have the best chance of decreasing genocide or politicide severity, while interventions in favor of the target are unlikely to reduce the severity of the killing. These are the only kinds of interventions that simultaneously appear legitimate, unbiased, and non-threatening. They also do not affect local power dynamics. If the intervening force can make it clear that stopping the killing, rather than victory for either side in the conflict, is the primary concern, then the assumption is that an impartial intervention should reduce the severity of state-sponsored mass murder.

Witness Model

One argument in favor of intervention in instances of state-sponsored mass murder is that simply placing troops on the ground that do not explicitly favor the perpetrator, whether they are impartial interveners or support for targets, alters the perpetrator's perceptions of the degree of permissiveness of the international context. Bystanders to initial moves toward genocide or politicide signal to perpetrators that there will be no severe consequences at the international level for continuing the killing (Harff, 1986:168; Power, 2002a:503). Perpetrators are not deterred because they view the likelihood of intervention to be minimal (Kuper, 1985; Stohl, 1987; Power, 2002a). When states intervene, they are no longer bystanders, but active participants and potential eyewitnesses to mass murder. Interventions signal at the very least an interest in the situation by the international community, and an unwillingness to be complicit by remaining passively on the sidelines. If perpetrators are deterred from conducting state-sponsored mass murder primarily by perceptions of a change in the degree of permissiveness in the international community, then interventions in ongoing genocides that do not act in support of the perpetrators should lead to a reduction in severity.

However, the Challenging Intervention Model assumes that simple signaling of world interest is not sufficient to deter a regime bent on eliminating a domestic group. Something must also be done to reduce the amount of resources that the perpetrator can expend on a policy of genocide or politicide. Moreover, the Challenging Intervention Model assumes that impartial interventions could be unlikely to reduce severity, and may even exacerbate the situation. Overall intervention salience or commitment by impartial interveners may be rather low (Carment and Rowlands, 1998). This can lead to the use of empty threats, which once exposed, can escalate rather than de-escalate severity (Power, 2002a; Harff, 2003:70). For example, UN failure to defend "safe havens" in Bosnia and Rwanda left a large group of targets unprotected, exposed, and centrally located, facilitating quicker extermination (Betts, 1994:24 25; Feil, 1998; Des Forges, 1999; Luttwak, 1999; Power, 2002a). Kofi Annan himself noted, " the pervasive ambivalence within the United Nations regarding the role of force in the pursuit of peace [and] an institutional ideology of impartiality even when confronted with attempted genocide" are in large part to blame for intervention disasters such as Srebrenica (United Nations, 1999:111). Indeed, impartial interventions are not likely to have an ameliorative effect, and might actually exacerbate genocide or politicide severity in the short to medium term (Betts, 1994). As such, I expect that neither impartial interventions nor simply signaling interest by placing troops on the ground are sufficient to reduce the severity of genocide or politicide.

Bystander Model

It is possible that challenging interventions may not be as effective as expected in reducing genocide or politicide severity. If both neutral and challenging interventions turn out to be ineffective, then choosing not to intervene militarily may be the optimal choice. Indeed, the most frequent choice made by international actors is to do nothing in the face of an ongoing genocide or politicide (Power, 2002a). Why intervene if the decision makers know that the mission is not likely to succeed (Luttwak, 1999)? A better option is to do nothing militarily, and either opt out of the situation or pursue other non-military options such as economic or political sanctions.

Balance of Power Model

It is also possible that my initial assumption that genocides and politicides are fundamentally different from other types of internal conflict is problematic. Perhaps a balance of power argument, demonstrated to have significant explanatory power in cases of civil war termination, might be appropriate here. Such an argument claims that interveners can tip the balance toward their favored side by providing much-needed resources or augmenting capabilities (Rothchild and Lake, 1998). The domestic political arena consists of a number of competing actors, including the perpetrator of genocide/politicide and its targets. By intervening, a foreign power effectively becomes an actor in the domestic arena in its own right. As such, it can form alliances, provide new sources of revenue or arms, stabilize a situation, embolden or enable their ally, or even act in lieu of their ally.6

The literature on interventions in internal wars suggests the utility of a balance of power approach. Balch-Lindsay and Enterline (2000) and Regan (2002) find that biased interventions (those in favor of only one side of the conflict event) shift the balance of power, thereby hastening the end of civil conflict. Similarly, they find that both impartial interventions and balanced biased interventions (interventions on both sides that cancel each other out) lead to a maintenance in the status quo (do not shorten the conflict) at best, and may exacerbate the conflict at worst (Balch-Lindsay and Enterline, 2000; Regan, 2002).

