Political Concepts Committee on Concepts and Methods Working Paper Series

Transcription

1 Political Concepts Committee on Concepts and Methods Working Paper Series 38 November 2009 Distinctions without Differences? Comparing Civil and Interstate Wars David E. Cunningham PRIO and Iowa State University Douglas Lemke Pennsylvania State University C&M The Committee on Concepts and Methods IPSA International Political Science Association CIDE Research and Teaching in the Social Sciences

2 Editor Andreas Schedler (CIDE, Mexico City) Editorial Board José Antonio Cheibub, University of Illinois at Urbana-Champaign David Collier, University of California, Berkeley Michael Coppedge, University of Notre Dame John Gerring, Boston University Russell Hardin, New York University Evelyne Huber, University of North Carolina at Chapel Hill James Johnson, University of Rochester Gary King, Harvard University Bernhard Kittel, University of Oldenburg James Mahoney, Brown University Cas Mudde, University of Antwerp Gerardo L. Munck, University of Southern California, Los Angeles Guillermo O Donnell, University of Notre Dame Amy Poteete, Concordia University, Montreal Frederic C. Schaffer, University of Massachusetts, Amherst Ian Shapiro, Yale University Kathleen Thelen, Northwestern University The C&M working paper series are published by the Committee on Concepts and Methods (C&M), the Research Committee No. 1 of the International Political Science Association (IPSA), hosted at CIDE in Mexico City. C&M working papers are meant to share work in progress in a timely way before formal publication. Authors bear full responsibility for the content of their contributions. All rights reserved. The Committee on Concepts and Methods (C&M) promotes conceptual and methodological discussion in political science. It provides a forum of debate between methodological schools who otherwise tend to conduct their deliberations on separate tables. It publishes two series of working papers: Political Concepts and Political Methodology. Political Concepts contains work of excellence on political concepts and political language. It seeks to include innovative contributions to concept analysis, language usage, concept operationalization, and measurement. Political Methodology contains work of excellence on methods and methodology in the study of politics. It invites innovative work on fundamental questions of research design, the construction and evaluation of empirical evidence, theory building and theory testing. The series welcomes, and hopes to foster, contributions that cut across conventional methodological divides, as between quantitative and qualitative methods, or between interpretative and observational approaches. Submissions. All papers are subject to review by either a member of the Editorial Board or an external reviewer. Only English-language papers can be admitted. Authors interested in including their work in the C&M Series may seek initial endorsement by one editorial board member. Alternatively, they may send their paper to wps[at]concepts-methods.org. The C&M webpage offers full access to past working papers.

3 The government of Ethiopia and Eritrean insurgents waged a nearly thirty-year war within Ethiopia beginning in In 1994, Eritrea became an independent state. However, sovereignty only brought a brief respite from warfare as the two sides returned to war from Relations between Ethiopia and Eritrea remain tense. The Republic of Vietnam was wracked by conflict in the 1940s through 1970s. From , France fought a war against anti-colonial groups across Indochina. Vietnam emerged as two independent states, and the new state of South Vietnam faced an immediate communist insurgency backed by forces in North Vietnam. By the middle of the 1960s, the conflict between the two Vietnams had heightened to the point of all-out inter-state war, a war which continued until North Vietnam won and unified the country in The violence between Ethiopia and Eritrea and that over whether Vietnam would be one independent communist state each lasted more than three decades and cost tens of thousands of lives. Because of their duration and consequences, political scientists should be interested in understanding them, and indeed they are frequently used as case studies. However, in quantitative conflict research these conflicts are generally not examined holistically, rather they are split into different types of conflict, housed in different datasets, and analyzed in separate statistical tests. The Correlates of War project includes these periods of conflict in different datasets of wars of different types a period of intrastate (or civil) and a period of interstate conflict in Ethiopia, and a period of extra-state intrastate and interstate warfare in the case of Vietnam. These cases are divided because one of the standard distinctions made between conflicts is whether they contain zero, one, or two recognized members of the international system

