When have violent civil conflicts spread? Introducing a dataset of substate conflict contagion

Abstract

The spread, diffusion, spillover, or contagion of violent civil conflict – including insurgencies, coups, or other internal armed conflict – across international borders is of great concern to civil war scholars and international security policymakers alike. For instance, great power military interventions are often predicated in part on the belief that if a given conflict is not stopped now, it may spread and destabilize an entire region. Nevertheless, our understanding of this phenomenon of ‘substate conflict contagion’ is hindered by the lack of a comprehensive and accurate universe of cases. In this article I introduce an original dataset of cases and non-cases of substate conflict contagion between 1946 and 2007. The key difference between my dataset and other datasets of this phenomenon is that I require in my definition of contagion not only the spatial and temporal proximity of two conflicts, but also a documented causal link between them. After introducing the dataset and the process by which it was constructed, I show that substate conflict contagion by my definition is significantly less common than previous scholarship and policymaker rhetoric suggest, and that its correlates – and potentially the best methods with which to measure those correlates – are different from prior research as well. Policy implications are considered, and applications of this dataset for future conflict research are explored.

Keywords

civil conflict civil war conflict contagion conflict dataset

Violent civil conflict sometimes spreads across international borders. For example, the conflict in Rwanda spilled over into what was then Zaire in 1996, and the separatist rebellion in Croatia triggered similar instability in Bosnia and Herzegovina in 1992. The fear of the spread of violent civil conflict across borders is frequently invoked as a justification for great power military intervention in conflicts – the logic being that if a given conflict is not stopped now, it may spread and destabilize the entire region. For example, US interventions in Southeast Asia (1950s–1970s), the Balkans (1990s), and Libya (2011) were undertaken in part to prevent conflict from spreading further afield (on Libya, see Obama, 2011).

Yet violent civil conflict does not always spread across borders. For example, the Balkan conflicts never destabilized the rest of Eastern Europe, as some feared at the time (Clinton, 1995). The question is begged, then, under what conditions violent civil conflict will spread across borders, and under what conditions it will remain contained. A better understanding of these contagion effects could potentially help policymakers make more informed decisions about which civil conflicts to intervene in, and which to leave alone. Although concern about contagion is just one of many reasons a state might choose to intervene in another state’s conflict, it is nevertheless an important, frequent, and understudied one.

In order to properly study the conditions under which violent civil conflict spreads across borders, we first need a universe of cases of the spread and non-spread of conflict. To date, a comprehensive and accurate universe of cases has been missing from the small academic literature on this topic, which as I will show has tended to overestimate the frequency of this phenomenon. Accordingly, in this article, I will introduce an original dataset of cases and non-cases of ‘substate conflict contagion’. I define substate conflict contagion as occurring when a ‘substate conflict’ – an armed conflict between a state and militarized rebels that has caused at least 25 cumulative battle-related deaths – makes a causal contribution to the onset of a substate conflict in another state. Between 1946 and 2007, I identify 122 such cases of substate conflict contagion, out of a total dataset of 21,364 directed dyad-year observations containing 616 cases of spatially and temporally clustered conflicts (all but 122 of which, as I discuss below, do not meet the aforementioned definition of contagion). That means that in the vast majority of cases violent civil conflicts do not spread across borders, an observation with significant policy implications discussed below.

In the next section I describe current attempts to identify the occurrence of substate conflict contagion, and two key shortcomings thereof.¹ Then I describe the process that I took to identify a more comprehensive and accurate universe of contagion and non-contagion cases. In the discussion section, I compare both the frequency and the correlates of my definition of substate conflict contagion to prior scholarship, finding markedly different results. I close the article with some suggestions for future research. Finally, the online appendix to this article provides the full dataset of contagion and non-contagion cases.

Existing datasets of substate conflict contagion

Previous work has defined universes of cases of substate conflict contagion, in the process of creating empirical scholarship about the spread of violent civil conflict across borders. This empirical scholarship deserves great credit for advancing knowledge about substate conflict contagion from virtually nothing to the beginnings of an understanding of the phenomenon. Nevertheless, existing definitions of the dependent variable suffer from two key challenges which my dataset aims to overcome. These challenges apply equally to works using directed dyad-years as the unit of observation (Forsberg, 2008, 2009) and to the other works cited below, which use country-years or a similar quantity as the unit of observation and use a dichotomous or ordinal variable to indicate that at least one nearby state is in conflict.