If external interveners provide material, and perhaps moral support, interventions on the side of the perpetrator of the atrocities might increase the severity of genocide or politicide. Perpetrators are already in a dominant position. In most cases, they are the state or are allied with the state, giving them access to the resources of the treasury and a monopoly on the legitimate mechanisms of coercion.7 However, state-sponsored mass murder is a costly policy, both to the perpetrator and to the state as a whole (Midlarsky, 2002:3). Significant resources must be spent or sacrificed (both in the short and long run) in order to carry such a policy out for long periods of time. Hence, a pro-perpetrator intervention should at the very least enable a continuation of the policy by supplementing the resources available to the perpetrator. International supporters of the perpetrator state may even become complicit in genocide or politicide themselves (Midlarsky, 2002). In addition, a regime that can bring home external resources often finds that it benefits from an increased level of legitimacy at home (Gurr, 1988:46). Hence, support from the outside can enhance their position, both in the eyes of their domestic constituents and in terms of relative capabilities. Similarly, an intervener that aids a perpetrator against other opposition groups makes it easier for that perpetrator to divert sufficient resources to be better able to continue its genocidal policy against the target of genocide or politicide. Hence, any intervention that favors the perpetrator, even in a conflict against a group not targeted for genocide or politicide, should have similar benefits for the perpetrator vis-à-vis the target, and therefore should have similar effects on genocide/politicide severity. By the same logic, any intervention that favors the target should decrease genocide/politicide severity.8 Finally, given a balance of power argument, impartial interventions should have no effect on severity. Neither side becomes allied with a new partner, nor do they acquire much-needed resources to change the balance of power in the domestic conflict arena. While the latter two expectations concur with those of the Challenging Intervention Model, the Balance of Power Model differs in that it also assumes that perpetrators yield benefits from additional external resources or alliances that allow them to escalate their murderous policies.

Threat-Based Model

Finally, some might argue that challenging interventions might actually escalate rather than reduce genocide/politicide severity. This approach assumes that genocides or politicides are the product of heightened threat perception by the perpetrator. Interventions may alter this perception of threat, and therefore affect the subsequent use of such murderous policies. Such an approach assumes that genocides and politicides are undertaken by weak or fragile states, or states facing major crises, as an attempt to influence events that affect their own security (Midlarsky, 2002). Perpetrator regimes may feel threatened, and adopt terror tactics to maintain their rule. Over time, these policies can metastasize and become full-scale murderous policies (Stohl, 1987:156; Midlarsky, 2002; Valentino, 2004). In some cases, the target group itself may pose a threat <!--[if !vml]--> political, economic, or social, real or imagined (Melson, 1992:274; Krain, 1997:333; Midlarsky, 2002). Perpetrators might also fear opposition groups that could potentially draw support from the civilian population. In such instances, regimes may target the civilians in an effort to cut into the opposition's support base (Valentino et al., 2004), or to undermine the rebels' ability to claim that it can protect the civilian population at large.9

International intervention against the perpetrator may provide support for the opposition, but it also increases the threat to the regime. Anti-perpetrator interventions may back regimes into a corner, and may "provoke even more vindictive action" against targets of genocide or politicide (McCloy, 1944; see also: Luttwak, 1999:42; "We Have Only Made it Worse ," 1999). Elites do their best to keep opportunities for challenge to the status quo at a minimum. However, openings in the political opportunity structure that favor the opposition often lead to challenges to the elites because they signal the vulnerability of the state (Krain, 1997). Even when this is not the case, elites may perceive themselves to be potentially weakened and vulnerable. Failure to act decisively may threaten the very existence of the regime: a regime that does not consolidate its power and eliminate the opposition risks having its authority called into question, and increases the possibility of other challengers arising from within (Tilly, 1978). Hostile international pressures lead to greater isolation of the elites, which in turn leads to an increased probability that these elites will use large-scale repression. At the extreme, this can lead to brutal suppression of opposition of groups perceived to be threats to the regime (Harff and Gurr, 1988).

Even if the perpetrators themselves do not feel threatened, they can use interventions against them to create the perception of threat among their constituents. In an attempt to garner more internal support, perpetrators might couch escalation in nationalist anti-intervener sentiment.10 As such, challenging external interveners provide a convenient enemy with which to justify the escalation of genocide or politicide campaigns. Ironically, such interventions may also strengthen the very actors that they are meant to weaken ("We Have Only Made it Worse ", 1999). Anti-perpetrator interventions also make it easier to hide the scale of the atrocities from a population by diverting their attention to the international threat.

Thus, we might expect interventions against the government to increase rather than decrease genocide or politicide severity, as they also widen windows of political opportunity within which elites can use state-sponsored mass murder (Fein, 1979; Harff, 1986; Melson, 1992; Krain, 1997). Empirical studies have shown that threatening interventions do tend to exacerbate domestic conflict in general (Pearson, 1974), and wars have been found to have escalatory effects on violations of human rights more generally (Poe and Tate, 1994), and the onset and severity of genocides and politicides more specifically (Krain, 1997). Thus, a Threat-Based Model suggests that intervention against the perpetrator would increase the severity of state-sponsored mass murder.