4 This division, however, is potentially problematic. Quantitative analyses of civil war analyze a war onset in Vietnam in 1955, despite the presence of a decade of conflict prior to that year (with COW data the Vietnamese Civil War does not start until 1960, and thus is completely separate from the extra-state COW war between France and Vietnamese colonial subjects that ended in 1954). Statistical tests of inter-state war identify a war onset between North and South Vietnam in 1965 and between Ethiopia and Eritrea in 1998, despite the decades of conflict preceding these. Studies of war duration are potentially more problematic, because the thirty-year conflict in Vietnam becomes three separate decade-long conflicts. These divisions occur because interstate and civil wars are analyzed separately in quantitative conflict research. Separate study is so pervasive that a consensus of opinion divides interstate and civil wars into two distinct types of conflicts. But, no one has ever demonstrated conclusively that interstate and civil wars are distinct types requiring separate analyses. Further, the consensus division of conflict studies obscures the fact that separation of interstate and civil war bodies of work is of relatively recent provenance. Pre-Correlates of War data collections (Sorokin 1937; Wright 1942; Richardson 1960) all reported both interstate and civil wars as one single category of violent conflict. No distinction was made in any of these early compilations to suggest interstate and civil war are distinct types of conflict. The question is thus worth asking: are interstate and civil wars distinct types of war? 1 It may be that the division of labor and the resulting consensus that interstate and civil wars are distinct types of conflicts is caused by nothing more than the fact that different types of 1 An unfortunate omission from our paper is detailed consideration of the potentially even-more-problematic distinction among interstate wars into interstate versus extra-state, where the latter are wars waged between a recognized member of the international system and another political entity that is neither part of the state nor recognized as an independent state on its own. While we do control for extra-state wars below, our comparisons are really between civil and interstate conflicts

5 data are available for interstate versus civil war. Specifically, since interstate wars are, by definition, waged between two recognized members of the interstate system, we have information about both sides of every interstate war. Also, we have information about all potential participants in interstate wars, since data about official states are widely available. Consequently, the standard research design for analysis of interstate war focuses on dyads of states, and predicts which dyads will experience wars or disputes. In contrast, scholars of civil war usually lack any information at all about the rebel side of the conflicts they study. Particularly missing is systematic information about the government s opposition absent the occurrence of civil war. Consequently, the standard research design for analysis of civil war focuses on states, and predicts which states will experience civil war, or recurrent civil war, etc. That interstate war studies tend to be dyadic and civil war studies tend to be national is a fact. However, that this fact has produced the consensus belief that civil and interstate wars are different types is only speculation. The purpose of this paper is to see whether there is some solid empirical justification for why civil and interstate wars are studied in isolation from each other. We wish to be clear: neither of us knows whether civil and interstate wars are comparable or distinct. We are not asserting that all conflict scholars should study both. Rather, we are suggesting that whether interstate and civil wars are comparable or distinct should be based on theory and evidence. But so far the division of labor is based on neither theory nor evidence. In this paper we undertake a number of steps designed to probe the civil interstate war distinction. First, we review existing conflict literature for hints about why scholars think civil and interstate wars are distinct. This literature review produces a list of characteristics that allegedly render civil and interstate wars distinct. Problematically, this list can also be used to make arbitrary distinctions within the interstate war type. We show that if we make such - 3 -

6 distinctions, that is, if we use these characteristics to define distinct types of interstate wars, wellknown regularities about interstate war onset disappear. This suggests that making distinctions between types of wars simply for the purpose of making distinctions, produces results which would be rejected by the community of interstate war researchers. This first empirical section thus is intended as a criticism of arbitrary, empirically-driven distinctions. It shows that if we start slicing wars into types of wars, we had better have a good reason for doing so because those slices determine the empirical results we get. Our second empirical section shows the potential benefits of thinking theoretically about relationships and combining types of wars. Building on bargaining arguments common to both war literatures, we show that in a combined dataset of civil and interstate wars variables representing the problems of committing to peace agreements (the number of actors involved, whether the conflict is a recurrent war, etc) have substantial influences on duration and whether there is a negotiated settlement. In fact, the inclusion of these variables reduces the influence of an indicator of whether each war was civil or interstate. Thus, we are able to show that controlling for conditions that complicate bargaining, the distinction between civil and interstate war is considerably mitigated. Our two empirical sections complement each other. The first demonstrates that arbitrary distinctions can generate nonsensical results. Our second empirical section demonstrates that when we include as variables concepts central to general theories about conflict, the distinction between civil and interstate wars becomes less meaningful. Having amassed evidence about the perniciousness of arbitrary distinctions, and about the power of theoretical arguments applied to both types of wars, we conclude the paper with our thoughts about a future research agenda uniting analyses of civil and interstate wars

7 Existing Claims about Civil and Interstate War It is by no means the case that all scholars advocate separate study of civil from interstate war. In fact, much of our thinking about the potential of combining civil with interstate war research is suggested by thoughtful scholars who themselves question the prevailing tendency to study civil and interstate wars separately. Some existing work questions whether our theories are sufficiently powerful to justify separating civil from interstate wars, while others question whether the precise distinctions we make when coding a conflict as civil or interstate are valid. An example of the former is offered by Wagner: the causes of civil and interstate wars tend to be studied separately. However, the justification for this division of labor is not as clear as it is commonly believed to be. (1993:235). Similarly, Lake writes: (u)ltimately, the differences between interstate and intrastate war may be found and recognized as important. But, we should not presume that such differences are large or profound or that one form of violence is wholly distinct from another. (2003:88) The hosts of the COW war datasets complain that artificial divisions of scholarly attention have made it difficult to focus on overall trends in warfare and interrelations among the types of war. (Sarkees, Wayman and Singer 2003:68). Sambanis goes further in cautioning against separate study of types of wars: (i)f civil wars are not unique, however, then by analyzing them in isolation we may be getting inefficient (or even biased) statistical estimates, because we are arbitrarily restricting our analysis to a subsample of the data. (2004a:261) Practical objections against separate analyses are frequently found in published work as well. Sambanis (2004b:815) suggests that (a)lthough a core set of ideal cases of civil war may exist, too many cases are sufficiently ambiguous to make coding the start and end of the war - 5 -