The first challenge is that substate conflict contagion has generally been defined as the occurrence of a conflict onset in a state which is adjacent to a state in conflict (Forsberg, 2008, 2009; Braithwaite, 2010; Kathman, 2010; Beardsley, 2011; de Groot, 2011). Only a few works define contagion as possible between non-contiguous states (Fox, 2004; Buhaug & Gleditsch, 2008). Only focusing on adjacent states underestimates the frequency of contagion in many situations. Although the means by which substate conflict spreads are at least partially geographic in nature, non-geographic factors play a role as well. For example, if it is foreign sponsorship that causes some rebellions, there is no strong reason to expect that sponsorship to move only across land borders. So by limiting the scope to contiguous states, some important cases of contagion – 33, according to my dataset – are omitted.

A second challenge with the existing dependent variable definition in all of these works is that the definition also overestimates contagion. The danger for overestimation stems from the fact that any substate conflict onset in a state adjacent to (or near) a state in conflict is included in the definition. Scholars have not required that the two conflicts in question be causally linked to each other – yet we could observe spatially clustered conflicts because of actual contagion (one conflict contributing to the onset of another), spatial clustering of other explanatory variables that cause conflict, or pure coincidence. Hence Forsberg (2008: 289, 292) identifies 186 cases of ethnic conflict contagion between 1989 and 2004,² only 26 of which I identify (see below) as instances of one conflict influencing the onset of another.

My dataset aims to overcome both of these challenges, first by defining contagion such that it can occur between non-contiguous states, and second by requiring that the first conflict actually have a causal link with the second, subsequent conflict onset.

Identification of cases of substate conflict contagion

I start with a ‘universe of possible cases’, consisting of a set of directed dyad-years between 1946 and 2007. Each dyad contains a State A and a State B. State A is a country experiencing an internal or ‘internationalized internal’ armed conflict, as defined in the Uppsala Conflict Data Program/Peace Research Institute, Oslo Armed Conflict Dataset (Version 4-2009) (Gleditsch et al., 2002). Specifically, this definition of substate conflict requires ‘a contested incompatibility that concerns government and/or territory where the use of armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths’ (UCDP & PRIO, 2009: 1). By convention, conflicts between metropoles and their nonstate colonies – known as ‘extrastate’ or ‘extrasystemic’ conflicts – are excluded from this definition.³ The ‘State A’ component of the dyad-year may also include a state in which a substate conflict ended five or fewer years ago. This allowance is meant to capture any lagged contagion effects.

State B is a country in the same ‘neighborhood’ as State A. In Table A.1 (in the online appendix), I list 19 neighborhoods that I initially defined based on geographic proximity, colonial histories, cultural similarities, and a rough comparison of states’ GDP per capita and democracy (POLITY IV) scores (Marshall, Jaggers & Gurr, 2008). (Note that states can be members of more than one neighborhood; Mexico, for instance, is a member of both North America and Central America.) All states in the world, except the Western European microstates (Luxembourg and smaller), island states with fewer than 500,000 inhabitants in 2007, and Pacific Island states for which there are no relevant State As end up in at least one neighborhood. The final definition of neighborhoods came about through an iterative process, in which individual neighborhoods were expanded when the coding process described below uncovered new networks of state-to-state influence, and collapsed when coding suggested that two states did not have sufficiently deep relations with one another to be considered ‘neighbors’. In total, this universe of possible cases numbers 21,364 directed dyad-years.

Needless to say, this universe would be much larger or much smaller if the definition of ‘neighborhood’ were expanded or contracted. A smaller version of the universe – contiguous states only – is considered in Table II below. As for the larger universe, time constraints prevented the construction of a dataset in which every state in conflict was paired with every other state in the world. But it is improbable that more than a handful of additional contagion cases would be identified by doing so, because my coding method already identifies 40 provisional cases of contagion that occurred outside the strict temporal and geographic scope conditions (more on this below).

