Académique Documents
Professionnel Documents
Culture Documents
ANDREW J. ENTERLINE
Department of Political Science, University of North Texas
KYLE A. JOYCE
Department of Political Science, Pennsylvania State University
What effect do third parties have on the evolution of civil wars? The authors argue that interven-
tion by third parties is central to the civil war process, a process that is characterized by the dur-
ation of hostilities and the type of outcome. The authors examine empirically the effect of third-party
intervention into civil wars during the period 1816–1997, using the event history framework of
competing risks. From the perspective of competing risks, as a civil war endures, it is at risk of experi-
encing a transition to one of three civil war outcomes in our sample: military victory by the gov-
ernment, military victory by the opposition group, and negotiated settlement. The competing risks
approach provides considerably better leverage on the dynamic qualities of civil wars and, in par-
ticular, the influence of interventions by third parties. The analysis suggests that third-party inter-
ventions can be decisive in the evolution of civil wars and that third-party interventions have a
different effect on the duration than different civil war outcomes. The results show that third-party
intervention decreases the time until the supported group achieves military victory. Furthermore,
third-party interventions, on both the government and opposition sides, increase the time until a
negotiated settlement.
* Dylan Balch-Lindsay, our co-author and friend, passed are grateful to Janet Box-Steffensmeier, Mark J. C.
away in an automobile accident in the fall of 2002. To Crescenzi, David Cunningham, Charles Boehmer, Suzanna
those who knew him, Dylan was a shy, gentle young man De Boef, Jeffrey Dixon, Errol Henderson, Brad Jones,
with an incredible knack for econometrics and fixing any- David Mason, Curtis Meek, Neil Mitchell, Mark
thing mechanical. Dylan left us at the very moment he was Peceny, Brandon Prins, Patrick Regan, Dan Reiter, Karen
about to embark on his dissertation project, and we miss Remmer, Christina Schatzman, William Stanley, Chris
him greatly. While the following article has changed in Zorn, and three anonymous referees for comments and
form since Dylan last saw it during summer 2002, we encouragement. All statistical analysis was conducted using
remain faithful to Dylan’s interest in the dynamic qualities Stata 9.2 (Stata Crop, 2005). The figures were produced
of conflict processes and his passion for event history model- using R 2.3.1 (R Development Core Team, 2006). The data
ing. This article originally was conceived in late 1998, in used in this article, along with a web appendix containing
preparation for the annual meetings of the American coding decisions and additional results, are available at
Political Science Association, 2–5 September 1999, and the http://www.prio.no/jpr/datasets. Correspondence should
International Studies Association, 14–18 March 2000. We be directed to kjoyce@psu.edu.
346 jour nal of P E A C E R E S E A RC H volume 45 / number 3 / may 2008
differentiated between the types, or ‘modes’, continued conflict’ (Mason & Fett, 1996:
of conflict termination (Carroll, 1969; Holsti, 548). The first half of the authors’ formal
1966; Phillipson, 1916). Other scholars model is balanced by each party’s expected
developed predictive models of war termin- utility for a negotiated settlement.
ation based on the attributes of conflicts, such Subsequently, Mason, Weingarten & Fett
as casualties (Klingberg, 1966). World War II (1999) assess the effects of factors similar to
stimulated several reflections on the dynamic those included in Mason & Fett (1996) on
process of victory and defeat in conflicts. For three civil war outcomes: government mili-
example, Calahan (1944) argues that the tary victory, rebel military victory, and nego-
loser dictates the terms of the peace, a theme tiated settlement. The authors find that
reasserted by Coser (1961). In a controversial interventions by third parties reduce the
assessment, Kesckemeti (1958) suggests that probability of a negotiated settlement; yet, as
the manner in which a conflict concludes is a civil war endures, third-party intervention
nearly predetermined. Other research resorts increases the probability of such a settlement.
to formal methods to model the process by Walter (2002) examines whether the costs of
which conflicts terminate, through rational the civil war, the balance of power between
decisionmaking (Porsholt, 1966; Wright, the disputants, the domestic political institu-
1965). This decision-theoretic approach is tions of the civil war state, ethnicity issues, the
further developed in recent approaches to stakes, mediation, and credible commitments
studying conflict, with Wittman (1979), for influence the probability of negotiations,
example, proposing that war endings are the signed peace bargains, and the success of
culmination of a process between war partici- peace settlements. Walter’s empirical analysis
pants who are constantly updating their of civil wars during the post-WWII period
expectations regarding the eventual outcome indicates that outside security guarantees and
of a conflict. Similarly, Pillar (1983: 345) power-sharing pacts are significantly related
argues that ‘the decision to negotiate is thus to peaceful civil war settlements.
not so much a choice of a discrete alternative Finally, DeRouen & Sobek (2004) formu-
but rather the selection of one moment along late and test empirically a model of the
an anticipated continuum of time in which dynamics of civil war duration and outcome
the military situation constantly changes’. grounded in the capacity of the civil war state.