8 problematic and to question the strict categorization of an event of political violence as a civil war as opposed to an act of terrorism, a coup, genocide, organized crime, or international war. Similarly, and echoing our opening paragraphs, Salehyan (2008:55) points out that (i)nter- and intrastate conflicts are often intertwined in complex ways, and a neat dichotomy between the two is often misleading. While many scholars recognize that distinctions between civil and interstate wars might be unnecessary and even misleading, it is certainly the case that the consensus among scholars is that civil and interstate wars are distinct types of conflicts, best studied separately. In total we find nine asserted distinctions between civil and interstate wars in existing literature. By far the most common claim about how civil and interstate wars differ is the observation that, at least since World War II, civil wars are more frequent, more deadly, and are more likely to recur than are interstate wars. So many civil war papers begin with a justification to this effect that a list of references would be too cumbersome to produce. But a sample of articles that begin with all or some of these claims includes: Mason and Fett (1996), Walter (1997), Henderson and Singer (2000), Fearon and Laitin (2003), Collier and Hoeffler (2004), and Lacina (2006). These are among the most influential articles about various aspects of civil wars, and while there are strong disagreements among these authors, they are united in their claims that civil wars differ from interstate wars in terms of frequency, severity, and the likelihood of recurrence. A fourth widely cited difference between civil and interstate wars concerns their durations: civil wars tend to be long while interstate wars tend to be much shorter. Collier et al., (2003:93) report that the average civil war lasts seven years, but the average interstate war lasts only six months. Other scholars emphasizing the difference in duration as an important - 6 -

9 distinction between civil and interstate wars include Balch-Lindsay and Enterline (2000), Regan (2002), and Collier et al. (2004). A fifth difference between civil and interstate wars is an allegation that the former are fought between adversaries very different in power, while the latter are more often fought between roughly equal opponents. Eckstein (1965:158) asserted in internal wars, unlike international wars, there is generally a great disparity in capacity for military effort between the incumbents and insurgents. The former tend to be in a much stronger position not always of course but more often than not. More recently, Fearon suggests that the very measures of capabilities must be different in civil wars than in interstate wars: The empirical obstacles to testing the impact of relative capabilities on civil war duration are also great, since governments and guerrillas deploy such different capabilities that it is difficult to know how to measure the balance. (2004:298) Contrary to Fearon s pessimistic expectations, Cunningham, Gleditsch and Salehyan (2009) measure relative capabilities of governments and rebels, and estimate the influence of the dyadic balance of power on civil war duration. A sixth alleged difference between civil and interstate wars concerns where they are fought. By definition civil wars occur within one state. This is, in fact, part of the standard definition of a civil war. Small and Singer (1982:211) write: One basic property of an internal conflict is obviously that it is internal to a state; that is, significant military action occurs between subjects within the boundaries of the Metropole. The Uppsala/PRIO Armed Conflict dataset does not require that the war be fought within one state, claiming: Internal armed conflict occurs between the government of a state and internal opposition groups without intervention from other states. (Gleditsch et al. 2002:619) While nothing in this more recent definition - 7 -

10 requires that civil wars be waged only within one state, Collier et al. (2003:11) claim civil differs from interstate war because it is fought entirely within the territory of the society. A seventh asserted difference also focuses on the physical process of war. Kalyvas (2006) asserts that in civil wars civilians are disproportionately likely to be victimized while in interstate wars it is soldiers that do the suffering and dying (2006:54). Similarly, Kalyvas alleges that (a) striking empirical observation is that very few civil wars are fought by means of conventional warfare Conversely, almost all interstate wars are fought conventionally. (2006:83). Thus, civil wars differ from interstate in the tactics employed: conventional armed struggle in interstate but irregular tactics in civil wars, and presumably these result in a higher civilian death toll in civil wars than in interstate wars. An eighth reported difference between civil and interstate wars concerns the stakes over which the sides fight. DeRouen and Sobek (2004:303) write: (c)ivil wars fundamentally differ from interstate wars defeat could mean the loss of existence. The high stakes generally make compromise difficult. Fortna (2004a:287) expresses similar thoughts when she writes: (p)erhaps the necessity of living in close contact with those responsible for the killing in civil wars accounts for the difference between intrastate and interstate wars. Losing an interstate war is seen as less threatening than is losing a civil war, because something approximating retreat is an option in interstate wars but not in civil wars. The growing literature on connections between civil war and episodes of mass killing is consistent with claims that the stakes may be higher in civil than in interstate wars (Valentino 2004; Valentino, Huth and Balch-Lindsay 2004). A ninth and final difference between civil and interstate wars frequently found in the literature concerns disparate ways the two types of war end. Walter (1997:335) begins her influential article with the observation that Unlike interstate wars, civil wars rarely end in - 8 -