The use of directed dyad-years as the unit of observation is relatively new to the study of substate conflict contagion (it was pioneered in Forsberg, 2008).⁴ Most other works have used country-years; the dependent variable in these studies is civil conflict onset and the explanatory variable is a dichotomous indicator of whether one of the state’s neighbors was involved in a civil conflict – or, alternatively, a count of the number of neighbors in conflict. Although I adapt my dataset into an alternate, country-year design below, I prefer the directed dyad-year setup. We are not concerned with substate conflict onset per se, but rather substate conflict onset related to a specific neighboring substate conflict. Many states, particularly in sub-Saharan Africa, have multiple neighbors in conflict at the same time. Thus, in many cases a country-year research design with a ‘neighbor in conflict’ explanatory variable would not permit us to distinguish which neighboring conflicts, if any, have spread to the state in question. The directed dyad-year design is suitable for precisely that purpose.

Next, I winnow the 21,364 observations in the universe of possible cases down to 616 cases of ‘potential contagion’. In this smaller subset of directed dyad-years, a substate conflict began in State B. (In general the ‘onset5’ variant of the Armed Conflict Dataset mentioned above was used to determine when conflicts began, although some exceptions were made due to inaccuracies in this dataset.⁵) Hence I have reason to suspect that the ongoing (or recently ended) substate conflict in State A contributed to the onset of the substate conflict in State B. These 616 potential contagion cases are listed in Table A.2, available in the online appendix.

To stop here would be to commit the same error as that found in much of the existing literature on substate conflict contagion: to equate spatial and temporal clustering of substate conflicts in State A and State B with the spread of conflict from State A to State B. Instead, I researched each of these 616 cases to determine whether it was a case of ‘actual substate conflict contagion’. In the ‘actual’ cases, at least one secondary source that I consulted identified the substate conflict in State A as a cause of the substate conflict onset in State B. State A’s conflict needed to be mentioned as a cause of State B’s conflict; it did not need to be elevated to be the cause. Thus the causal threshold for actual substate conflict contagion is rather low (albeit higher than in previous scholarship); the State B conflict could have been caused by 25 different factors, and I code the presence of actual contagion if State A’s conflict is one of those factors. Likewise, if conflicts in multiple State As contributed to the conflict in State B – for example, the conflicts in Rwanda, Burundi, Uganda, and Angola all contributed to the subsequent conflict in Zaire – each directed dyad-year (i.e. Rwanda–Zaire 1996, Burundi–Zaire 1996, Uganda–Zaire 1996, and Angola–Zaire 1996) is coded as actual contagion. In the remaining cases, by contrast, I find no causal link asserted between State A’s substate conflict and State B’s substate conflict onset. Generally I consulted at least three English-language secondary sources before identifying an absent causal link, and whenever possible I checked these codings with an area expert (usually an author of one of the secondary sources that I read). Full details on my codings are available in the online appendix.⁶

I chose this low threshold for the identification of actual contagion because substate conflict causation is an extraordinarily complex phenomenon. No such conflict has ever been caused by a single factor. Insisting that a State A conflict be the sole cause of a State B conflict, or even that State A conflict be a necessary condition for State B conflict, would have led me to throw out dozens of contagion cases in which State A conflict was an important but not necessarily a determinative factor in the onset of the State B conflict. Defining contagion more broadly – as the influence of State A conflict on the onset of State B conflict – allows me to consider the full spectrum of relationships between pairs of conflicts, while allowing readers to exclude ‘weaker’ cases at their discretion rather than mine. So it is important to note that the identification of a causal link between a State A conflict and a State B conflict should not be interpreted as the assertion that State A’s conflict was necessary for the onset of State B’s conflict.

This coding process yields 122 cases of actual substate conflict contagion (33 between non-contiguous dyads), with contagion defined as one substate conflict contributing to the onset of another. These cases are listed in Table A.3 in the online appendix. A flowchart showing how I arrived at this list – in effect summarizing this section – is shown in Figure 1.

Discussion

In this section, I will make some initial observations about the data that have been created here. There is vast potential for future research using this dataset, some ideas for which will be discussed in the concluding section of this article. In the meantime, although I can only scratch the surface of these data in the space available, some of the most basic generalizations may already have significant implications for both civil conflict scholarship and international security policymakers.

First, there is an enormous variety of ways in which substate conflicts spread. Table I gives just five examples, randomly selected from the population of 122 contagion cases. We see contagion between both contiguous and non-contiguous states (Sierra Leone and Nigeria are not contiguous); we see agency both on the part of state governments (Angola, China, Zaire) and non-state actors (refugees, arms dealers); and we see that the vectors of transmission include refugees, small arms, and ideology. The larger book project that this dataset is a part of seeks to make sense of this cacophony of causes; suffice it to say for now that contagion is a complex and poorly understood phenomenon.