Interventions that support the perpetrator do not simply increase resources available to the regime. They may also act to reduce the threat levels felt by the regime. If so, one might expect a counterintuitive result that interventions in favor of the government act to reduce state-sponsored slaughter. It is also possible that regimes may still perceive a threat, even after favorable interventions. Yet, these interventions would hardly escalate that threat. Hence, one would expect that at the very least, interventions on the side of the perpetrator would not act to increase severity, even if severity did not decline.

The history of the genocide or politicide itself is likely to have an effect on the magnitude of the severity of genocide or politicide. The longer the murderous policy has been in place, the more time the perpetrator has to learn how to kill more efficiently. The duration of a genocide or politicide has been found to increase its severity significantly (Krain, 1997). Conversely, the longer a perpetrator engages in large-scale slaughter, the fewer potential victims remain. Regardless of the expected direction, it is important to account for how long the genocide or politicide has been ongoing. Moreover, one of the most robust findings in the literature on repression is that previous repression increases the likelihood that repression will be used in the future (Gurr, 1986, 1988; Poe and Tate, 1994; Davenport, 1995, 1999). Whether the reasons lie in habituation of the user to the use of repression, the reduction in costs to the user once the mechanisms of repression are in place, or the bureaucratic inertia that occurs once the specialists in repression are given power (Gurr, 1986, 1988), previous levels are likely to affect current or future levels of slaughter. Therefore, I control for both the Duration of the genocide or politicide and its Previous Magnitude of Severity.

States contiguous to the site of an ongoing genocide or politicide have a higher degree of interest in change and instability in neighboring states than in less proximate locations (Gurr, 1994; Maoz, 1996). States may intervene in other states' affairs to stabilize their own political environment. In the international community, neighbors do not have the luxury of being bystanders.11 Internal conflicts have a tendency to diffuse across borders (Rothchild and Lake, 1998). Neighboring states are likely to want to intervene to stop the killings, if only to insure that refugees do not flee to their own state, and to reduce the chance of conflict contagion by pre-empting an escalating slaughter (Regan, 1998:766). Still others may intervene to assist the perpetrator in eliminating a common threat, and would therefore intervene to help escalate the killing. Regardless of the reason, contiguous states are more likely to intervene, more likely to have a higher commitment to the intervention, and more likely to be successful in their intervention (Regan, 1998:766). While it is unclear what effect to expect, it is clearly important to control for the effects of Contiguity of interveners in instances of genocide or politicide.

Genocides and politicides are almost always a consequence of State Failure, defined as internal wars, adverse or disruptive regime transitions, or other genocides or politicides.12 State failures promote domestic instability, and open windows of opportunity during which murderous policies become more likely (Fein, 1979; Harff, 1986, 2003; Melson, 1992; Krain, 1997). In most cases, internal wars are the first in a complex series of events, often including other destabilizing events such as war, decolonization, or extra-constitutional changes in leadership (Krain, 1997; Harff, 2003). Evidence suggests that the greater the number of state failures experienced, the more severe the instance of state-sponsored mass murder (Krain, 1997). Given the predictive power of this variable, it is important to control for state failures.

Regime Type may also affect the severity of state-sponsored mass murder. A major finding in the international relations literature is that democracies tend not to fight one another. This democratic peace proposition has more recently been extended to internal wars (Krain and Myers, 1997; Hegre, Ellingsen, Gates, and Gleditsch, 2001), human rights (Davenport, 1999), and state-sponsored mass murder (Rummel, 1994). Rummel (1994) finds that democracies have killed substantially fewer of their own citizens than other forms of government. His explanation is that democracies tend to diffuse power, making such large-scale killings less likely. Yet, at least one study has concluded that while high amounts of power concentration may contribute to an environment in which the use of state-sponsored mass murder is more likely, it alone neither determines when in the history of a given state such practices will be used, nor how severe the resulting state-sponsored mass murder will be (Krain, 1997). An alternative hypothesis is that power diffusion (as found in the most liberal democratic states) reduces the likelihood of state failure, which in turn reduces the probability of onset of state-sponsored mass murder (Krain, 2000). Most of the state failures experienced by democracies were neither state-sponsored mass murder nor internal war, but rather "adverse or disruptive regime transitions," or rollback of democracy (Esty, Golstone, Gurr, Harff, Levy, Dabelko, Surko, and Unger, 1998). Autocracies are three and a half times more likely than other types of regimes to have state failures that lead to genocide or politicide (Harff, 2003:66). Regardless of whether the effect is direct or indirect, it is clear that regime type may have some effect on the likelihood of a state's willingness or ability to kill their own people. Therefore, I control for regime type's potential effect on the severity of state-sponsored mass murder in this study.