11 negotiated settlements. She cites a variety of earlier scholars who report the same difference between the two types of war (Modelski 1964; Pillar 1983; Stedman 1991, and Licklider 1995). This difference between the types of war is cited in most work following on Walter s analysis (e.g. Collier et al. 2003:147). Scholars claim to have identified a lot of differences between civil and interstate wars. They tell us that civil wars are more frequent, more deadly and are more likely to recur than are interstate wars. Civil wars are generally a lot longer in duration than are interstate wars. In civil wars it is usually the case that the two sides are very different in power capabilities, while in interstate wars the opponents are more closely matched. Civil wars, largely by definition, are constrained geographically to occur disproportionately within the territory of one state, while interstate wars are more likely to spill over borders. Civil wars are more likely to involve irregular tactics and as a result, more civilians die than is the rule in the conventional warfare characterizing interstate conflicts. Several scholars point out that the stakes of civil wars are higher than they are for combatants in interstate wars, because in interstate wars retreat is a much more viable option while in civil wars the losing side faces an existential threat. Finally, civil and interstate wars differ in how they end, with the former very unlikely to end by negotiated settlement, while the latter are usually resolved via some sort of negotiations. We do not doubt that some of these differences exist. But we are skeptical that the existence of these differences is sufficient justification for asserting that civil and interstate wars are separate types of conflict. In the next section we demonstrate empirically the consequences of making similar distinctions among wars within the interstate war category

12 Exploring Differences Among Interstate Wars Many of the distinctions presented above can also be drawn within the population of interstate wars. For example, some interstate wars have tragically high fatalities (World War II tops the COW project s list with 16 million dead) while others have relatively low fatalities (the Falklands War anchors COW s low end with an unexplainable 910 battle fatalities). If the high average death toll in civil wars is a valid justification for studying them separately from interstate wars, then it is equally sound to study high fatality interstate wars separately from low fatality interstate wars. Drawing out the list of distinctions, it would similarly be appropriate to study recurrent, longer, geographically-constrained, guerrilla wars between unequal combatants separately from non-repeating, short, widely-dispersed, conventional wars among equals. Doing so, however, produces results contradicting some of the strongest findings about interstate war. In order to demonstrate the perverse inferential consequences that result from using observed differences among wars to analyze them separately, we begin with an all-dyads dataset covering the years 1816 to During this time period there were 124 war initiations among dyads according to the COW project (with war initiation occurring when a dyad s members are listed on opposite sides of a war on the first day of the war). The most influential study of war onset (Bremer 1992) demonstrates that these war onsets are more likely to occur when dyads are composed of neighbors that are not grossly unequal in power, are both major powers, are not allied, are non-democratic, underdeveloped, and are highly militarized. Given how influential Bremer s study is, we treat his findings as the conventional wisdom about interstate war onset. Not all of the factors claimed to render civil and interstate wars distinct from each other are easy to measure for a demonstration like this. Yet, of the nine presented above six are reasonably straightforward. We thus offer analysis of war onset in six subsets of interstate wars

13 Each analysis has the same number of observations, but how many of them, and which ones of them, are coded as war onsets varies from analysis to analysis. We compare the results from these six analyses with the results from a normal analysis that combines all war onsets. A first distinction we can make is between high and low fatality interstate wars. The average fatality level in COW interstate wars is 402,016 dead. But only two wars are above that average (WWI and WWII). Since a subset of cases with only two wars, and particularly those two wars, is unappealing for estimation purposes, we instead use as our low/hi fatality cut-point a figure of 85,017 dead (this is the average in COW interstate wars excluding the two world wars). There are 21 high fatality war onsets according to this definition. The large literature on interstate rivalry (e.g. Diehl and Goertz 2000) is motivated by the observation that while rivalry is rare, rivals nevertheless account for the majority of wars and disputes. This is only possible because conflicts between rivals are so much more likely to recur than are conflicts among non-rivals. Consequently, we define wars among rivals as more likely to recur than are wars among non-rivals, and use this as our second distinction among interstate wars. There are 51 onsets of wars among rivals employing Diehl and Goertz s (2000: ) definition of enduring rivalry. The duration distinction, that civil wars should be studied separately from interstate wars because they last so much longer, is easy to apply among interstate wars. The average duration of a COW interstate war is 427 days (the longest is the post-1965 internationalized part of the Vietnam War [3735 days] and the shortest is the 1969 Football War between Honduras and El Salvador [5 days]). Identifying COW wars of 428 or more days as long wars, produces 24 war onsets