Second, another striking early impression from this substate conflict contagion dataset is that it portrays substate conflict contagion as being far less frequent than prior scholarship suggests. In 21,364 possible cases of substate conflict contagion – listed in full in the online appendix – there were only 122 cases (about 0.57%) in which a substate conflict in State A actually had some causal link to a subsequent substate conflict onset in State B. In Table II, I compare this low apparent frequency of substate conflict contagion to the frequencies estimated in data from four previous articles on the spread of violent civil conflict across borders.

As the Table II shows, using my own dataset and the same temporal scope as each of the authors,⁷ I find

Figure 1.

Flowchart of substate conflict contagion coding methodology

Table I.

Mechanisms involved in five randomly selected substate conflict contagion cases

Case	How contagion occurred
Angola to Zaire, 1996	Zaire was supporting Angolan rebels; Angola retaliated by supporting Zairean rebels.
Sierra Leone to Nigeria, 2004	Weapons moved from the Sierra Leonean conflict into the hands of Nigerian rebels.
Guatemala to Mexico, 1994	Refugees fleeing the conflict in Guatemala strained land resources in Chiapas, contributing to grievances that led to revolt.
China to Laos, 1959	The Chinese Civil War ended in communist victory; Communist China then trained Pathet Lao cadres.
China to Burma, 1949	The Chinese Civil War ended in communist victory; Karen rebels were inspired by Maoist strategy.

For full case details and sources consulted, see the online appendix.

substate conflict contagion to be less common than all of them. This continues to be true even if I only look at contiguous dyads within my own dataset, or if I use a country-year version of my dataset rather than a directed dyad-year version. (In the country-year version of my dataset, available in the online supplementary materials, the dependent variable is whether a state was a State B or ‘receiver’ of contagion in a given year.) Thus, although the rarity of contagion that I find is partially a function of the directed dyad-year design and partially a function of including non-contiguous dyads, it is also a function of fewer substantive cases of contagion being identified. Only Beardsley (2011) estimates a frequency of substate conflict contagion that is similar to mine, although he arrives at this low frequency by using country-months as the unit of observation rather than country-years. If I divide the frequencies of contagion from my dataset by 12 (assuming that in most cases, if contagion happened in year X, it only happened in one month of year X and not in the other 11), I get 0.05% for the full dataset, 0.11% for the contiguous-only dataset, and 0.11% for the country-year dataset – all well below his 0.66%.

It is clear, then, that prior scholarship has on net overestimated the frequency of the spread of violent civil conflict across borders. By restricting my definition of substate conflict contagion to cases in which one substate conflict made an actual causal contribution to the onset of another, I arrived at a more accurate and much lower frequency of the phenomenon. (Details on the decadal and regional frequencies of substate conflict contagion

Table II.

Comparison of frequency of substate conflict contagion across datasets

Article from which replication data were drawn and temporal scope	Denominator (and description)	Numerator (and description)	Frequency of substate conflict contagion alleged	Frequency of substate conflict contagion found using my dataset with same temporal scope	Frequency of substate conflict contagion found using my dataset with same temporal scope (contiguous states only)	Frequency of substate conflict contagion found using my country-year dataset with same temporal scope
Buhaug and Gleditsch (2008); scope: 1950–2001	2,648 (country-years in which a contiguous neighbor was in intrastate conflict)	131 (onsets of intrastate conflict within this subset)	4.95%	0.55% (100 actual contagion cases/18,069 directed dyad-years)	1.31% (76 actual contagion cases/5,806 directed dyad-years)	1.31% (73 actual contagion cases/5,582 country-years)
Forsberg (2008); scope: 1989–2004	2,532 (directed dyad-years with one state experiencing/having recently ended ethnic intrastate conflict)	186 (onsets of ethnic intrastate conflict)	7.35%	0.64% (59 actual contagion cases, ethnic or not/9,283 directed dyad-years	1.56% (44 actual contagion cases/2,825 directed dyad-years)	1.60% (37 actual contagion cases/2,316 country-years)
Braithwaite (2010); scope: 1960–2001	2,497 (country-years in which a contiguous neighbor was in intrastate conflict)	123 (onsets of intrastate conflict within this subset)	4.93%	0.55% (92 actual contagion cases/16,777 directed dyad-years)	1.33% (71 actual contagion cases/5,325 directed dyad-years)	1.34% (67 actual contagion cases/5,005 country-years)
Beardsley (2011); scope: 1946–2005	37,473 (country-months in which a contiguous neighbor was in intrastate conflict)	248 (onsets of intrastate conflict within this subset)	0.66%	0.56% (115 actual contagion cases/20,446 directed dyad-years)	1.29% (85 actual contagion cases/6,600 directed dyad-years)	1.31% (83 actual contagion cases/6,312 country-years)