Many also argue that Ethnic Fractionalization is a key element in understanding state-sponsored mass murder. Groups targeted for elimination are routinely defined by ethnicity. Even in situations of politicide, groups targeted primarily for political reasons also frequently exist as communal groups. Groups that are marginalized are at the greatest risk (Gurr, 1993). Such groups are more easily identifiable as "different" than the rest of society, and are therefore more likely to be targeted for scapegoating and dehumanization (Staub, 1989). This in turn makes it more likely that large groups in the majority population will support the murderous policy (in the name of ethnic purity, nationalism, security, etc.), thereby increasing the severity of the killings. Preliminary empirical analyses suggest moderate support for this relationship. Alone, the degree of concentration of ethnic groups has no significant effect on either the onset or severity of state-sponsored killings (Krain, 1997; Harff, 2003; Valentino, 2004). However, when duration is accounted for, ethnic fractionalization appears to have the expected effect: the more ethnically concentrated a society, the greater the severity of a genocide or politicide (Krain, 1997:353). This suggests that it would be wise to control for varying levels of ethnic fractionalization.

Economic Marginalization is one of the prime determinants of large-scale intergroup conflict (Gurr, 1994:359). The weakness of states in the world economic system is often associated with ruthless repression and large-scale killings (Gurr, 1986). Recent evidence suggests that states that are less open to trade, and thereby less connected to the global economic system, are more likely to experience genocide or politicide (Harff, 2003). Yet, economic marginalization is not a significant determinant of severity (Krain, 1997). However, economic status may affect the presence or intensity of international diplomatic and/or economic pressures faced by perpetrators. Stohl (1987) argues that economic interests cloud the judgment of potential interveners within the international system, often creating bystanders. Gurr concurs, arguing, "peripheral status in the world system increases the likelihood that regimes that rule by violence can do so with impunity" (Gurr, 1986:61). Nations that have no interest in economic relations with murderous states have less incentive to intervene to stop the killing (Gurr, 1994:359). Therefore, economically marginalized states, with fewer ties to the international system, receive fewer political pressures to cease and desist murderous policies (Harff, 1986:168). Without such pressures, these states can engage in longer and/or more concentrated periods of mass killing. All of these arguments suggest that economic marginalization should be controlled for in analyses of intervention's effects on genocide/politicide severity, despite evidence that suggests that it should have little or no effect (Krain, 1997; Valentino, 2004).

Finally, it is possible that effects of intervention on genocide or politicide severity are a function of the structure of the international system. Interventions may have had different effects on severity before and after the Cold War as a result of different dynamics in the UN and other international organizations, in polarity, or in the evolving norms of intervention. Therefore, it is important to control for the differences between the Cold War and post-Cold War eras.

Unit of Analysis

My unit of analysis is the country-year. I examine only country-years in which a genocide or politicide is ongoing, plus the year immediately following the end of that instance of state-sponsored mass murder. Since intervention data must be lagged to insure the ability to infer causality through temporal ordering, examining the year after state-sponsored mass murder has concluded allows me to include the effects of overt military interventions in the final year of genocide/politicide. I examine only ongoing cases of state-sponsored mass murder because I am interested in answering questions about the international community's ability to affect cases that have already begun. Overt military intervention tends to be a reactive policy, as humanitarian disasters are rarely pre-empted by military means. Of course, pre-emption is an important theoretical, moral, and policy question, and may be empirically testable. However, given its importance, such an analysis must be carried out with equal care, and in the context of very different theoretical arguments. As a result, I defer this question to a future study.

Dependent Variable: Genocide/Politicide Severity

The State Failure Task Force developed a list of all genocides and politicides from 1955 to 2001 (Esty et al., 1998; Marshall, Gurr, and Harff, 2002).13 This list is an updated version of the data previously compiled by Harff (1992) (see also Harff and Gurr, 1988),14 which has been used as the basis for a number of comparative empirical studies examining genocide and politicide (Fein, 1993; Licklider, 1995; Krain, 1997; Harff, 2003; Valentino et al., 2004). A slightly abbreviated version of the list, reproduced here in Table 1, includes location and start and end years of the instance of state-sponsored mass murder for the years 1955 to 1997.15

The data set also includes information about the magnitude of severity of the genocide or politicide. All data on the severity of state-sponsored mass murder have an inherent flaw the more successful the policy, the less we may know about it (Lopez and Stohl, 1984). In particular, body counts, the usual measures of severity, are problematic. There will be under-reporting of fatalities by the aggressors in most cases in order to hide the extent of the atrocity. There will also be over-reporting of fatalities by the victims in order to highlight the degree of the atrocity (Rummel, 1994). Therefore, no such data can be precise.16 The State Failure Task Force data consist not of exact body counts, but a more approximate scaled magnitude index representing the likely range for the estimated number of deaths per year.17 The original scale, as presented by Marshall et al. (2002), is reproduced here in Table 2. I adopt their severity scale as the dependent variable in this study. Note that the dependent variable is an ordinal categorical variable, suggesting the necessity for an appropriate statistical method. As a result, I used ordered logit as my method of analysis in this study.