14 A fourth arbitrary distinction we make among interstate wars divides those among grossly-unequal adversaries from those among more evenly-matched opponents. A 10:1 ratio is widely regarded as representing overwhelming preponderance (Weede 1976; Bremer 1992). We use it to separate out interstate wars among unequals, finding 24 instances in which war originator dyads were characterized by overwhelming preponderance. The last two arbitrary distinctions we make concern where interstate wars were fought and how they were fought. No COW dataset indicates which specific states experienced conflict physically within their territory, or indicates whether guerrilla tactics were employed in the war. Lacking such information, we combed historical summaries of each COW war in military encyclopedias (Clodfelter 2008; Dupuy and Dupuy 1993) to gather information about these alleged characteristics unique to civil wars. It turns out that a near majority (59) of COW interstate wars were waged within only one state, and similarly that a surprisingly high number (24) of COW interstate wars included guerrilla combat. High fatalities, frequent recurrence, long duration, vast inequality among belligerents, geographic constraints, and guerrilla tactics are six of the nine distinctions researchers have used to justify studying civil wars separately. These same distinctions can be made among interstate wars. If interstate war researchers used the same arguments to justify which wars they studied, they would produce something like Table 1: *****Table 1 About Here***** The first column of Table 1 reports a replication and extension of Bremer s 1992 study. He analyzed all originator war onsets in COW interstate wars over the time frame, employing an all-dyads/all-years case selection procedure. He included seven predictor variables, all of which were significant influences on the probability of war onset. We include

15 the same variables as in Bremer s canonical study, measured exactly the same way. The only difference between our analysis in column 1 and his is that our temporal domain extends 35 years past his. Neighbors is a dummy equal to one when the dyad members are contiguous by land or up to 150 miles of water. Preponderance is a dummy equal to one when the dyad members relative power ratio (stronger to weaker) is at least 10:1. Major Powers is the number of dyad members qualifying as Major Powers in that year. Allies is a dummy equal to one if the dyad members are aligned in any type of COW military alliance. Joint Democracy is a dummy equal to one if both dyad members Polity III Democ-Autoc score is greater than five. Development is the number of dyad members whose share of COW economic capabilities (iron/steel production and energy consumption) is greater than their share of demographic capabilities (total population and urban population). Militarization is the number of dyad members whose share of COW military capabilities (number of troops and total military expenditures) is greater than their share of demographic capabilities. In Dangerous Dyads, Bremer reports results substantively identical to those we report in column 1 of Table 1. Neighbors, Major Powers, and Militarization make war onset more likely, while Preponderance, Allies, Joint Democracy, and Development make war onset less likely. So far, so good. We replicate the conventional wisdom about interstate war onset. However, when we turn to analysis of war onset constrained by characteristics allegedly important enough to justify separate analysis of civil wars, we find the conventional wisdom falls apart. The most important difference across columns 2 through 7, compared to column 1, is that in every single one of the subsets of wars, there is no democratic peace. There is not a single subset of interstate wars in which a significant Joint Democracy effect is uncovered (see the lower highlighted row in Table 1). The democratic peace is clearly the most robust, widely

16 recognized, politically-influential finding about interstate war to emerge in the last two decades of quantitative research on the subject. It would have gone undiscovered if the selection criteria guiding civil war researchers were generalized to interstate war researchers. That is an outrageous inferential cost to pay for adopting an arbitrary distinction as a case selection rule. In addition to the elimination of the democratic peace, interstate war researchers adopting civil war research practices would not have any consistent findings about the pacifying influence of overwhelming preponderance. In their comprehensive summary of quantitative research on interstate war, Geller and Singer (1998:75-76) conclude that the three strongest empirical patterns in interstate war onset are the Democratic Peace, the strong positive association between contiguity and war onset, and the strong negative relationship between preponderance and war mirrored by the strong positive relationship between parity and war. As can be seen in the upper highlighted row in Table 1, in only two subsets of interstate wars (Rival Wars and Long Wars) does Preponderance significantly reduce the likelihood of war. It appears to do so in the Unequal Wars subset, but given that our measure of preponderance and of Wars Among Unequals both employ the 10:1 ratio, it is necessarily the case that all war onsets in this subset would have a value of 1 on the Preponderance independent variable. To avoid this perfect identification, we replace the Preponderance dummy variable with the simple power ratio of the weaker to the stronger dyad members. Given conventional wisdom, this variable should increase the likelihood of war (since lower values of it represent overwhelming preponderance). Instead, we find a negative, that is, wrong signed, significant coefficient for power ratio in the fifth column of Table 1. This completely contradicts the pacifying influences of Preponderance in Rival and Long Wars. The remaining columns of Table 1 (for Hi Fatality, One State, and Guerrilla Wars) only further muddy the waters, as in all of them the influence of Preponderance

17 is insignificant. Consequently, were Interstate War scholars to follow Civil War research practices and isolate analysis of wars based on the various criteria here, some would find that Preponderance is pacifying, some that it increases the risk of war, and others that it has no influence on war onset. Another of the three strongest findings about war onset would be lost. Of the other variables, only Neighbors, the dichotomous contiguity measure, is consistently related to war onset across all of the subsets. This means that if we regard Table 1 s first column as representing the conventional wisdom, and think of columns 2 through 7 as the range of findings interstate war researchers would have generated had they separated interstate wars by types, then six-sevenths of the conventional wisdom would have been lost. Perhaps most importantly, the two-sevenths most unambiguously lost are two of the most widelyaccepted findings about interstate war onset. If interstate wars are separated along the same criteria used to isolate civil war studies, what we know about interstate war onset is almost entirely lost. Theory Driven Combinations of War Types The above analysis demonstrates that using arbitrary justifications to divide conflicts can yield nonsensical results. However, what about theoretical justifications for dividing the two? Maybe civil and inter-state war are not just different because of separate empirical patterns, but these empirical patterns are in fact driven by theoretical differences between them. If so, then it is particularly interesting to note that despite the general consensus on dividing these wars, the dominant theoretical approaches to war do not generally differentiate between the two. That is, both the literature on inter-state and on civil war tend to focus on the same factors the ability to make credible commitments, the divisibility of issues, and

18 information asymmetries as critical determinants of where wars occur, how long they last, and so on. While many of the formal models that analyze the determinants of war theoretically look at the interaction between State A and State B (see, for example, Fearon 1995; Wagner 2000; Slantchev 2003), similar insights have been applied to war between a state and a non-state actor. In fact, Wagner s (2007) bargaining and war model is much more general than any of these and looks at the interaction between two predatory rulers, coming to similar conclusions as do models which look at two states. If both civil and inter-state wars result from common factors such as commitment problems, issue indivisibility, and information asymmetries why are they divided into separate types and analyzed in separate studies? The most common justification seems to be that these barriers to bargaining are more acute in intra-state conflicts. Walter (1997, 2002), for example, argues that commitment problems present particularly high barriers to negotiated settlement in civil wars because at least one side will have to disarm in the implementation phase of the agreement, which makes them vulnerable. Toft (2003) suggests that issues in ethnic conflicts can become indivisible when states are concerned with the precedent of giving up territory and geographically concentrated ethnic minorities see independence as mandatory. Another common argument is that informational asymmetries are greater in civil war, because in inter-state conflict the military capabilities of each side are generally known in advance, while in civil war rebels and governments are often both uncertain about the extent of mobilization the insurgents will be able to achieve. Finally, Cunningham (2006) argues that many civil wars are quite long because they contain multiple combatants and that multiparty bargaining makes reaching an agreement to end war difficult

19 It may very well be true that conflicts within states on average generate greater commitment problems and information asymmetries, and are fought over less divisible issues than those between states. In fact, if we are right that these are the main determinants of warfare the fact that civil wars are much more prevalent than inter-state wars in the post-world War II era would suggest this to be true. However, even if these problems are more extreme in civil wars, we argue that this is really a justification for better specified models, i.e., those that include more theoretically motivated independent variables, rather than for dividing these wars and analyzing them separately. Take the argument about commitment problems as an example. Walter (1997, 2002) finds strong support for her prediction that civil wars are more likely to end in an implemented negotiated settlement when a third-party offers a credible guarantee to enforce the agreement than when they do not. The logic of the argument rests on the difficulty that combatants have trusting the other side when they have to come together in the aftermath of the conflict to form one government and one military. Her analysis is limited to civil wars because it is typically in those types of conflicts that combatants end up sharing the same state. However, this is not true by definition. Rather, in some civil wars (particularly separatist conflicts) it is possible that each side will be able to retain its own government and army, and in those cases an external guarantor may not be needed. Some inter-state wars, likewise, lead to the creation of one state (such as the Wars of Italian Unification or the Franco-Prussian War), and presumably in those cases issues of trust in the disarmament and government creation phase are likely to be particularly difficult to resolve. This suggests, then, that the ability of combatants to trust each other varies across conflicts. There is theoretical reason to believe, in fact, that many interstate wars present many

20 of the same barriers to trust that civil wars do. Fortna s (2008) recent book concludes convincingly that peacekeeping works in civil wars, but her earlier book (Fortna 2004b) demonstrated that it also prolongs peace in interstate wars. Many of the interstate wars occur between states with great difficulty trusting each other (Israel-Syria, India-Pakistan) even if neither side anticipates sharing the same government in the near future. The argument that information asymmetries are higher in civil wars is also not a sufficient justification for splitting these categories. While information asymmetries may be more acute on average in civil war, there is likely to be large variance in both civil and inter-state wars. This variance could be caused by a number of factors. States and/or non-state actors are likely to be more certain of their opponents capabilities if they have fought before so recurrent civil wars might actually have less acute informational problems than first time inter-state conflicts. We also know that many inter-state wars take place between enduring rivals and so interstate wars may generally be shorter because they are often fought between states with high levels of information about each other. Finally, while Cunningham (2006) focuses his analysis on civil war, there is no reason that a similar logic should not apply to interstate war. He argues that conflicts containing more veto players are more difficult to resolve through negotiation and therefore last longer. In order for combatants in conflict to be veto players they need to meet three conditions they have to have separate preferences over the outcome of the conflict, they need to be militarily strong enough to continue the war unilaterally in the face of agreement by the other parties, and they need to be cohesive enough to stick to a coherent set of demands. States in interstate wars almost always meet the last criteria, and generally the military viability one as well (except in extremely asymmetric conflicts). The main difference between interstate and civil wars may be

21 that in many multilateral interstate wars, states do not have separate preferences over the outcome of the conflict (particularly when they fight as alliance partners), but in cases where they do the logic should work the same. What each of these arguments suggests is that there is some underlying factor that causes the barriers to bargaining such as commitment problems, information asymmetries, or issue divisibility that is then causing warfare. A proper empirical analysis, we believe, would then try to measure that underlying factor and examine how it impacts, for example, the onset, duration, severity, or outcome of warfare across a set of cases without limiting the analysis to one or another sub-set of cases. An analogy to the type of argument that we suggest would be the distinction between differences between democracies and non-democracies. Scholars of comparative politics have pointed out a variety of differences that commonly occur between governments which are popularly elected and beholden to the rule of law and those which are not. Democracies often have higher audience costs than non-democracy, democratic leaders are generally more constrained by legislatures than their autocratic counterparts, democratic societies may have more internalized norms of compromise and peaceful resolution of conflict, and these differences are likely to affect outcomes in which we are interested. Increasingly, however, scholars have looked beyond the crude democracy/nondemocracy distinction and have attempted to see which of these factors affect outcomes political scientists care about. In a prominent book, for example, Bueno de Mesquita et al. (2003) argue that the critical difference between political systems is driven by the size of two groups the selectorate and the winning coalition and their relation to each other. They use measures of these concepts to examine variation in economic growth, war, the survival of leaders, and so

22 on. For them, then, democracy and non-democracy may represent only crude proxies for what really are the critical distinctions between different regimes. We are interested in determining whether we can accomplish something similar that is, can we identify the causes of conflict onset, duration, severity, outcome, etc. that are common across civil and inter-state wars. If we are able to do so, we may be able to move beyond statements such as civil wars are unlikely to be resolved without external guarantees, and instead say something like conflicts in which combatants have high levels of mistrust are unlikely to be resolved without external guarantees. Before we proceed to the empirical analysis, one more analogy might help to illustrate what we are doing. In macroeconomic studies of economic growth, it is common for a simple dichotomous measure of whether the country is in Africa to be highly statistically significant with a large negative coefficient. However, as researchers we are generally interested in understanding why African countries have worse economic performance than non-african countries. As such, there is a lot of worked aimed at making the Africa dummy insignificant by identifying the underlying factors that cause the poorer performance. We are interested in doing the same with an all conflicts analysis in which we include a dichotomous measure of whether a conflict is a civil war. We know that, at least in the contemporary era, civil conflicts are more frequent, longer, (generally) more intense, and so on than inter-state wars. Our goal is to identify the dynamics of conflict that lead to those patterns. A Combined Analysis of War Duration and Outcome In this empirical analysis, rather than conducting separate quantitative tests of a sample of civil and interstate wars, we examine a combined dataset and treat the civil/interstate war

23 distinction as an independent variable. We conduct two sets of analyses, the first a study of war duration and the second of war outcome. First, we describe the creation of the dataset that allows us to conduct these analyses. As any quantitative conflict scholar knows, there are multiple datasets available that identify conflicts. 2 The two most prominent datasets are the Correlates of War (COW) and the PRIO/Uppsala Armed Conflict Dataset (ACD). The Correlates of War (COW) project has a longer time frame ( ) and uses a higher battle death threshold (1000 per year), while the ACD has more recent data ( ) and uses a lower battle death threshold (25 per year). In this analysis, we use the ACD because there is more available data allowing us to code the independent variables we would like to include. The ACD includes wars of four different types extra-systemic wars, interstate wars, civil wars, and internationalized civil wars. Quantitative studies of civil war generally include the last two categories, although some also include the extra-systemic wars, which are almost all anti-colonial conflicts. We include all conflicts in one analysis. Because we are conducting duration and outcome analyses, the crucial decisions that have to be made are when to mark the beginning and ending of war. The ACD includes data on the start and end dates of episodes of violence, and allows researchers to make decisions about when to code new conflicts. 3 We follow the lead of other studies using this dataset and code a new war as starting following a twelve month period without 25 battle deaths. 2 See Sambanis (2004b) for a description of the various civil war datasets and for a discussion of the consequences of different coding choices in each. 3 The ACD actually includes two start dates, the first when the conflict generated its first fatality and the second when the conflict reached 25 battle deaths. Because the second is the threshold for conflict used by PRIO/Uppsala, we use the second start date

24 While we largely take the set of conflicts in these datasets as they are coded, there are a few exceptions. In both datasets, there are conflicts that are coded as separate not because there is a twelve month break in fighting but rather because the conflict changes type. However, because here we are interested in analyzing the duration and outcome of violent conflict, not specifically of civil or interstate war, we have combined six of these wars. In Angola, Cameroon, and Malaysia, there were anti-colonial conflicts which continued after independence and we have combined these wars. In Vietnam, we have combined an anti-colonial conflict, a subsequent civil war, and the North-South Vietnam interstate war into one long thirty-year war. In Israel, the ACD codes an end to the war in 1964 because the non-state actor changes from Non-PLO groups to the PLO, but we code this as one long conflict. Finally, the ACD divides the current Iraq war into an interstate and a civil conflict, and we have combined them. These coding results give us a dataset with 436 conflicts. We use this dataset to conduct analyses of war duration and outcome. For the duration analysis, the dependent variable is the duration of conflict, in days, between the start of conflict and the period when the conflict had at least a twelve month break in fighting. For the outcome analysis, the unit of analysis is the conflict and the dependent variable is a dichotomous measure of whether the conflict ended in a negotiated settlement. This variable is coded from the Uppsala Conflict Termination Project (Kreutz 2010). The main independent variable in these tests is the type of war. We conduct two sets of analyses on each dataset. The first includes a dichotomous distinction between civil and interstate wars, and excludes the extra-systemic conflicts. The second includes all three types of war, and includes dummy variables of extra-systemic and civil conflicts (with interstate as the reference category). The interest in this analysis is to see if the Civil War variable decreases in

25 size or significance when we identifying underlying factors that drive duration and outcome in both (or all) types of conflict. Duration Analyses It is well established that civil wars last longer than do interstate wars. We want to determine why. To see if we can identify patterns across all types of conflicts, we include three variables into an analysis of the duration of all conflicts in the post-world War II period. First, we include a measure of the number of participants in each conflict. This is a variable that varies by year and includes the number of parties identified as Side A and Side B actors in the ACD. Second, we examine whether recurrent conflicts may be shorter. To measure recurrence, we examine whether the period between the end of the previous war and the current war including the same dyad was no longer than the length of the previous war. Finally, because a disproportionate number of conflicts in the ACD involve Myanmar as an actor, we include a dichotomous variable measuring whether the conflict took place in Myanmar. Table 2 reports the results of a series of Cox regressions. The first model just includes the civil war dummy variable, Models 2-4 add each of these other variables individually, and Model 5 includes all four variables. Models 6-10 replicate these analyses including dichotomous variables measuring whether the conflict was a Civil War or an Extra-systemic War (with interstate wars as the excluded category). These results report hazard ratios. To interpret hazard ratios, remember that the difference from 1 is the percentage change in the probability of a war ending in the period under observation. A hazard ratio of 0.2, for example, indicates that a one unit change in the variable lowers the probability of the war ending by 80 percent, while a hazard

26 ratio of 2 indicates that a one unit change in the variable increases the probability of a war ending by 100 percent. *****Table 2 About Here***** The first thing to notice from Table 2 is that the dichotomous variable measuring whether a conflict is a civil war is always significant and negative, as is the variable measuring extrasystemic wars. Civil and extra-systemic wars are always significantly longer than inter-state wars, regardless of what other variables we include in the model. This confirms the conventional wisdom that civil wars are longer than interstate wars. Table 2 does show, however, that controlling for the number of combatants, whether the conflict is a recurrent war, and whether the conflict takes place in Myanmar diminishes the difference in duration between civil and interstate wars. In Models 2-5 the hazard ratio on the civil war dummy is closer to one than it is in the bivariate Cox regression in Model 1. This is also true when comparing Models 7-10 to Model 6 as well. While our inclusion of theory-driven variables does not render the difference between the duration of civil and interstate wars insignificant, we have identified three factors that explain at least some of this difference. Each of these variables has an odds ratio below one, meaning that conflicts with more parties, recurrent wars, and wars in Myanmar last longer than do (any type of) wars without these three characteristics. The direction of the effect of war recurrence is surprising, and opposite our expectation. However, it may be the result of low intensity conflicts popping in and out of the ACD in the early years and then staying in as they become more intense. To compare the impact of these difference variables, Figure 1 shows survival functions for civil and interstate wars that also vary on the three other variables included in the model. The