Table III.

Replication of Forsberg (2008) logistic regressions, 1989–2004

Variable	Forsberg (2008), Model 3 (DV: ethnic conflict contagion)	Replication with corrected ethnic contagion DV	Forsberg (2008), Model 4 (DV: ethnic conflict contagion)	Replication with corrected ethnic contagion DV
Polarization	0.876 (0.241)**	0.238 (0.513)	0.838 (0.221)**	0.497 (0.371)
Ethnic kin	0.557 (0.213)**	0.794 (0.518)	0.383 (0.187)*	0.805 (0.417)†
Ethnic kin * Polarization	–0.199 (0.369)	0.351 (0.793)	–	–
ln GDP/capita	–0.650 (0.129)**	–0.063 (0.295)	–0.632 (0.128)**	0.024 (0.307)
Polity	0.038 (0.016)*	0.058 (0.033)†	0.040 (0.016)*	0.069 (0.032)*
Polity squared	–0.002 (0.004)	–0.005 (0.009)	–0.002 (0.004)	–0.005 (0.009)
ln Population	0.147 (0.069)*	0.029 (0.143)	0.160 (0.074)*	0.062 (0.148)
Territorial conflict	0.110 (0.178)	–0.350 (0.399)	0.110 (0.210)	–0.446 (0.400)
Intensity of conflict	–	–	–0.015 (0.197)	0.767 (0.474)
Length of boundary	–	–	–0.000 (0.000)	–0.000 (0.000)
Proximity to conflict zone	–	–	0.303 (0.286)	0.600 (0.550)
Decay function	2.641 (0.386)**	0.869 (0.666)	2.643 (0.375)**	0.839 (0.705)
Constant	–1.464 (1.130)	–4.979 (2.985)†	–1.875 (1.161)	–6.565 (3.164)*
Log Pseudo-likelihood	–537.105	–132.899	–536.152	–130.842
Observations	2,377 (181 contagion cases)	2,377 (25 contagion cases)	2,377 (181 contagion cases)	2,377 (25 contagion cases)

Robust standard errors in parentheses; † significant at 10% level; * significant at 5% level; ** significant at 1% level. Listwise deletion due to missing polarization, GDP per capita, and Polity data reduces the number of positive observations from 186 to 181 in Forsberg’s models, and from 26 to 25 in my replications. Similar substantive results are obtained using rare events logistic regression, and using Forsberg’s Models 1 and 2.

are available in Tables A.4 and A.5 in the online appendix.)

The rarity of substate conflict contagion persists if we add to Table A.3 40 additional cases of contagion which violate either the temporal or geographic scope conditions. In Table A.6 (also available in the online appendix), I list these 40 cases. In each, either (1) State A’s conflict ended more than five years prior to the substate conflict onset in State B (i.e. Cuba to El Salvador, 1979), or (2) State A and State B are not in the same neighborhood, as defined in Table A.1 (i.e. Peru to Nepal, 1996). I discovered these cases during the coding process described above, and did not want to exclude them from the dataset entirely even though they violated the coding rules. Needless to say, even if these 40 cases are added, substate conflict contagion is still quite rare, at 162 cases out of 21,404 possible cases (0.76%).⁸

If substate conflict contagion is some ten times rarer than previously believed, there is a clear policy implication. Contrary to the frequent invocation of the fear of substate conflict contagion in debates over military intervention in developing world civil conflicts (see the introduction for some examples), it should now be evident that contagion almost never occurs. The rarity of the spread of violent civil conflict across borders suggests that the fear of contagion, in and of itself, is an insufficient justification for great power intervention in the vast majority of cases. This does not necessarily mean that great powers should not intervene in developing world civil conflicts for other reasons – after all, the fear of contagion has never been the sole cause of a great power military intervention. Regarding Libya, for example, fears of contagion via destabilizing refugee flows into Egypt and Tunisia were voiced by President Obama, but humanitarian concerns appear to have been far more prominent in the White House’s decisionmaking (Obama, 2011). But fears of contagion have almost always been a cause for such interventions, and thus international security policymakers could at least marginally improve their intervention decisionmaking by appreciating the empirical rarity of the spread of violent civil conflict across borders.