Independent Variables: International Military Intervention (IMI) Variables

The Pearson and Baumann (1993) data set, updated by Pickering (1999), codes all overt IMIs from 1946 to 1996.18 In practice, this allows me to examine genocides and politicides occurring through 1997, since all intervention indicators were lagged one year to avoid problems in determining causality. The variables below are all coded using data available in the IMI data set.

Using the IMI data set, I created an Intervention Dummy Variable to measure the presence or absence of any international intervention force that does not support the perpetrator (either against the perpetrator or as an impartial force) in a given country-year. This allows me to test the Witness Model more directly. Interventions that were not pro-perpetrator were coded as "1." Country-years without interventions, or with pro-perpetrator interventions were coded as "0."

The IMI data set records in what direction (or on whose behalf) the intervener acts. I compress a number of IMI categories into three main intervention types: I code as pro-perpetrator interventions all interventions that are explicitly pro-perpetrator, as well as those that are anti-target. I code all interventions that are explicitly anti-perpetrator, as well as those that are pro-target, as anti-perpetrator interventions. Those that support neither side explicitly or are expressly impartial are coded as impartial interventions. I examined all other interventions (including those that support or oppose third-party governments, or those that oppose rebel groups in sanctuary) and recoded them as to whether they indirectly were pro-perpetrator or pro-target. Those that were neither were considered "directionless" and/or not relevant to the domestic political situation, and therefore not included in any of these categories. I was thus able to code the number of interventions for and against the perpetrator, as well as those that were impartial.19

I also created a measure of the Balance of Interventions, coded as the sum of the number of pro-perpetrator interventions minus the sum of the number of anti-perpetrator interventions in each country-year. A positive value indicates that more interventions occurred that favored the perpetrator. A negative value indicates that more interventions occurred that opposed the perpetrator. A value of zero suggests that interventions balanced one another out.

Control Variables

Contiguous countries are coded as states that share a common land border, or those separated by less than or equal to 150 miles of water. The IMI data set records when the source and target of an intervention are contiguous. From this information, I construct a dummy variable that records in a given country-year whether at least one intervener was a contiguous state. Like the rest of the intervention data, this variable, Contiguity of Intervener(s), is lagged 1 year.

In order to capture the effects of the Duration of state-sponsored mass murder on its severity, I code in which year of the genocide or politicide the observation occurs. For example, the first year of the genocide or politicide is coded as "1," the second year as "2," and so on. I also lag the dependent variable by 1 year to enable me to control for the effects of Prior Severity of Genocide or Politicide on current genocide or politicide severity. This helps to control for the effects of autocorrelation (Poe and Tate, 1994; Davenport, 1995).

The State Failure variable is a dummy variable measuring whether the state experiencing genocide or politicide is also experiencing another kind of state failure in that same year. The State Failure Task Force developed a list of all state failures from 1955 to 2001 (Esty et al., 1998; Marshall et al., 2002).20 State failures are defined as all revolutionary wars, ethnic civil wars, genocides, or politicides, and disruptive regime transitions.21 However, since all the cases examined here are ongoing genocides or politicides, this variable would be a constant if it included the ongoing state-sponsored mass murder. Hence, for the purposes of this study, I exclude the ongoing genocides or politicide from the state failure variable.

The Polity IV data set includes a number of measures of Regime Type for all countries from 1800 to 2000 (Marshall and Jaggers, 2002). I use the measure of regime type recommended by Marshall and Jaggers (2002), the Polity IV democracy measure minus the autocracy measure. The democracy and autocracy scores are each ordinally scaled, ranging from zero to ten, each measuring institutional aspects of the regime. The regime type score, therefore, ranges from 10 to 10, with lower scores denoting more autocratic states and higher scores more democratic states. I use the recommended coding rules for transitional states (interpolation) and those in a state of interregnum (set at zero).22

I use Krain's (1997) index to measure Ethnic Fractionalization. This continuous index is calculated as follows: the proportion of the population of each ethnic group to the total population of the country is squared; the squared proportions for all groups are then summed and that number is subtracted from one to come up with the fractionalization measure for that country. A low score indicates asymmetry between groups and/or relative homogeneity. A high score indicates many groups with small or relatively equal percentages of the population.23 This variable was originally coded by Krain (1997) at 10-year intervals: 1948, 1958, 1968, and 1978. A country's score for a given year was then calculated as the average of the two scores between which the year fell. In other words, the score for a country in 1964 would be the average of the scores for 1958 and 1968. For this study, if an instance of genocide or politicide began on or after 1979, I used the formula described above to calculate that country's scores for the nearest year. Information on a country's ethnic makeup in that year was extracted from the Political Handbook of the World and the CIA World Factbook from that year.24