These formal approaches provided a fruit- The authors examine four outcomes: govern-
ful foundation for the subsequent study of ment victory, rebel victory, truce, and treaty.
civil wars. For instance, Mason & Fett (1996: DeRouen & Sobek find that state capacity, in
548) rely on Wittman (1979) to argue that the the form of an effective state bureaucracy,
choice by civil war combatants of whether to reduces the odds that anti-government rebel
seek an end to a civil war through a negotiated groups will achieve victory. The authors con-
settlement or continue fighting ‘is a function clude that ‘duration decreases the probability
of the difference in each party’s expectations of a government victory and increases the
concerning the eventual outcome of the con- probability of treaty’ (2004: 315). This con-
flict if a negotiated settlement is not reached clusion generally supports Zartman’s (1985,
immediately’. Each combatant’s expected 1993) notion of a hurting stalemate, in which
utility for continuing a conflict is a function a government that is unable to defeat a rebel
of its subjective estimate of the probabilities group rapidly is less likely to do so successfully
for victory and defeat and their attendant as time passes. The authors also find that the
utility for these two outcomes. In turn, this longer a civil war endures, the greater the costs
expected utility calculation is conditioned by to the government for prosecuting it further,
the ‘rate at which [a] party absorbs costs of and the more likely the government will seek
348 jour nal of P E A C E R E S E A RC H volume 45 / number 3 / may 2008
to settle a conflict through peaceful means. the intensity of a civil war is high and more
However, DeRouen & Sobek (2004) model is likely to intervene when a humanitarian crisis
decidedly state-centric, thereby excluding is imminent. Despite some disagreement,
involvement by third parties independent of a central thrust of this first agenda is that
United Nations’ sponsored interventions. third parties are frequent interveners in civil
wars and these interventions are more often
Third Parties and the Civil War Process than not grounded in self-interested motives,
Despite a traditional emphasis on national- motives that are central to the evolution of
level causes and consequences of civil wars, intrastate conflicts.
recognition of the international dimensions The second agenda focuses on the effect
of these conflicts is longstanding. For of third parties on the outcome of civil wars.
example, Modelski (1964) asserts that inter- For example, Zartman (1993) concludes that
nal conflicts are part and parcel of the inter- civil war belligerents rely on third parties, not
national system, an argument reiterated in only for resources and sanctuary, but also for
Deutsch (1964). Rosenau (1964) theorizes leverage, both on the battlefield and at the
that the type and evolution of a civil war negotiating table, a condition that he refers
influences third-party intervention, and he to as ‘triangulation’. In his empirical analysis
argues that the goals of third parties condition of civil wars in the post-WWII period, Regan
the way in which civil wars end. Drawing (1996, 2000) finds that third parties are more
explicitly or implicitly on this early work, likely to be successful in their interventions in
current research on the role of third parties civil wars when they resort to a strategy mixing
divides into three roughly defined research economic and military policies. Carment &
agendas: (1) the causes of third-party inter- James (1998: 591) report that, while ‘more
vention in civil wars, (2) the effect of third- intensive initial interventions are more likely
party intervention on civil war outcomes, and to result in cooperative outcomes’, non-
(3) the effect of third-party intervention on conflictual outcomes are also a function of the
the duration of civil wars. rebel group’s salience for the conflict and its
The first research agenda explores the con- military capability. Mason, Weingarten & Fett
ditions that lead third parties to join ongoing (1999) find that third-party interventions
civil wars. For example, Suhrke & Noble make a negotiated settlement less likely, unless
(1977) rely on case studies to demonstrate this intervention occurs once a civil war
that third parties often intervene in ethnic becomes protracted, at which point the likeli-
civil wars for reasons entirely exogenous to hood of a negotiated settlement increases,
the wars themselves, and, thus, civil wars are a finding that is bolstered by Walter’s (2002)
often fought against a backdrop of multiple analysis of third-party guarantees.
levels of interest, allegiance, and involvement The third research agenda focuses on the
by third parties. Holl (1993) argues that third effect of third-party intervention on the dur-
parties are more likely to enter civil wars as ation of civil wars. Several studies of civil war
these conflicts approach termination, thereby outcomes develop theoretical arguments
introducing the complicating factors associ- linking the outcomes of these conflicts to their
ated with multiple preferences and coalition duration. Generally, the arguments advanced
formation, and, ultimately, when and how in these studies tie civil war duration to the
civil wars end. Finally, Regan’s (2000) analy- costs of fighting, whereby long wars increase
sis of third-party involvement in civil wars the costs to the combatants and, in turn,
during the post-WWII period suggests that increase the likelihood that a civil war will
third parties are less likely to intervene when end in a specific way. Indicative of these
D ylan Balch-Lindsay et al. T H I R D -P A RT Y I N T E RV E N T I O N 349
studies is reliance on post hoc knowledge of finance its civil war effort. In turn, the reduced
civil war duration to inform statistical models necessity to exact resources from the popula-
intended to predict civil war outcomes. tion, as well as the potential for an increase in
Alternatively, several studies examine the the degree of security enabled by third-party
effect of third parties on civil war duration resources, allows a government to shore up
itself, without explicitly incorporating the political support, while at the same time reduc-
manner in which civil wars end. For example, ing the ability of an opposition group to gain
Collier, Hoeffler & Söderbom (2004) con- a foothold on popular allegiance. Additionally,
clude that civil war states that are ethnically third-party intervention on behalf of a govern-
fractionalized are likely to experience longer ment enables the latter to alter the probability
civil wars, yet the authors do not consider the of various civil war outcomes occurring, as the
role of third parties. Balch-Lindsay & availability of greater military, political, and
Enterline (2000) explore the effect of simul- economic capability increases a government’s
taneous interventions on the government odds of winning militarily, provides it with
and opposition sides in a civil war and con- greater leverage over the terms of negotiation,
clude that these interventions result in longer and reduces its odds of losing militarily.
civil wars, a finding corroborated in Elbadawi Similarly, third-party interventions on the
& Sambanis (2000) and Regan (2002). Third side of an opposition group influence the
parties, therefore, are central to the civil war behavior of the opposition group. Third-party
process. Even decisions by third parties not to intervention facilitates the ability of the oppos-
intervene overtly in a civil war has meaning- ition to impose costs on a government. An
ful consequences for the manner in which opposition group supported by a third-party
civil wars evolve. is more likely to challenge the government
Clearly, a central theme in the aforemen- with a capable military threat, thereby increas-
tioned literatures is that third parties are central ing the cost to the government of confronting
to the civil war process. We study this issue the opposition, as well as increasing the oppos-
further in the remainder of this section by ition group’s chances of achieving victory.
assuming that third-party intervention in civil Third-party support for the opposition
wars manifests itself in three ways: (1) third- enables the latter to demonstrate to the
party intervention on behalf of the govern- civil war state’s populace that it presents a
ment of the civil war state; (2) third-party credible threat to the government. By provid-
intervention on behalf of the opposition group ing resources to the opposition, third-party
challenging the government; and (3) simul- support increases the capacity of the opposi-
taneous third-party interventions on behalf of tion to disrupt a government’s provision of
both the government and opposition sides. security, goods, and services to its citizens.
Indeed, third-party support may enable the
Third-Party Support for Government or opposition to bid for popular support by pro-
Opposition Third-party intervention on the viding a rival source of goods and services and,
side of the government alters a government’s in turn, challenging the government’s monop-
decision calculus – specifically, the cost and oly over the ‘hearts and minds’ of its citizens.
benefits to the government for pursuing In general, third-party support increases the
specific policy choices (e.g. continued likelihood that a combatant will more rapidly
fighting) – and, in turn, the probability of spe- win the conflict militarily.
cific outcomes occurring. The presence of
third-party support enables the government to H1: Third-party support increases the likeli-
reduce the rent it extracts from its populace to hood of a military victory by the supported
350 jour nal of P E A C E R E S E A RC H volume 45 / number 3 / may 2008
government that would secure a separate state Civil War Costs Central to much of the
over a negotiated settlement that might result contemporary literature’s treatment of the
in only partially fulfilling the separatist group’s link between costs and the civil war process is
demand for a separate state, but the presence the assumption that such costs are roughly
of a resolute government likely foreshadows a evenly distributed across the combatants.3
costly conflict. Ultimately, there is some basis Although the destruction of government assets
for the argument that a government and its might suggest that an opposition group is
opposition will prefer a negotiated settlement increasing its odds of military victory, the
to other outcomes. A central part of this argu- core of government and opposition power
ment is that the preferences of the combatants (i.e. popular allegiance) is, in part, a function
change over time, such that military victory is of the degree to which a civil war is costly. If
preferred to negotiated settlement initially, this assumption of roughly equitable costs
and later, as a civil war endures, this preference holds, the strategic decisions that are likely to
by the combatants is reversed. flow from this assumption should translate
The logic of the second argument is into an increase in the probability of a nego-
grounded in the assumption that, because tiated settlement occurring at any given point
separatist demands challenge the very basis of in time, a linkage similar to the aforemen-
state sovereignty (i.e. territorial integrity), tioned notion of a hurting stalemate.
governments are extremely reluctant to sur-
render the prerogative to determine policy Bargaining Reputation The theoretical lit-
within the geographic boundaries of their erature suggests that civil war is a bargaining
states. Throughout the history of the inter- process (Pillar, 1983; Wagner, 1993; Licklider,
state system, states rarely condone partition. 1995; Walter, 2002, 2004). The bargaining
Additionally, the preference to maintain ter- process between civil war combatants is
ritorial integrity is, in part, a product of gov- carried out under varying degrees of incom-
ernment fear that allowing one region to gain plete information and uncertainty. Civil war
autonomy establishes precedent for subse- combatants base their expectations, in part,
quent separatist demands. Therefore, sep- on the war-fighting reputations of their
aratist demands are likely to result in a opponents. Several studies demonstrate that
concerted effort by the government to resist such inter-civil war processes are likely at
separatist demands. We expect that consider- work within countries that experience civil
ations of sovereignty and fear of demonstra- war. For example, Licklider (1995) concludes
tion effects increase government motivation that countries in which a civil war ends in a
to continue fighting, if only to forestall victory decisive military victory are less likely to
by the separatist opposition. Moreover, gov- experience subsequent outbreaks of civil war.
ernments will likely thwart efforts for a nego- Walter (2004: 385) concludes that civil war
tiated settlement, as such an agreement would, duration and outcome (i.e. short wars that
by definition, require sovereignty to be com- end in partition) are likely to increase the
promised to some degree, again providing odds that a civil war state will experience
grounds for further demands for autonomy subsequent civil war.
on the government. Under these conditions,
a government’s preference to continue fight-
ing suggests that the government prefers its 3 In part, this assumption flows from the dearth of data on
own military victory to virtually any other costs to rebel and opposition groups for prosecuting civil
war. This is, of course, not particularly sound logic and will
outcome, and the outcome is likely to be likely be rectified by the availability of such data in the
continued fighting and a longer civil war. future.
D ylan Balch-Lindsay et al. T H I R D -P A RT Y I N T E RV E N T I O N 353
We follow this line of reasoning here. Here, we posit that civil war states with rela-
Specifically, we focus our attention on the tively greater economic resources have two
actor for which we have consistent cross-time effects on the civil war process. First, civil war
data on strategic choices during civil war: states with relatively greater economic devel-
the government. We begin by assuming that opment are capable of absorbing the costs of
states develop a single reputation for the prosecuting a civil war, as well as offsetting the
strategic choices they pursue in civil war, even cost of the conflict for loyalist constituents,
if the policymakers controlling a given state’s and, in doing so, blunting the ‘hurting’ com-
government in one civil war are distinct from ponent of the ‘hurting stalemate’. Second,
the policymakers influencing that same state’s civil war states that are relatively more devel-
policy choices in a previous civil war. States oped can bring greater force to bear on the
with reputations for negotiating in civil wars battlefield, thereby increasing the chances of
do not necessarily prefer negotiated settle- military victory or at least forestalling defeat.
ments relative to other outcomes at any point
in time; in fact, we expect that governments Democracy The theoretical reasoning under-
prefer negotiated settlements only when mili- lying the inclusion of democracy in a model
tary defeat by the opposition is imminent. predicting civil war duration is similar to that
However, many factors can handicap a gov- described above with regard to economic devel-
ernment’s ability to achieve an outcome more opment. Specifically, many of the causal argu-
preferable then a negotiated settlement, ments relating democratic institutions in the
thereby contributing to a reputation charac- civil war state with the duration of hostilities are
terized by negotiated settlement as it appears grounded in causal arguments tied to the pre-
in the historical record. In turn, government civil war phase, such that democracy deters
reputation provides a potential opposition fighting or increases the likelihood of a negoti-
group a basis for developing expectations ated settlement. We argue that democratic
about how a state’s current government will states are, relative to non-democratic states,
respond to a challenge by the opposition. acutely aware of the detrimental effect of civil
Given this logic, we anticipate that govern- war breaking out, and they are even more
ments with reputations for negotiating acutely aware of the detrimental negative con-
settlements in civil wars will experience sequences of an opposition victory. For demo-
difficulties winning subsequent civil wars. cratic polities, allowing constituents to achieve
their policy preferences through a resort to
Economic Development Several recent arms, rather than through peaceful negotiation
studies explore the link between economic and compromise, incises the heart of the demo-
development and civil war duration (e.g. cratic process. Additionally, allowing an oppos-
Collier, Hoeffler & Söderbom, 2004; ition group to succeed may set a dangerous
DeRouen & Sobek, 2004; Fearon, 2004; precedent and pollute the democratic process
Walter, 2004). However, the causal logic by presenting resort to arms as a viable alterna-
underlying this relationship between devel- tive mode of policy negotiation, rather than a
opment and civil war duration is not entirely nearly unthinkable alternative. In the light of
evident; indeed, the plausible theoretical this logic, we anticipate that civil war states with
relationship between development and civil a democratic tradition will be relatively better
wars would appear to be in the pre-civil war able to mobilize resources against an opposition
phase, where greater levels of development act group, will fight harder to defeat opposition
as a deterrent to the outbreak of hostilities, groups military, and will be reluctant to nego-
rather than once a civil war is under way. tiate a settlement.
354 jour nal of P E A C E R E S E A RC H volume 45 / number 3 / may 2008
defeat) experienced by the civil war state. accommodating multiple outcomes.9 In the
Government reputation ranges from 0 to 1.8 competing risks framework each observation
(i.e. a civil-war-day) is at risk of experiencing
Economic Development We use the COW a transition to three different civil war out-
National Material Capability Data (Singer, comes: (1) military victory by the govern-
Bremer & Stuckey, 1972) to create a measure ment, (2) military victory by the opposition,
of economic development, calculated as the and (3) negotiated settlement. In this way,
natural log of the sum of energy consump- we can allow covariates to have different
tion and iron/steel production. We create a effects across the competing risks.
pre-civil war level of economic development We specify three separate models, one for
by lagging this variable one year prior to the each potential outcome, where failures due
civil war. to the other two risks are treated as randomly
censored. For example, we estimate a model
Regime Type We use the Polity IV Data where the outcome is military victory by the
(Marshall & Jaggers, 2002) to create a government and treat failures due to military
dichotomous variable, democracy, that is victory by the opposition and negotiated set-
coded 1 if the civil war state’s polity score in tlement as randomly censored. This model
the year prior to the civil war was greater than makes the assumption that the risks are con-
5 and 0 otherwise. ditionally independent; that is, if an obser-
vation fails due to one risk, it could have
Competing Risks Model experienced one of the other outcomes, had
Because a civil war can end in different ways, the civil war lasted long enough. For
a covariate can potentially have a different example, failure due to military victory by
effect on the time to different civil war out- the government is independent of failure due
comes. For example, a covariate may have a to military victory by the opposition and
negative effect on the hazard (i.e. the risk of negotiated settlement. Thus, there is a poten-
a civil war ending at a particular point in tial failure time associated with each of the
time given that the civil war has survived up civil war outcomes.10 However, once a civil
to that point in time) of one risk and a posi- war ends, it exits the sample and is no longer
tive effect on another. In order to account for the at risk of experiencing any of the other civil
different effects a covariate can have on the war outcomes, that is, we only observe the
time to different civil war outcomes, it is neces- occurrence of the first outcome.
sary to use a statistical model that can We estimate three separate Cox propor-
account for the duration of a civil war and tional hazards models. We prefer the Cox
the various ways a civil war can end. Toward model to the parametric alternatives because
that end we estimate a ‘competing risks’ it makes no assumptions about the distribu-
event history model that treats time as a tional form of the baseline hazard rate
function of some underlying risk of an event (Box-Steffensmeier & Jones, 2004). We test
occurring at each time point, as well as
9 An introduction to competing risk event history
models can be found in Box-Steffensmeier & Jones
(2004: 166–181).
10 If the outcomes, conditional on the covariates, are not
8 For cases in which civil war states did not experience a
independent, then the assumption that we can treat
prior civil war, we recoded this variable with a value of 0, the other outcomes as randomly censored is violated.
thereby capturing the fact that the opposition in these con- In this case, a dependent risks model is appropriate
flicts had no historical basis to expect the government to (Han & Hausman, 1990; Hill, Axinn & Thornton, 1993;
negotiate. Gordon, 2002).
356 jour nal of P E A C E R E S E A RC H volume 45 / number 3 / may 2008
for violations of the proportional hazards demonstrate the advantage of the competing
assumption (Box-Steffensmeier, Reiter & risks approach, namely, that the effect varies
Zorn, 2003) in each of our three Cox models across the three civil war outcomes. Figure 1
and correct for non-proportionality by inter- presents the estimated baseline survivor
acting the offending covariates with the function for each civil war outcome when
natural log of time. there is an intervention on the side of the
government.11 This figure demonstrates that
a third-party intervention on the side of the
Analysis
government has a different effect on the
The results of our Cox models of the dur- expected probability of a civil war surviving
ation to the three civil war outcomes are past a given point in time for each civil war
reported in Table I. Our first hypothesis outcome. Importantly, these differing effects
anticipates that third-party support will would not be evident had we estimated a
increase the likelihood of the supported Cox model where we pooled the three civil
group achieving military victory, while war outcomes.
decreasing the likelihood of the unsupported Turning to the results corresponding to
group achieving military victory. The posi- intervention for opposition, the statistically
tive and statistically significant coefficient for significant coefficient for the Government
intervention for government in the Govern- Military Victory model indicates that third-
ment Military Victory model indicates that party support for the opposition decreases
third-party intervention on behalf of the the risk that a civil war will end in a govern-
government of the civil war state increases ment military victory. The positively signed
the risk that the civil war will end in a gov- and statistically significant coefficient for
ernment military victory and, therefore, this variable in the Opposition Military
decreases the time until such a victory will Victory model suggests that third-party
occur. The interaction between this variable support increases the risk that a civil war will
and the natural log of time is negatively end in an opposition military victory and,
signed and statistically significant, indicating thus, decreases the time until such a victory
that, as a civil war persists, the benefit of will occur. Finally, the positively signed and
third-party support for the government statistically significant coefficient in the
decreases the risk of a civil war ending in a Negotiated Settlement model suggests that
government military victory. intervention on behalf of the opposition
The coefficient for intervention for gov- increases the risk of a civil war concluding
ernment in the Opposition Military Victory more rapidly in negotiated settlement,
model is not statistically significant and is lending support to our second hypothesis,
positive and statistically significant in the though the risk is smaller than military
Negotiated Settlement model, suggesting victory by the opposition (see Figure 2). We
that interventions on behalf of the govern- suspect that this final relationship is the result
ment have no statistically significant effect of third-party intervention making an oppos-
on the time until an opposition military ition group a more credible and potentially
victory and decrease the time until a negoti-
ated settlement. Interestingly, civil wars have
a higher risk of ending in a negotiated settle-
11 For each figure, the estimated survivor function was cal-
ment than a government military victory
culated based on the mean values for continuous variables,
when a third-party intervenes on the side of modal values for discrete variables, and one for the inter-
the government. The results for this variable vention variable of interest.
D ylan Balch-Lindsay et al. T H I R D -P A RT Y I N T E RV E N T I O N 357
Table I. Cox Competing Risks Model of Civil War Duration & Outcome, 1816–1997
long-term opponent for the government and, intervention is positive and statistically sig-
in turn, making negotiated settlement an nificant in the Government Military Victory
attractive outcome for a government. model and negative and statistically signifi-
In the third hypothesis, we anticipated cant in the Negotiated Settlement model,
that interventions of third parties on both suggesting that the presence of a balanced
the government and opposition sides in a distribution of third parties increases the
civil war would decrease the likelihood of the risk of a government military victory and
civil war ending in a military victory and decreases the risk of a negotiated settlement
increase the likelihood of a negotiated settle- (see Figure 3), a finding supporting the rea-
ment. However, the coefficient for balanced soning of our fourth hypothesis, whereby
358 jour nal of P E A C E R E S E A RC H volume 45 / number 3 / may 2008
1.0
Government military victory
Opposition military victory
Negotiated settlement
0.8
0.6
Probability of survival
0.4
0.2
0.0
Duration in days
0.6
0.4
0.2
Duration in days
D ylan Balch-Lindsay et al. T H I R D -P A RT Y I N T E RV E N T I O N 359
1.0
0.8
Civil war costs are often argued to increase Finally, while the coefficients corresponding
the likelihood that civil wars will end with a to democracy indicate that democratic states
negotiated settlement. Here, we examine this are more likely to fight shorter civil wars and
relationship in the context of two additional win them than either lose to the opposition
outcomes that are probable in intrastate con- group or negotiate a settlement, these coeffi-
flicts, military victory by the government or cients are not statistically significant.
the opposition. The coefficients for war costs
in our three models suggest that, as the costs
Conclusion
of the civil war increase, the time until an
opposition military victory significantly The competing risks approach provides con-
decreases, while the time until a negotiated siderably better leverage on the frequently
settlement significantly increases. While we cited dynamic qualities civil wars and, in par-
expected that costly civil wars were likely to ticular, the influence of interventions by
increase the attractiveness of negotiated settle- third parties. At the same time, this method
ment to combatants, costly civil wars might of analysis reveals the limits of contemporary
also reflect conditions in which one combat- empirical studies of civil war in sorting and
ant is winning the conflict rapidly by exact- testing the bevy of theoretical claims regard-
ing tremendous cost on an opponent. Given ing when, how, and why intrastate conflicts
that the COW civil war data measures only evolve and end, a criticism that we extend to
the battle-deaths suffered by the state partic- our efforts in this article. For example, our
ipants in a civil war, we are unable to sort out analysis reveals that increases in economic
the conditions that result in negotiated out- development increase the time to all three
comes from those that result in military outcomes. Albeit interesting, these empirical
victory by either side. findings fail to provide much insight into the
The negative and statistically significant different effects of economic development
coefficient for government reputation in the on the time to different civil war outcomes.
Government Military Victory model sug- In part, this inconclusiveness is a function of
gests that states that have a reputation for the quality and richness of our data on civil
negotiating in past civil wars are likely to wars, but it is also likely due to the precision
have to fight longer to win future civil wars, of our theoretical expectations. Thus, while
a finding that underscores the relevance of we argue that a competing risks approach is
long-term conflict behavior and the infor- generally better than alternative methods,
mation this provides to combatants engaged relying upon this method does not rectify
in the civil war process. shortcomings in theory and information per-
The coefficients corresponding to eco- taining to civil wars.
nomic development indicate that, as the eco- The differences in our findings with
nomic development level of the civil war respect to third-party interventions empha-
state increases, the time until a government sizes the importance of distinguishing be-
or opposition military victory or a negotiated tween the time to different civil war outcomes.
settlement decreases, but, as a civil war per- While third-party intervention increases the
sists, the influence of economic development likelihood of the supported group achieving
weakens. Thus, in general, more econom- military victory, third-party intervention on
ically developed states are expected to fight behalf of the government or the opposition
shorter civil wars, and these civil wars have a also increases the likelihood of a negotiated
higher risk of ending in a military victory for settlement. However, third-party interven-
either side than a negotiated settlement. tions on the side of both the government
D ylan Balch-Lindsay et al. T H I R D -P A RT Y I N T E RV E N T I O N 361
and opposition decrease the likelihood of a Box-Steffensmeier, Janet M.; Suzanna De Boef &
negotiated settlement. The results highlight Kevin Sweeney, 2005. ‘Multilevel, Stratified,
the important differences in the dynamics of Frailty Models and the Onset of Civil War’,
civil wars and underscores the importance of paper presented at the Annual Meeting of the
considering civil wars as processes, rather Midwest Political Science Association, Chicago,
IL, 7–10 April.
than static events. In doing so, we anticipate
Box-Steffensmeier, Janet M. & Bradford S. Jones,
the emergence of better theory, better empir- 2004. Event History Modeling: A Guide for
ical tests of these theories, and better policy Social Scientists. Cambridge: Cambridge
recommendations. University Press.
With respect to policy, our study provides Box-Steffensmeier, Janet M.; Dan Reiter &
insight into recent civil wars. For example, in Christopher Zorn, 2003. ‘Nonproportional
the Balkans, efforts by the North Atlantic Hazards and Event History Analysis in
Treaty Organization (NATO) to isolate International Relations’, Journal of Conflict
Yugoslavia from third-party support (princi- Resolution 47(1): 33–53.
pally, Russia) were an important element in the Calahan, Harry A., 1944. What Makes War End?
allied strategy to aid the Albanian Kosovars New York: Vanguard.
Carment, David & Patrick James, 1998. ‘The
and limit the duration of this conflict. The
United Nations at 50: Managing Ethnic
central thrust of our study suggests that, had
Crises–Past and Present’, Journal of Peace
NATO allies failed in their efforts to deter Research 35(1): 61–82.
Russia from overt intervention on the side of Carroll, Berenice A., 1969. ‘How Wars End: An
Yugoslavia, the civil war in Kosovo most likely Analysis of Some Current Hypotheses’,
would have raged for much longer than it did. Journal of Peace Research 6(4): 295–321.
Conversely, in central Africa, the success or Clodfelter, Michael, 2002. Warfare and Armed
failure of peace settlement talks in the former Conflicts: A Statistical Reference to Casualty and
Zaire hinge primarily on coordinating the Other Figures, 1500–2000. Jefferson, NC:
preferences of a half-dozen intervening African McFarland.
states, a very convincing example of the central Collier, Paul; Anke Hoeffler & Måns Söderbom,
importance of collective action problems in 2004. ‘On the Duration of Civil War’, Journal
of Peace Research 41(3): 253–273.
intrastate conflicts, multiple sets of prefer-
Coser, Lewis A., 1961. ‘Termination of Conflict’,
ences, and civil war duration and outcome. Journal of Conflict Resolution 5(4): 347–353.
Indeed, with several states overly committed to DeRouen, Karl R., Jr., & David Sobek, 2004. ‘The
aiding the government and opposition com- Dynamics of Civil War Duration and Outcome’,
batants, our model suggests conditions of Journal of Peace Research 41(3): 303–320.
stalemate in which a negotiated settlement is Deutsch, Karl W., 1964. ‘External Involvement in
not likely in the near term. Internal War’, in Harry Eckstein, ed., Internal
War: Problems and Approaches. New York: Free
Press (100–110).
References
Elbadawi, Ibrahim & Nicholas Sambanis, 2000.
Balch-Lindsay, Dylan & Andrew J. Enterline, ‘External Interventions and the Duration of
2000. ‘Killing Time: The World Politics of Civil Civil Wars’, World Bank Policy Research
War Duration, 1820–1992’, International Working Paper 2433.
Studies Quarterly 44(4): 615–642. Fearon, James D., 2004. ‘Why Do Some Civil
Bercovitch, Jacob & Richard Jackson, 1997. Wars Last So Much Longer Than Others?’,
International Conflict: A Chronological Encyclo- Journal of Peace Research 41(3): 275–301.
pedia of Conflicts and their Management, Gordon, Sanford C., 2002. ‘Stochastic
1945–1995. Washington, DC: Congressional Dependence in Competing Risks’, American
Quarterly Press. Journal of Political Science 46(1): 200–217.
362 jour nal of P E A C E R E S E A RC H volume 45 / number 3 / may 2008
Han, Aaron & Jerry A. Hausman, 1990. ‘Flexible Mason, T. David; Joseph P. Weingarten, Jr. &
Parametric Estimation of Duration and Patrick J. Fett, 1999. ‘Win, Lose, or Draw:
Competing Risk Models’, Journal of Applied Predicting the Outcome of Civil Wars’,
Econometrics 5(1): 1–28. Political Research Quarterly 52(2): 239–268.
Hartzell, Caroline; Matthew Hoddie & Donald Modelski, George, 1964. ‘The International
Rothchild, 2001. ‘Stabilizing the Peace After Relations of Internal War’, in James N.
Civil War: An Investigation of Some Key Rosenau, ed., International Aspects of Civil
Variables’, International Organization 55(1): Strife. Princeton, NJ: Princeton University
183–208. Press (45–91).
Hill, Daniel H.; William G. Axinn & Arland Phillipson, Coleman, 1916. Termination of War
Thornton, 1993. ‘Competing Hazards with and Treaties of Peace. New York: Duton.
Shared Unmeasured Risk Factors’, Sociological Pillar, Paul, 1983. Negotiating Peace: War
Methodology 23: 245–277. Termination as a Bargaining Process. Princeton,
Holbrooke, Richard, 1998. To End a War. New York: NJ: Princeton University Press.
Random House. Porsholt, Lars, 1966. ‘On Methods of Conflict
Holl, Jane E., 1993. ‘When War Doesn’t Work: Prevention’, Journal of Peace Research 3(2):
Understanding the Relationship between the 178–193.
Battlefield and the Negotiating Table’, in R Development Core Team, 2006. R: A Language
Licklider (269–291). and Environment for Statistical Computing.
Holsti, Kalevi J., 1966. ‘Resolving International Vienna: R Foundation for Statistical Computing.
Conflicts: A Taxonomy of Behavior and Some Regan, Patrick M., 1996. ‘Conditions for Suc-
Figures on Procedures’, Journal of Conflict cessful Third-Party Intervention in Intrastate
Resolution 10(3): 272–296. Conflicts’, Journal of Conflict Resolution 40(2):
Kesckemeti, Paul, 1958. Strategic Surrender: The 336–359.
Politics of Victory and Defeat. Stanford, CA: Regan, Patrick M., 2000. Civil Wars and Foreign
Stanford University Press. Powers: Outside Intervention in Intrastate Conflict.
Klingberg, Frank L., 1966. ‘Predicting the Ann Arbor, MI: University of Michigan Press.
Termination of War: Battle Casualties and Regan, Patrick M., 2002. ‘Third-Party Interventions
Population Losses’, Journal of Conflict and the Duration of Intrastate Conflicts’, Journal
Resolution 10(2): 129–171. of Conflict Resolution 46(1): 55–73.
Licklider, Roy, 1993a. ‘How Civil Wars End: Rosenau, James N., 1964. ‘Internal War as an
Questions and Methods’, in Licklider (3–19). International Event’, in James N. Rosenau, ed.,
Licklider, Roy, ed., 1993b. Stopping the Killing: International Aspects of Civil Strife. Princeton,
How Civil Wars End. New York: New York NJ: Princeton University Press (14–44).
University Press. Sarkees, Meredith, 2000. ‘The Correlates of War
Licklider, Roy, 1995. ‘The Consequences of Data on War: An Update to 1997’, Conflict
Negotiated Settlements in Civil Wars, Management and Peace Science 18(1): 123–144.
1945–1993’, American Political Science Review Singer, J. David; Stuart Bremer & John Stuckey,
89(3): 681–690. 1972. ‘Capability Distribution, Uncertainty,
Luttwak, Edward, 1999. ‘Give War a Chance’, and Major Power War, 1820–1965’, in Bruce
Foreign Affairs 78(4): 36–44. Russett, ed., Peace, War, and Numbers. Beverly
Marshall, Monty G. & Keith Jaggers, 2002. Polity Hills, CA: Sage (19–48).
IV Project: Political Regime Characteristics and Small, Melvin & J. David Singer, 1982. Resort to
Transitions, 1800–2002. Dataset Users’ Arms: International and Civil Wars, 1816–1980.
Manual. College Park, MD: Center for Beverly Hills, CA: Sage.
International Development and Conflict Suhrke, Astri & Lela Garner Noble, eds, 1977.
Management, University of Maryland. Ethnic Conflict in International Relations.
Mason, T. David & Patrick M. Fett, 1996. ‘How New York: Praeger.
Civil Wars End: A Rational Choice Approach’, StataCorp, 2005. Stata Statistical Software: Release
Journal of Conflict Resolution 40(4): 546–568. 9. College Station, TX: StataCorp, LP.
D ylan Balch-Lindsay et al. T H I R D -P A RT Y I N T E RV E N T I O N 363
Wagner, Robert Harrison, 1993. ‘The Causes of conflict processes in world politics and political
Peace’, in Licklider (235–268). methodology, particularly event history
Walter, Barbara F., 2002. Committing to Peace: models. We dedicate this article to his wife
The Successful Settlement of Civil Wars. Suzanne.
Princeton, NJ: Princeton University Press.
Walter, Barbara F., 2004. ‘Does Conflict Beget
Conflict? Explaining Recurring Civil War’, ANDREW J. ENTERLINE, b. 1966, PhD in
Journal of Peace Research 41(3): 371–388. Political Science (SUNY, Binghamton, 1998);
Wittman, Donald, 1979. ‘How a War Ends: Associate Professor, University of North Texas
A Rational Model Approach’, Journal of (2000– ); current main interests: third-party
Conflict Resolution 23(4): 743–763. involvement in conflicts, and the impact and
Wright, Quincy, 1965. A Study of War. Chicago, performance of externally imposed polities.
IL: University of Chicago Press.
Zartman, I. William, 1985. Ripe for Resolution:
Conflict and Intervention in Africa. New York: KYLE A. JOYCE, b. 1977, PhD candidate
Oxford University Press. in Political Science (Pennsylvania State
Zartman, I. William, 1993. ‘The Unfinished University); current main interests: interna-
Agenda: Negotiating Internal Conflicts’, in tional conflict, war expansion, computational
Licklider (20–36). and mathematical modeling, and quantitative
political methodology.