A third early impression from this new dataset is that it suggests different correlates of substate conflict contagion from those currently identified in published research. In Table III, I show a replication of Forsberg (2008), Models 3 and 4, replacing her dependent variable of ‘ethnic conflict contagion, 1989–2004’ with my own version. I checked her 186 identified cases of ethnic conflict contagion to see whether they were supported by my own data collection, and found only 26 cases on which we agree (see Appendix Table A.7 for a line-by-line comparison). Thus my dependent variable has 26 ones and 2,351 zeroes, while Forsberg’s has 186 ones and 2,191 zeroes.⁹ (In fairness, Forsberg also uses a more inclusive definition of conflict onset. She codes an onset every time a new warring group enters a conflict, whereas I only code onset for entirely new conflicts, or those reactivated from dormancy. Our threshold for reactivation differs as well; Forsberg allows conflicts to re-onset after one year of inactivity, rather than the four-year threshold I prefer. These definitional differences appear to explain 53 of the 160 cases, identified in Appendix Table A.7, for which she codes contagion and I code non-contagion.)

Unsurprisingly, substantial changes to the dependent variable yield substantially different results. One of Forsberg’s two principal explanatory variables, the ethnic ‘polarization’ of State B’s population, loses statistical significance in my replications, while her other principal variable, the presence of an ethnic kin link between States A and B, loses significance in one model and drops just below the 5% threshold in the other. In addition, both State B’s GDP per capita and population no longer have the expected significant correlation with ethnic conflict contagion. Of the substantive results from Forsberg (2008), only the statistical significance of State B’s democracy score remains unaffected. These replication results should be regarded with extreme caution, however, because the use of logistic regression on data with a small absolute number of ‘events’ (here, cases of ethnic conflict contagion) relative to the number of variables is considered problematic for several statistical reasons, including biased coefficients and reduced power. Generally a ratio of at least ten events to one variable is desirable (Peduzzi et al., 1996), but these replications have ratios of 2.78 for Model 3 and 2.27 for Model 4. The more fundamental point may be that given the new estimates of contagion’s rarity and the need for a robust set of control variables in conflict research, logistic regression may be an altogether inappropriate methodology of study, and a methodology like matching may be more appropriate. Either way, the mystery behind the causes of substate conflict contagion has only deepened – suggesting this topic is ripe for further exploration.

Conclusion and suggestions for future research

In this article, I introduced an original dataset of cases and non-cases of the spread of violent civil conflict from one state to another. My dataset improves on existing universes of cases in the small literature on this subject, by (1) expanding the geographic scope of ‘substate conflict contagion’ beyond contiguous states and (2) requiring that the ‘contagious’ conflict in the first state actually be causally linked to the conflict onset in the second, ‘infected’ state. The resulting universe of cases shows substate conflict contagion as substantially less common than either past scholarship or past and present policymakers acknowledge.

While the rarity of substate conflict contagion shown by this dataset is striking, even more impactful policy implications would likely arise from a better understanding of what causes contagion to occur about 0.6% of the time, and to not occur the other 99.4% of the time. Table III’s replication has only scratched the surface of such research, which would help policymakers decide not only on the risk of a given conflict spreading, but also what precisely they should do to stop conflicts from spreading when they assess the risk of contagion to be high.

This dataset was originally built for research into precisely this set of questions – what causes contagion and how can it be prevented. At this point my own answers are preliminary but counterintuitive. In the book-length version of this research project, I disaggregate my universe of contagion cases (including the 40 added provisionally) into 84 high-intensity and 78 low-intensity cases, where ‘high-intensity’ simply designates those cases in which at least 1,000 cumulative battle-related deaths resulted from the State B conflict (according to the Armed Conflict Dataset). I find that almost all of the high-intensity cases – about 75% – experienced contagion only when one of the sovereign state governments involved took a deliberate action that enabled the spread of conflict to occur (for example, sponsoring a rebel group), while in the low-intensity cases, these same state government actions were uncommon, occurring in about 33% of these cases. These findings suggest that state governments play a critical and potentially underappreciated role in the most catastrophic contagion cases.