A country's international economic interconnectedness is measured as the degree of Marginalization within the World Economy. This variable is operationalized as a function of that country's percentage of world trade. The percentage of world trade is computed by calculating the total imports plus total exports of a given nation divided by the total imports plus exports of the world economy. The inverse of that figure is then divided by 100, yielding the economic marginalization score. The score is thus the degree of marginalization low scores on this index indicate greater centrality within the world economy; high scores indicate greater marginalization. The data were collected from various years of the International Monetary Fund's Direction of Trade Statistics Yearbook (1998).25

Finally, in order to account for the potential temporal differences caused by changes in the structure of the international system, I created a dummy variable for the Cold War, with the years 1955 1989 coded as "1," and the years 1990 1997 coded as "0."

Below, I present the results of ordered logit models of factors affecting the severity of ongoing instances of state-sponsored mass murder.26 Models are estimated using STATA, version 8.0, and using the White estimator of robust standard errors to correct for heteroskedasticity. White's (1980) estimators of variance are particularly useful when estimating ordered logit models using unbalanced panel data (each panel has a different number of observations because each genocide/politicide lasts a different number of years). White's robust standard errors help to produce estimates that account for the fact that "observations are likely to be independent across countries but not within them" (Davenport, 2004:550; see also Poe and Tate, 1994:859 <!--[if !vml]--> <!--[endif]--> 860).

I used three statistical models to test the six different sets of expectations regarding the effects of intervention on the magnitude of severity (Table 3). All three models use the same set of control variables, but each includes different versions of the intervention variables. The first model includes a dummy variable that records the presence or absence of any international intervention force that does not support the perpetrator (either against the perpetrator or an impartial force). The second model includes measures of the number of interventions for and against the perpetrator, as well as the number of impartial interventions. The third model uses a measure of the overall bias of the balance of interventions (number of pro-perpetrator interventions minus the number of anti-perpetrator interventions), along with the number of impartial interventions.

The results reported in Model #1 suggest that we can reject the Witness Model. The presence of an international intervention force that does not support the perpetrator (either against the perpetrator or as an impartial force) does not in and of itself have a statistically significant effect on the severity of genocide or politicide. Simply signaling that the international context is not "permissive" by having witnesses on the ground may be a necessary condition for slowing or stopping the killing, but it is not sufficient.

We can also reject the Impartial Intervention Model based on the results reported in Model #2. Impartial interventions have no statistically significant effect on the severity of state-sponsored mass murder. This result confirms recent conventional wisdom regarding the ineffectiveness of UN and other impartial interventions,27 but also suggests that they do not do as much harm as has sometimes been argued <!--[if !vml]--> <!--[endif]--> at least not in the medium term, up to a year after the intervention. It is also consistent with much of what we know from other recent studies of impartial interventions, including the finding that UN intervention has no effect on the severity of future conflict, regardless of the level of violence of the conflict (Diehl et al., 1996).

Model #2 does provide support for the Challenging Intervention Model. The number of interventions against the perpetrator, or for the target, has a statistically significant negative effect on the magnitude of severity of genocides and politicides, even given the effects of contiguity, state failure, duration, and the previous year's magnitude of severity. Moreover, as expected, neither interventions that favor the perpetrator nor impartial interventions have a statistically significant effect on severity. Interventions that directly challenge the perpetrator or aid the target appear most likely to reduce the severity of genocide or politicide. Since there is evidence that challenging interventions reduce the severity of genocides or politicides, we can also reject the Bystander Model as based on faulty assumptions about the ineffectiveness of intervention.

The Balance of Power Model also expects the number of interventions supporting the target to have a negative effect on the severity of genocide or politicide. However, the Balance of Power Model predicts that as the number of interventions in favor of the perpetrator increases, the severity of genocide or politicide should increase. Model #2 shows that interventions in favor of the perpetrator have no statistically significant effect on severity. And as Model #3 demonstrates, the balance of interventions has no statistically significant effect. Evidence suggests that the Balance of Power Model, while useful to explain civil war duration (Balch-Lindsay and Enterline, 2000; Regan, 2002), does little to explain the severity of genocides and politicides.

State-sponsored mass murder is a policy founded on the absolute preponderance of the power of perpetrators over highly vulnerable targets (Harff, 1992; Valentino, 2004). Shifts in balance of power are not what alter calculations. Additional support for perpetrators does not make an already lethal policy more lethal. These results suggest that military support for targets or in opposition to perpetrators changes the murderous calculations of perpetrators by altering the almost complete vulnerability of unarmed civilian targets. This confirms yet again that the underlying dynamics of genocides and politicides are significantly different from those of other internal conflicts, and suggests that solutions that successfully slow or halt the killing in civil wars are not likely to work to reduce the severity of genocides or politicides.

Finally, we can reject the Threat-Based Model out of hand as a viable explanation for changes in the severity of state-sponsored mass murder. The model's main expectations do not hold. The number of interventions supporting the target has a negative rather than the expected positive effect on severity, although the number of interventions in favor of the perpetrator has no measurable effect. Model #3 demonstrates that the balance of interventions has no statistically significant effect on severity. Policy maker concerns that intervention on the behalf of target populations will escalate the killing appear to be unfounded.

The only overt military interventions that appear to be effective in reducing the severity of genocides or politicides are those that explicitly challenge the perpetrator of the atrocities. The question remains as to the magnitude of the effect of challenging interventions on genocide/politicide severity. The ordered logit coefficients presented in Model #2 are difficult to interpret on their own. However, one can use them to generate predicted probabilities that are much more intuitive. Therefore, I explore the degree of the effect of such interventions in Tables 4 and 5, examining predicted probabilities of two hypothetical cases of interest.

Table 4 shows the probability that a "typical" post-Cold War era case of genocide or politicide will experience different magnitudes of severity up to a year after a given number of interventions against the perpetrator of the atrocities. The case examined is an "average" case in that most values of the independent and control variables are set at the mean or modal values. The hypothetical case occurs in a relatively ethnically divided (0.53), somewhat economically marginalized (37.02) autocracy ( <!--[if !vml]--> <!--[endif]--> 4.54) that is experiencing another state failure besides the ongoing genocide or politicide. The case is in its sixth year of the genocide/politicide (6.28). The previous year witnessed between 4,000 and 8,000 killed (magnitude=2.13). Neither pro-perpetrator nor neutral interventions have occurred, nor have contiguous actors intervened.

The first column indicates the probabilities predicted for the case if no actor challenges the perpetrator militarily. Given that this case has already experienced a magnitude of killing of about 2.0, absent such an intervention, it appears likely that the killings will escalate. The predicted probability of escalation is 0.6422, or the sum of the predicted probabilities of the case experiencing magnitudes above 2.0. The probability that it will taper off on its own is 0.2836, while the probability that the magnitude of the killing will remain about the same is 0.0743.

Even a single intervention against the perpetrator has a measurable effect on the severity of genocide or politicide in the "typical" case. When a single international actor challenges the perpetrator, the predicted probability that the killings will escalate drops from 0.6422 to 0.5510, while the probability that the killings will decrease jumps from 0.2836 to 0.3664. If two actors challenge the perpetrator, the probability of escalation drops further to 0.4564, while the probability that the killings will abate increases to 0.4580. Three challenging interventions increase the probability of lives saved to 0.5527.

Table 5 depicts simulated effects of interventions against the Sudanese regime in 2003 on severity of state-sponsored mass murder in Darfur in 2004. I chose this case because of the policy debate at the time of this writing as to how to deal with an ongoing case of genocide/politicide. The slaughter in Darfur began in 2003, when an estimated 16,000 <!--[if !vml]--> <!--[endif]--> 32,000 people were killed (magnitude of severity=3.0) by the Sudanese government in Khartoum, in concert with the Janjaweed militia.28 Throughout the crisis, the autocratic government (regime type= <!--[if !vml]--> <!--[endif]--> 7) of Sudan has been in the midst of a civil war (presence of state failure). The state is highly fractionalized ethnically (ethnic fractionalization=0.74), and is relatively marginalized within the international economic system (economic marginalization=37.949, using 2003 data from the IMF's Direction of Trade Statistics Quarterly to estimate 2004 trade data). In 2003, there were no overt military interventions by any international actors, partisan or neutral, contiguous or not, nor have there been any as of this writing (January 2005).

The magnitude of the severity of slaughter in Darfur in 2003 is estimated to have been 3.0. Absent any challenging intervention, the model expects that the killings would have escalated beyond that level in 2004, all else held constant. The predicted probability of escalation of the magnitude of severity is 0.6410. The probability that it would have tapered off on its own in 2004 is 0.2120. The probability that the magnitude of the killing would have remained about the same in 2004 is 0.1470.

Intervention against the Sudanese government and the Janjaweed within the first year of the genocide would likely have had a measurable effect on the severity of state-sponsored mass murder in the following year. Had a single international actor challenged the perpetrator in 2003, at the start of the killings, the predicted probability that the slaughter would have escalated declines from 0.6410 to 0.5499, while the probability that the killings would have been slowed increases from 0.2120 to 0.2823. If two actors had challenged the perpetrator, the probability of escalation would have dropped further to 0.4554, while the probability that civilians in Darfur would have been saved would have increased to 0.3649. Three challenging interventions would have been expected to increase the probability of lives saved in Darfur to 0.4565.

These two hypothetical cases are merely indicative of the type of effect that challenging interventions can have on the likelihood of reducing or stopping the killing in cases of ongoing genocides or politicides. Of course, there is no certainty that international interventions that challenge the perpetrator would reduce the magnitude of severity of the killings. Yet, the evidence presented here suggests that such interventions are the only effective type of military response, and do increase the probability that the magnitude of the slaughter can be slowed or stopped.

This study is the first to systematically examine the role of overt international military interventions in affecting the severity of ongoing instances of state-sponsored mass murder. The results suggest that the most effective way for the international community to intervene militarily to reduce the severity of an ongoing genocide or politicide is to directly challenge the perpetrator or to aid the target of the brutal policy. Perhaps equally important, this study confirms that attempts by external actors to intervene as impartial parties seem to be ineffective. The implications for policy, particularly for international organizations such as the United Nations, are clear. While impartial interveners such as the UN can and should remain integral to military-led humanitarian interventions, their emphasis on impartiality may be best suited to rebuilding and reconciliation efforts after the genocide or politicide has been ended. If actors wish to slow or stop the killing in an ongoing instance of state-sponsored mass murder, they are more likely to be effective if they oppose the perpetrators of the brutal policy.

This study's findings are consistent with recent arguments that attempts to prevent or alleviate mass killings should focus on opposing, restraining, or disarming perpetrators and/or removing them from power (Stanton, 2004; Valentino, 2004). At the very least, this study contributes more evidence to suggest the utility of such a policy over alternatives such as impartial military intervention. Additionally, perhaps these results will convince policy makers that some measures taken against perpetrators are significantly better than the choice to do nothing. As Power notes:

The <!--[if !vml]--> <!--[endif]--> response usually offered to the question of why the major powers did so little to stop genocide is that any intervention would have been futile. Each time states began slaughtering and deporting their citizens, Western officials claimed that the proposed measures would do little to stem the horrors, or that they would do more harm than good <!--[if !vml]--> <!--[endif]--> For all the talk of the futility of foreign involvement, in the rare instances that the United States and its allies took even small steps, they appear to have saved lives.

(Power, 2002b:73)

By finding that increasing the number of interventions against perpetrators reduces severity, this study confirms that international interventions against perpetrators do save lives.

Of course, it is possible that the effects of overt international military interventions found in this study are short lived, and that in the long run such interventions actually affect the overall severity or the duration of the slaughter differently. Two strategies need to be used in order to test such a hypothesis. First, longer time lags need to be introduced into the current analysis of severity. This will also help researchers determine whether any of the variables examined here have different effects in the long and short term. Second, duration analysis of genocides and politicides would now be useful in ascertaining whether these variables act to reduce the time factor rather than yearly severity. Shorter genocides do not necessarily mean less severe ones. However, given the right circumstances, they might. Perhaps interventions act to shorten the time spent killing rather than reduce annual severity. Given our newfound understanding of severity, knowing now about duration would help us paint a more complete picture of the impact of overt international military intervention on the dynamics of genocides and politicides.

It is also possible that the magnitude of military intervention might influence the severity of state-sponsored slaughter. The success of interventions depends in part upon the size of the forces deployed (McDermott, 1998:23). Missions involving large numbers of troops may signal a high commitment level on the part of the intervener, while lower troop counts may call into question the salience of the conflict to the third party (Carment and Rowlands, 1998:588). Future research should examine these potential effects to enable policy makers to calculate the extent of intervention necessary to slow or stop genocidal killing.

It is also important to remember that policy makers have a range of options available to them in the face of an ongoing genocide or politicide. This study only examines one of those options <!--[if !vml]--> <!--[endif]--> military intervention. Given Regan's (1996) findings that mixed (military, diplomatic, and/or economic) strategies are more effective in ameliorating internal conflict than military ones alone,29 future research should extend this analysis to include other types of international action.

Finally, while this study addresses the most effective way of reacting militarily in the face of an ongoing genocide or politicide, the most effective way of reducing the severity of the slaughter is by being proactive. The development of effective early warning systems now enables us to better predict which states and which groups are most likely to fall victim to large-scale atrocities (Gurr, 1993; Esty et al., 1998; Harff, 2003). Yet, even successful identification of a potential genocide or politicide does not necessarily lead to prevention. Policy makers need information regarding the effectiveness of intervention and other policy options in pre-empting state-sponsored mass murder. Future research needs to be directed toward this important goal if policy makers are to be convinced that pre-emption is possible and likely to be effective.

Nevertheless, large-scale atrocities often occur despite early warnings and some measure of preventative action, thus the importance of continued study of the effectiveness of reactive measures in the face of an already ongoing genocide or politicide. The more information about policy options available to decision makers, the more likely that the world community can begin to successfully act upon the as yet unmet challenge of the last century: Never Again!