Future research could continue exploring contagion’s causes and consequences by attempting to disaggregate the cases of contagion by type of conflict or other forms of intensity besides battle-related deaths. There is also a critical need to avoid selection on the dependent variable, by focusing on the more than 21,000 cases of non-contagion identified in this dataset (available online). Any variables consistently present in both these cases and the 122–162 cases of actual contagion probably cannot explain much variation in the occurrence of this phenomenon. These are just some of the potential research projects that could follow from this data collection effort.

Footnotes

Replication data

The datasets, coding documentation, and do-file for the empirical analysis in this article can be found at .

Acknowledgments

I thank JPR’s editor and three anonymous reviewers, Kenneth A Oye, Fotini Christia, Roger Petersen, Barry Posen, Daniel Altman, Chad Hazlett, Erika Forsberg, Sameer Lalwani, Miranda Priebe, Josh Shifrinson, David Weinberg, and seminar participants at MIT, the Harvard Kennedy School, and the Tobin Project for their helpful feedback on earlier versions of this work.

Notes

References

Beardsley

Kyle

(2011) Peacekeeping and the contagion of armed conflict. Journal of Politics 73(4): 1051–1064.

Braithwaite

Alex

(2010) Resisting infection: How state capacity conditions conflict contagion. Journal of Peace Research 47(3): 311–319.

Buhaug

Halvard

Gleditsch

Kristian Skrede

(2008) Contagion or confusion? Why conflicts cluster in space. International Studies Quarterly 52(2): 215–233.

Clinton

William

(1995), quoted in Brown, Michael E (1996) Introduction. In: Brown

Michael E

(ed.) The International Dimensions of Internal Conflict. Cambridge, MA: MIT Press, 23–24.

Groot

Olaf

(2011) Culture, contiguity, and conflict: On the measurement of ethnolinguistic effects in spatial spillovers. Journal of Development Studies 47(3): 436–454.

Forsberg

Erika

(2008) Polarization and ethnic conflict in a widened strategic setting. Journal of Peace Research 45(2): 283–300.

Forsberg

Erika

(2009) Neighbors at Risk: A Quantitative Study of Civil War Contagion. PhD dissertation, Uppsala University.

Fox

Jonathan

(2004) Is ethnoreligious conflict a contagious disease? Studies in Conflict & Terrorism 27(2): 89–106.

Gleditsch

Kristian Skrede

Salehyan

Idean

Schultz

Kenneth

(2008) Fighting at home, fighting abroad: How civil wars lead to international disputes. Journal of Conflict Resolution 52(4): 479–506.

10.

Gleditsch

Nils Petter

Wallensteen

Peter

Eriksson

Mikael

Sollenberg

Margareta

Strand

Håvard

(2002) Armed conflict 1946–2001: A new dataset. Journal of Peace Research 39(5): 615–637.

11.

Kathman

Jacob D

(2010) Civil war contagion and neighboring interventions. International Studies Quarterly 54(4): 989–1012.

12.

Marshall

Monty G

Jaggers

Keith

Gurr

Ted Robert

(2008) POLITY IV project: Political regime characteristics and transitions, 1800–2010 (http://www.systemicpeace.org/inscr/inscr.htm).

13.

Obama

Barack

(2011) Remarks by the President in Address to the Nation on Libya. 28 March (http://www.whitehouse.gov/the-press-office/2011/03/28/remarks-president-address-nation-libya).

14.

Peduzzi

Peter

Concato

John

Kemper

Elizabeth

Holford

Theodore R

Feinstein

Alvan R

(1996) A simulation study of the number of events per variable in logistic regression analysis. Journal of Clinical Epidemiology 49(12): 1373–1379.

15.

UCDP & PRIO (2009) UCDP/PRIO armed conflict dataset codebook (http://www.prio.no/Global/upload/CSCW/Data/UCDP/2009/Codebook_UCDP_PRIO%20Armed%20Conflict%20Dataset%20v4_2009.pdf).

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB