DRAFT – not for citation or distribution without permission
Civil War Intervention and the Problem of Iraq
Stephen Biddle, Jeffrey Friedman, and Stephen Long*
August 22, 2008
Civil war is the most common form of armed conflict worldwide, and has killed hundreds
of millions of people in the decades since World War II. It is also among the most pressing issues
in US foreign policy today, in the form of the ongoing civil war in Iraq.
There are many important theoretical and empirical questions in the study of civil warfare,
but one of the most immediately consequential concerns foreign intervention. Warfare internal to
a state is bad enough, but when neighboring states intervene, a local internal tragedy can become
a region-wide conflagration with far worse consequences for much greater populations. Its stakes
make the causes and incidence of civil war intervention an inherently important question for
scholarship.
But this is also a question at the very heart of today’s debate over US policy for Iraq. This
debate now turns on the need for a continued US troop presence and the consequences of US
withdrawal. And the case for a large US presence turns on the argument that withdrawal risks
destabilizing Iraq in a way that could cause the Iraqi civil war to spread across the region as
neighbors are drawn into the fighting, yielding a conflict that could engulf the whole of the
Middle East’s energy production and plunge hundreds of millions of additional people into open
warfare (e.g., National Intelligence Council 2007, Byman and Pollack 2007, Boot 2008).
Opponents of the US presence, by contrast, argue that such fears are exaggerated, and that the
war would either burn out within Iraq’s borders or that a US withdrawal would actually solve the
underlying problem and enable a peaceful settlement without a broader war (eg, Simon 2007,
*
Stephen Biddle is Senior Fellow for Defense Policy at the Council on Foreign Relations. Jeffrey
Friedman is a doctoral student at the Harvard Kennedy School of Government. Stephen Long is
Assistant Professor of Political Science at Kansas State University. Authors are listed in
alphabetical order. Correspondence should be addressed to Stephen Biddle, Council on Foreign
Relations, 1779 Massachusetts Avenue NW, Washington DC 20036, sbiddle@cfr.org, 202-5183476.
1
DRAFT – not for citation or distribution without permission
Gause 2008, Korb et al 2008). The outcome of this debate will shape perhaps the most
consequential foreign policy decision facing the next President, with stakes that stretch far
beyond the United States to the stability of an entire region.
Yet for all its importance, this debate has been almost devoid of any systematic scholarly
analysis of the actual risk that others would intervene in the Iraqi civil war should the US
withdraw from an unstable Iraq. There is an extensive body of empirical experience covering
over 140 civil wars since 1945, and there is a significant literature on civil war intervention in
these cases on which such an analysis could draw; this is an issue on which international relations
scholarship could potentially offer important insight. This existing literature is not yet directly
applicable, however. Its choices of dependent variables and units of analysis, for example, are
structured for related but different purposes, making its findings suggestive but not dispositive for
Iraq. And perhaps most important, several of the sub-issues most important for the Iraq debate –
and especially the causal roles of ethnic-sectarian linkages, regional political culture, and military
strength – have not been interconnected in a systematic or comprehensive way to date.
The purpose of this paper is thus twofold. First, we seek to extend the theoretical and
empirical literature on civil war intervention to encompass the joint causal roles of ethnicsectarian linkages, regional political culture, and military materiel among neighboring powers.
Second, we apply the results to the specific problem of Iraq by using the resulting model to
estimate the probability that this war would spread beyond Iraq’s borders in the event of a US
withdrawal.
We find that the particular circumstances of Iraq and the Gulf region today create an
important risk that the war could indeed spread if the United States exits and internal violence
escalates. Intervention is a commonplace feature of civil warfare generally, but Iraq’s particular
combination of multiple ethnic and sectarian linkages outweighs the material weaknesses of its
neighbors or the accommodationalist political culture of the region to make this an unusually
intervention-prone case. Even so, intervention is by no means a certainty. But our findings
2
DRAFT – not for citation or distribution without permission
suggest a nearly 40 percent probability that a ten-year civil war following a US withdrawal would
eventually draw in three or more of Iraq’s neighbors – and a much higher probability that the war
would spread to at least one. This scale of risk warrants serious consideration in the debate over
Iraq policy.
We present these findings in five steps. First, we review the extant literature on civil war
intervention, discuss the problems of ethnic-sectarian linkage, regional political culture, and
military materiel, and motivate our theoretical treatment of those variables. Next, we discuss our
dataset and operationalize our variables. We then present the statistical results. We conclude with
a series of implications from these results for scholarship and policy.
Explanations of Civil War Intervention
Civil war has been getting increasing attention in the international relations literature, and
two overlapping schools of intervention analysis have emerged from this. One school emphasizes
material geo-politics, and sees the causes of intervention in formal alliance ties, variations in
conflict type and stakes (identity wars are seen as more prone to intervention than ideological
ones), high conflict intensity and elevated casualty levels, the potential threat that civil wars pose
to the stability of their neighbors, and geographic proximity between civil war states and potential
interveners (Werner 2000, Regan 1998, Regan 2002, Lemke and Regan 2004, Findley and Teo
2006, Salehyan and Gleditsch 2006, Mullenbach 2008, Kathman np). Another school emphasizes
cultural and ethnic affinity, and argues that intervention is more likely when sectarian or ethnic
ties link civil war parties with potential interveners or when interveners and civil war states are
connected by a former colonial relationship (Heraclides 1990, Carment, James, and Rowlands
1997, Davis and Moore 1997, Davis, Jaggers, and Moore 1997, Khosla 1999, Saideman 2001,
Sambanis 2001, Centinyan 2002, Woodwell 2004, Austvoll 2006, Buhaug and Gleditsch 2008).1
1
Both groups see an important role for regime type (democracies are seen as less prone to
intervene), Cold War dynamics (Cold War cases are seen as more prone to intervention), and
African experience (African cases are generally seen as more prone to conflict) (Kathmann np,
Khosla 1999).
3
DRAFT – not for citation or distribution without permission
Iraq involves both geopolitical and ethno-sectarian factors simultaneously. The civil war
pits Iraqi Shia against Iraqi Sunnis in the middle of a region that has experienced long historical
conflict between these groups both within and between states, and in the midst of rising regional
tensions along just such sectarian lines (Nasr 2006). At the same time, many who argue that the
Iraqi war will not spread do so based on a geopolitical argument that Iraq’s Sunni neighbors are
too weak militarily to intervene (Takeyh et al. 2008, Posen 2007). To assess the net risk of
intervention in Iraq thus requires adjudication of potentially conflicting geopolitical and ethnosectarian influences.
In the literature, however, geopolitical and affinity variables tend to be treated in different
studies using different datasets, making net assessment of these effects for Iraq difficult. The
affinity literature tends to use the Minorities at Risk (MAR) data, which treat only contiguous
states and offer limited coverage of geopolitical issues. The geopolitical literature, by contrast,
tends to use the Correlates of War (COW) data, which do not cover ethnic or cultural affinity. A
recent paper by Martin Austvoll (2006) indicates that material and cultural factors may each be
significant determinants of civil war intervention, but the scope of this research is limited to
twenty-seven conflicts, which are suggestive but not conclusive of broader patterns.
Much of the literature, moreover, uses very broad definitions of “intervention.” Perhaps the
most common definition is Regan’s “convention-breaking military and/or economic activities in
the internal affairs of a foreign country targeted at the authority structures of the government with
the aim of affecting the balance of power between the government and opposition forces” (1998:
756, also used by Lemke and Regan 2004; Findley and Teo 2006; Austvoll 2006; Kathman np).
This has the virtue of excluding little, but it also includes much that would fall below the
threshold of concern in the Iraq debate, treating modest economic sanctions and large military
deployments equally. In addition, Regan codes separate observations of intervention every time a
country escalates its activities, weighting certain cases very heavily in the findings and producing
a higher intervention count than many in the Iraq debate would intuit. For instance, Rwanda
4
DRAFT – not for citation or distribution without permission
intervenes once in the Congo (1996-97), but Cuba intervenes 11 times in Angola’s civil war and
Vietnam intervenes 23 times in Cambodia. Others distinguish multiple levels of external
involvement, as in a range from “ideological encouragement” to “active combat units in country”
for the MAR data (eg, Saideman 2001; Cetinyan 2002); for the Iraq debate, only the most forceful
of these intervention definitions speak to primary concerns.
A final issue with the existing scholarly literature concerns its units of analysis. Most
recent empirical work on intervention has focused on interstate dyad-years. The benefit of this
approach is that it includes every potential intervener in a civil war, and thus avoids selecting on
the dependent variable. A drawback of this approach, however, is that it gives equal statistical
weight to all potential interveners, even those whose small size and remote location make them
implausible candidates for military action. By examining dyads where there is essentially no
chance of conflict, researchers may end up overshadowing other statistical patterns and reducing
a model’s relevance for assessing the likelihood of intervention in important cases like Iraq. The
utility of restricting analysis to “politically-relevant” dyads limited to proximate or great-power
interactions has attracted increasing attention (eg, Bennett 2006, Quackenbush 2006, Lemke and
Reed 2001), but most existing work on civil war intervention does not explicitly distinguish
relevant from other dyads.
Since 2007, the scholarly literature has been joined by policy analyses on the potential
causes of intervention in Iraq per se. In particular, Daniel Byman and Kenneth Pollack have
argued that civil wars involving a combination of ethnic links between neighbors and conflict
parties, refugee flows into neighboring states, and regional arms racing create a particular risk of
intervention, and that Iraq poses just such a combination (Byman and Pollack, 2007). Others, by
contrast, argue that Middle Eastern political culture promotes accommodation and compromise
rather than confrontation, and that Iraq’s neighbors are too weak to sustain a major cross-border
intervention in any case – hence the Iraq case should be considered a low risk for regionalization
(Takeyh et al. 2008, Posen 2007).
5
DRAFT – not for citation or distribution without permission
Neither Iraq camp’s analysis, however, has yet been informed by any systematic large-n
empirical investigation. Nor can this be provided by simple reference to the existing literature on
intervention. The affinity-oriented literature excludes the military materiel issues so important in
the Iraq debate, whereas the geopolitics-oriented literature excludes the ethnic and sectarian
factors. Neither scholarly sub-literature controls for Middle Eastern political culture and its
effects on intervention proclivity, nor does either sub-literature consider the effects of refugee
flows or arms-race dynamics on intervention.
To account for the effects considered important in the Iraq debate thus requires a synthesis
and extension of the available research on intervention. In particular, this requires a new dataset
with coverage of both affinity and geopolitical variables; it requires a dependent variable
operationalization that focuses on the more forceful forms of intervention pertinent to the Iraq
debate; and it requires an explicit treatment of region-specific features of the Middle East, the
effects of potential interveners’ military capability, the regional military balance, and the effects
of change in this balance as a prospective regional arms race in the Mideast unfolds.
Data
To test these hypotheses, we examine 142 civil wars between 1950 and 1999, each with a
minimum of 200 battle fatalities. Our baseline data are configured in dyad-years: for every year in
which a conflict is ongoing, there is a separate observation for every state in the international
system paired with the civil war state. We use this master set to produce results based on a
politically relevant dyads design.
Our dependent variable, intervention, is a dummy coded as 1 when a third party
intervenes in a conflict by sending combat troops into the civil war state. Each intervention must
involve state soldiers being sent across borders by an intervener for the first time in the civil war;
subsequent escalation, reinforcement, or other policy changes are not coded as additional
6
DRAFT – not for citation or distribution without permission
“interventions.”2 Others have operationalized “intervention” in much less restrictive ways. Regan
(2002), Lemke and Regan (2004), Austvoll (2006), and Kathman (np), for example, include
economic assistance, arms transfers, intelligence cooperation, or military advising, in addition to
armed border crossings; these authors also code secondary troop movements such as
reinforcements or escalation as additional interventions. While valid for their authors’ purposes,
these broader codings include as “interventions” an enormous range of cases that most in today’s
Iraq debate would not consider the primary focus of policy concern. (In Regan 2002, for example,
912 of 1036 total cases of “intervention” involve reinforcements of ongoing military action or
instances of assistance short of cross-border troop movements.) We thus adopt a very
conservative coding designed to speak both to the scholarly theoretical literature, whose interests
include high-end as well as more modest forms of external involvement, and the Iraq policy
debate – where the issue of central concern is the fear that the Iraq war will spread to engulf the
neighbors in active warfare per se.
To bridge the divide between affinity and geopolitical treatments of intervention and their
respective data sources, we begin with the COW data (Singer, Bremer and Stuckey 1972, Sarkees
2000, Hensel 2001, Gibler and Sarkees 2004b) given its broader coverage, and add data from
other sources as necessary to account for affinity variables and a variety of controls. Following
Lemke and Regan (2004), we adopt a less restrictive civil war definition than COW’s (200 or
more battle deaths, as opposed to COW’s 1000); this also requires us to add conflicts fitting this
definition but missing from COW, again following Lemke and Regan (2004), using Kathman’s
transformation of these data into a dyad-year unit of analysis via EUGene 3.0.3. We draw our
dependent variable values from Lemke and Regan (2004), but limited to cross-border troop
movements alone as noted above. We also use data on regime type from the POLITY IV dataset
(Marshall and Jaggers 2007) and Freedom House (2008); fatalities from the International Peace
2
This coding rule implies that dyad-years for ongoing interventions would be coded as zeros even though
the case involved a continuing intervention. To avoid bias from this effect, dyad-years for states that
intervene are dropped for years subsequent to the first cross-border troop movement.
7
DRAFT – not for citation or distribution without permission
Research Institute (Lacina and Gleditsch 2005) and Lemke-Regan (2004); economic performance
from Collier and Hoeffler (2004); conflict type from Lemke-Regan (2004); and refugee flows
from UNHCR (2000).
The most challenging data issues involve ethnic linkages. Fearon, Kasara, and Laitin
(2007) record the ethnicity of the “top political leader” in each state since 1945, but no data set
provides systematic information on rebel group ethnicity.3 We therefore compiled new data on
this, following Fearon et al.’s logic of coding groups by the ethnic background of their leaders.
Where rebellions comprised multiple factions we included each. We were able to document rebel
ethnicity for 139 of 142 civil wars in the data set.4 We then reviewed the merged data extensively,
with a particular emphasis on fatalities. As a result of this review, 29 of the 142 fatality values
were changed, five double-counts or erroneous civil war state identifications were corrected, and
we were able to code missing ethnicity values for more than 30 state leaders (recovering several
thousand observations in the master dataset).5
A key issue in the new coding was to determine whether an ethnic group in one country is the
“same” as in another. For robustness, we used two sets of coding rules, but both were quite
restrictive:

Rebel link v1, State link v1: Rebel link v1 is a dummy variable coded as 1 if the rebel
group and the potential intervener are the same group with the same name in each
country but the civil war state government is not, and 0 otherwise. State link v1 is a
dummy variable coded as 1 if the civil war state government and the potential intervener
3
The authors thank Kimuli Kasara at Columbia University for making these data available to us.
Sources for information on rebel groups include Clodfelter (2002), Minorities at Risk “Minority Group
Assessments” (Minorities at Risk 2002), Library of Congress country studies, Encyclopedia Britannica, as
well as individual sources particular to each conflict. We only included information on rebel ethnicity that
could be confirmed across multiple sources. Specific citations for each coding are provided in the data set.
5
We thank Patrick Regan for his assistance in identifying and correcting several errors in the 2004 dataset.
Examples of our corrections include the elimination of the double counting of the Biafran civil war and the
replacement of the civil war state in four conflicts from Congo-Brazzaville to Congo-Kinshasa. All
changes and additions to preexisting data are documented in the data manual, which is available from the
authors.
4
8
DRAFT – not for citation or distribution without permission
are the same group with the same name in each country but the rebel group is not, and 0
otherwise.

Rebel link v2, State link v2: In addition to the rule above, actors are coded as linked if
they are members of one of 18 ethnic “clusters” (e.g., Moldovans/Romanians, AmericoLiberians/Creole, Turks/Turkmen/Uighurs).
Operationalizations for other key variables are:

PotIntColonizer: a dummy variable with a value of 1 if the potential intervener was
formerly a colonizer of the civil war state, and 0 otherwise, including if the civil war state
was never a colony (Hensel 2001b).

CWSPowerShare: the share of material military power in the dyad controlled by the civil
war state, expressed as the civil war state’s CINC (Composite Index of National
Capability) score divided by the sum of the civil war state’s CINC score and the potential
intervener’s CINC score (Singer, Bremer and Stuckey 1972 v3.02).

PotIntAlliance: a dummy variable with a value of 1 if the potential intervener and the
civil war state have a formal alliance consisting of a “defense pact, neutrality or nonaggression treaty, or entente agreement,” and 0 otherwise (Gibler and Sarkees, 2004
v3.03).

Intensity: the natural logarithm of the average number of battle-related fatalities per
month of the conflict. We based our data on Regan (2002), and checked them against
figures from Clodfelter (2002), Correlates of War (Sarkees 2002), the International Peace
Research Institute (PRIO) Battle Deaths data (Lacina and Gleditsch 2005), and other
sources cited in the PRIO documentation. When Regan’s values differed from multiple
crosschecks by more than a factor of two, we replaced them with the figure given in
PRIO.6
6
There are two exceptions to this coding rule. Conflict #922 (Iran) was coded from Leitenburg (2003),
since the PRIO figure did not include battle-related civilian deaths for that conflict; conflict #971 (Iraq) was
9
DRAFT – not for citation or distribution without permission

Mideast: a dummy variable with a value of 1 if the civil war state is Iran, Turkey, Egypt,
Syria, Lebanon, Jordan, Israel, Saudi Arabia, the Yemen Arab Republic, Yemen, the
Yemen People’s Republic, Kuwait, Bahrain, Qatar, the UAE, or Oman.7

Refugees: the natural logarithm of the average number per year of refugees from the civil
war state residing in the potential intervener, using data from UNHCR (2000).8
Control variables include:

Coldwar: The Cold War superpower competition created an intervention incentive for
great powers that may not have been present since then. The Coldwar variable controls
for this potential influence, and is coded as a dummy variable with a value of 1 if the
dyad-year is before 1989, and 0 otherwise.

African: Sub-Saharan Africa is widely considered an unusually intervention-prone
region.9 The African variable controls for this potential influence, and is coded as a
dummy variable with a value of 1 if the civil war state is in Sub-Saharan Africa (COW
country codes 402 through 591) and 0 otherwise.

Resources: States whose economies turn on the export of oil or other primary
commodities are often considered attractive predation targets and subject to unique
internal political dynamics that might make them unusually intervention-prone. The
Resources variable controls for this potential influence, and is coded as the ratio of the
coded from Clodfelter, since PRIO did not record any value for battle deaths in that conflict. We also
produced robustness checks using an intensity variable based solely on PRIO codings, but this resulted in
the loss of more than 65% of our observations used in the main analysis due to missing data.
7
As a check on the robustness of this definition of the Middle East region, we examined another coding of
this variable that included North Africa and the Horn of Africa, the results of which are reported below.
8
UNHCR refugee data for years prior to 1965 contain a very high fraction of missing values; inclusion of
this variable thus effectively drops 25 of the 142 civil wars from the master dataset. The best fit model
reported in Table 1 below nevertheless includes refugees, but dropping this variable yields statistical results
very similar to Model 1 (the only notable changes are that CWSPowerShare becomes insignificant, while
PotIntAlliance moves to the 0.1 threshold of significance). Given that dropping refugees adds almost 2,000
observations, these are fairly minor changes. Where the UNHCR reported no refugees in a dyad, we
changed the value to 1, making the natural logarithm of the entry zero.
9
Cf. Regan (2004) which, contrary to most literature, finds Africa less prone to intervention than other
regions.
10
DRAFT – not for citation or distribution without permission
civil war state’s primary commodity exports to total GDP, coded in 5-year increments
following Collier and Hoeffler (2004).

Democratic state, Democratic intervener, Joint Democracy: Democracies are widely
expected to display distinctive security behavior that might reduce their proclivity to
intervene in civil warfare, or to be subject to intervention by others when beset by civil
warfare. To control for this potential influence, Democratic state is coded as a dummy
variable with a value of 1 if the civil war state scores 6 or greater in the POLITY IV data
set (Marshall and Jaggers, 2007); Democratic intervener is coded as a dummy variable
with a value of 1 if the potential intervening state scores 6 or greater in POLITY IV; Joint
Democracy is coded as a dummy variable with a value of 1 if the potential intervening
and civil war states both score 6 or greater in POLITY IV. POLITY contains a large
number of missing values, and where that was the case, we considered a state a
“democracy” if it was listed by Freedom House as “Free” (Freedom House 2008),
following the procedure in Regan (2002).
Analysis
Table 1 presents statistical findings for a probit analysis with robust standard errors on our
politically relevant dyad year data for our binary intervention/no intervention dependent
variable.10 Table 2 presents a comparison of the magnitude of substantive effects for each
10
All statistical analyses are conducted in Stata 10 using the probit command. All analyses have been
checked for specification error (using the “linktest” command) and multicollinearity (using the “collin”
command). The AIC and BIC measures of model fit show that Model 1 is superior in fit to alternatives that
change control variable operationalizations individually and in various combinations, such as using Rebel
link v2 and State link v2 instead of v1; using intensity numbers based only on PRIO instead of Regan et al
and PRIO; and including North Africa and the Horn of Africa in the Middle East. While the pseudo-r2
measures did not agree with the AIC and BIC on the best-fitting model, when selecting a primary model,
theoretical decisions should be weighed along with model fit statistics, and we believe that Model 1
represents the best combination of theoretically sound variables. We also performed negative binomial and
Poisson analyses on a version of the data in which civil war years, rather than civil war year dyads, made
up the observations, adjusting the variables to reflect that data structure, but the reduced number of
observations had severe effects on the statistical significance that we were able to observe in the models.
Details of the results, model fit, and regression diagnostics for each of these individual changes and each of
the combinations of these changes tested are available upon request from the authors. In all, the robustness
11
DRAFT – not for citation or distribution without permission
statistically significant variable from Model 1, showing changes in estimated net intervention
probabilities as key variables’ values are altered around their means. The results suggest several
key findings.
First, as the affinity literature suggests, ethnic linkages can be important for civil war
intervention. But not all linkages are created equal. In particular, Rebel link v1 is positive,
statistically significant at the .01 level, and substantively important: a potential intervener with an
ethnic link to rebels is more than five times as likely to intervene as one without a link (Table 2).
State link v1, however, is statistically insignificant with an opposite sign. The results thus imply
that links with rebels in the civil war state are much more conducive to intervention than links
with governments. Analyses that do not disaggregate thus risk underestimating the importance of
ethnic affinity for intervention, and the scale of the difference in empirical performance suggests
important theoretical gaps in the affinity literature: the logic of affinity clearly operates very
differently for rebels and governments.
Colonial history similarly matters for intervention. As the affinity literature expects, former
colonial powers are more likely to intervene in their former colonies’ civil wars: the
PotIntColonizer variable is significant at the .01 level and almost as important substantively as
ethno-sectarian linkages to rebel groups (the probability of intervention rises by more than a
factor of four for former colonizers: see Table 2).
checks make up more than 50 pages of regression output, so it is impractical to attempt to summarize the
results of every alternative model here.
12
DRAFT – not for citation or distribution without permission
Table 1: Probit Model of Intervention by a Politically Relevant Potential Intervener
Rebel link v1
State link v1
Coldwar
PotIntAlliance
CWSPowerShare
PotIntColonizer
Democratic state
Democratic intervener
JointDemocracy
Refugees11
Ethnic Conflict
Intensity
Resources
Mideast
African
Constant
Model I:
Full Model
Model II:
Affinity
and Controls
0.5944***
(0.2121)
-0.3816
(0.4000)
0.4184***
(0.1060)
0.2401*
(0.1322)
-0.3802**
(0.1685)
0.5355***
(0.1883)
-0.1234
(0.2480)
-0.4681***
(0.1452)
0.2033
(0.3642)
0.0384***
(0.0122)
0.1274
(0.1132)
0.0821***
(0.0323)
1.1250***
(0.4385)
-0.1623
(0.1540)
0.3638***
(0.1213)
-3.4991***
(0.2733)
0.6705***
(0.2034)
-0.1271
(0.4038)
0.3825***
(0.1070)
0.5978***
(0.1872)
-0.2263
(0.2353)
-0.4424***
(0.1403)
0.2248
(0.3484)
0.0441***
(0.0118)
1.2681***
(0.4251)
-0.1273
(0.1588)
0.4331***
(0.1165)
-3.0991
(0.1574)
N=8594
Wald 2 (15)=107.55
Prob >2=0.0000
M&Z’s R2=0.249
McF’sAdj R2=0.132
N=8612
Wald 2 (11)=99.31
Prob >2=0.0000
M&Z’s R2=0.202
McF’sAdj R2=0.123
Model III:
Geopolitics
and Controls
0.4150***
(0.1034)
0.2495*
(0.1301)
-0.5285***
(0.1687)
-0.1788
(0.2509)
-0.3084**
(0.1211)
0.2050
(0.3633)
0.0467***
(0.0113)
0.1692
(0.1100)
0.0737**
(0.0308)
0.9340**
(0.4107)
-0.1309
(0.1473)
0.3699***
(0.1145)
-3.4188
(0.2593)
N=9432
Wald 2 (12)=94.97
Prob >2=0.0000
McK&Z’s R2=0.226
McF’s Adj R2=0.116
*** p-value 0.01 or less; ** p-value 0.05 or less; * p-value 0.10 or less
11
Refugees, fatalities, and intensity use the natural log of each, using a minimum value of 1 for refugees to
undefined expressions.
13
DRAFT – not for citation or distribution without permission
A second major finding is that affinity alone is an inadequate explanation of intervention:
geopolitical variables matter, too. Material military power is significant at the .05 level, and its
effects are substantively important (though less so than for ethnic linkages with rebel groups or
colonial relationships): the materially weakest interveners in the dataset have a predicted
probability of intervention 71 percent lower than the strongest. Combat intensity is significant at
the .01 level and is extremely important: intervention is nearly 16 times more likely when the
intensity level is at its lowest in the estimation sample than when it is at its highest.12 But not all
geopolitical variables noted in the literature are consequential: conflict type does not pass even
the .1 threshold for statistical significance, while alliance relationships are significant only at the
.1 level.
The results also show that to consider either affinity or geopolitics in isolation is to produce
an unnecessarily narrow and potentially biased picture. To illustrate the point, Model II in Table 1
considers only affinity variables plus controls; Model III considers only geopolitics plus controls.
Neither partial model performs as well as Model I; statistical performance for the partial models
varies by measure but can exceed a 30 percent falloff, as in the Efron’s r2 for Model III.13 And
omitted variable bias, while generally modest, can sometimes affect coefficients in problematic
12
While this change in predicted probability seems remarkably high, it is important to remember that the
predicted probability of intervention when all variables are at their means (continuous variables) or modes
(dummy variables), the probability of intervention in a given civil war state-third party dyad year is less
than one fifth of one percent. Percentage changes in such a low probability of intervention can appear
large, while still reflecting a relatively low chance of intervention (a 383% increase over the base level
holding all variables at their means or modes, for instance, still results in only a 0.87% predicted
probability of intervention by a given politically relevant potential intervener in a given year of a given
civil war).
13
There is no consensus on a single best measure of fit quality for probit. We considered eight different
approaches, six variants on pseudo-r2 measures (McKelvey and Zavoina, McFadden, McFadden adjusted,
Cox-Snell, Cragg-Uhler and Efron), and two information criterion measures (Akaike, or AIC, and
Bayesian, or BIC). All pseudo-r2 measures show a loss of fit quality for Models II and III, with a degree
ranging from 7% (0.132 to 0.123) for McFadden’s adjusted r 2 in Model II to 33% (0.03 to 0.02) for Efron’s
r2 in Model III. AIC and BIC show less difference and the BIC actually implies a slight improvement in fit
quality for the partial models (the BIC score for Model I, for example, is -77092.333; the score for Model II
is -77293.963 and that for Model III is -85531.857), though this is unrepresentative of the range of
measures overall. For the Model I-II comparison specifically, loss of fit quality by pseudo-r2 approaches
varied between 7% for McFadden’s adjusted r2 (0.132 to 0.123) and 19% for McKelvey and Zavoina’s r2
(0.249 to 0.202); the loss of fit quality between Model I and Model III varied between 9% for McKelvey
and Zavoina’s r2 (0.249 to 0.226) and 33% for Efron’s r 2 (0.03 to 0.02).
14
DRAFT – not for citation or distribution without permission
ways: in Model III, for example, the estimated effect of military materiel is biased upward by
about 40 percent relative to the more complete analysis in Model I, and the effect of democracy in
the potential intervener is diminished by more than 50 percent.
Table 2: Model 1 Effects on the Predicted Probability of Intervention
Covariates
Rebel link v1
Coldwar
PotIntColonizer
CWSPowerShare
Democratic intervener
Intensity14
Refugees
Resources
African
Min
0
0
0
0.0002
0
-0.4336
-3.4657
0.006
0
Predicted p
0.0018
0.0018
0.0018
0.0028
0.0018
0.0004
0.0008
0.0012
0.0018
Max
1
1
1
0.9993
1
10.0971
14.6840
0.7940
1
Predicted p
0.0102
0.0063
0.0087
0.0008
0.0004
0.0067
0.0073
0.0158
0.0054
%∆
+467%
+250%
+383%
-71%
-78%
+1575%
+813%
+1217%
+200%
Predicted probabilities calculated holding continuous variables at their means and dummy variables
at their modes. Means, minimums, maximums and modes are for the estimation sample. Only
variables with a significance level of 0.05 or better are presented.
The results suggest several other findings of note. Democracy’s effects on international
politics have been a major theme in recent scholarship, and Model 1 suggests that regime type is
indeed important for intervention: Democratic intervener is significant at the .01 level and
substantively important. Yet Model 1 also suggests that, like ethnic linkage, democracy’s effects
on intervention are differentiated. Democratic outsiders are significantly less likely to intervene in
civil wars, but democracies undergoing civil wars are no more or less likely to suffer intervention
than others, and democratic outsiders are not significantly more likely to intervene when the civil
war state is a fellow democracy than when it is not.
The Refugees variable is significant at the .01 level, and its effects are substantively
important: the potential intervener containing the highest average of refugees per civil war year in
the sample is over eight times more likely to intervene in a given year than a potential intervener
14
While it may seem odd for intensity and refugees to have negative minimum values, remember that these
are the natural logs of the actual values. In some cases, the actual value of these per-month averages are
between 0 and 1, leading to a negative result after logging. This does not mean that we are assuming or
imputing the possibility of negative casualties or refugees, and the logic of using and interpreting natural
logs here still holds.
15
DRAFT – not for citation or distribution without permission
with the lowest average of refugees per civil war year in the sample. The nature of the civil war
state’s economy also matters: the Resources variable is significant at the .01 level, and even more
important than ethno-sectarian linkage in its effects: the case with the highest ratio of primary
commodity exports to GDP is more than twelve times more likely to experience intervention than
the case with the lowest ratio. States with primary commodity-centered economies are thus
substantially more likely to be targeted for intervention in the event of civil war than are states
with other economic foundations. The Cold War appears to have involved distinctive dynamics:
Cold War civil wars were more prone to intervention than subsequent conflicts, and the effect is
both statistically significant at the .01 level and fairly important substantively (post-Cold-War
civil war year dyad observations are about 71 percent less likely to receive intervention: see Table
2).
Regional distinctions can be important. African civil wars, for example, are two times more
susceptible to intervention than elsewhere. But here, too, the effect is not universal – and in
particular, Middle Eastern civil wars are not meaningfully less likely than others to see
intervention. Thus there is no evidence in these data to support the claim that Middle Eastern
states are unusually free of intervention risk by virtue of a distinctive political culture.
Conclusions and Implications
Our findings thus suggest that ethno-sectarian affinity and geopolitical dynamics can be
significant contributors to the risk of outside intervention in civil warfare. In particular, links
between the civil war rebel group and the governments of neighboring states significantly
increase the risk that neighbors will intervene by sending troops across the border, as do material
power advantages for potential interveners and high levels of combat intensity in the civil war. In
addition, a number of factors noted in the policy debate but largely absent from the scholarly
literature on civil war intervention also have important effects: refugee flows and primary
commodity dependence both demonstrate empirical performance with causal importance
comparable to or greater than that of many variables which have heretofore received greater
16
DRAFT – not for citation or distribution without permission
attention in the civil war literature. The results improve our understanding of intervention in
particular and civil war in general, and suggest the importance both of combining geopolitical and
affinity approaches to the study of internal warfare, and of extending analysis to new explanatory
variables outside either tradition.
These findings also have important implications for US foreign policy and the debate over
Iraq. In particular, they imply an important danger that the Iraq war could spread if the US
withdraws and internal violence in Iraq escalates. Many proponents of prompt withdrawal have
argued that the risk of intervention is exaggerated, and this assessment is often supported by
arguments that Iraq’s neighbors are too weak, or that Middle Eastern states resolve differences by
appeasement rather than invasion (e.g. Takeyh et al. 2008, Simon 2007, Posen 2007, Gause 2008,
Korb et al 2008). The findings above, however, imply that the unique features of Iraq and its
neighborhood could have the opposite effect – increasing, not decreasing, the risk of intervention
in this war relative to others. Iraq presents an unusually interconnected ethno-sectarian conflict in
a neighborhood with a large number of potential interveners who share the Sunni majority
populations and/or regimes of Iraq’s Sunni insurgency. The neighbors are relatively weak now,
but so is Iraq, and the region’s ongoing arms race stands to increase those neighbors’ material
capacity to intervene over time. Iraq is also a major oil exporter, a major source of destabilizing
refugee flows into neighboring countries, and a state with a civil war of very high average
intensity. These factors are all strongly linked with an elevated risk of intervention in the data as a
whole. And the findings above show no reason to expect that anything unique to the Middle East
region per se should imply any unusual freedom from danger: whereas Africa, for example, is an
especially intervention-prone region, the Mideast is not significantly different from the rest of the
world in this regard.
This does not imply a certainty of intervention in a post-withdrawal Iraq. The findings
above are based on necessarily imperfect data, and statistical analyses never explain the totality of
the variance in the data; a degree of caution is always in order in drawing policy implications
17
DRAFT – not for citation or distribution without permission
from empirical analysis. It is also possible that US withdrawal could reduce rather than increase
internal violence within Iraq (though this is unlikely: Biddle, O’Hanlon, and Pollack 2008). Nor
does the empirical analysis above suggest anything like a guarantee of disaster in the event that
Iraqi internal violence does escalate in the wake of a withdrawal.
But the odds of intervention implied by the findings above are daunting all the same. To
explore this more concretely, we extrapolated from the individual-state intervention probabilities
estimated in the probit analysis above to compute the net probability of one-, two-, or more-thantwo-state interventions in a post-withdrawal Iraq war using a Monte Carlo simulation
methodology. That is, we computed unique time-dependent intervention probabilities per year for
each of Iraq’s neighbors using the coefficients in Model I; compared these to random number
draws for each neighbor in each simulated year; scored a neighbor as having intervened if the
random draw was below the computed probability for any simulated year; then summed the
interventions over potential interveners and over simulated time.15 Table 3 presents the mean
15
In these simulations, Bahrain and Iran are coded as having ethnic links to the civil war state;
Jordan, Kuwait, Qatar, and Saudi Arabia have links to the rebels; Syria is majority Sunni but is
ruled by Alawites, who are a Shi'a offshoot, so we ran the simulation twice, once with Syria
coded as linked to the rebels and once with it linked to the state. The results presented in Table 3
are based on coding Syria as linked to the rebels (Syria’s Sunni majority to Iraq’s Sunni rebels).
The alternative coding (linking Syria’s Alawite leaders to Iraq’s government) yields different, but
still high, estimates for at least one intervention within 5 years (63%), 10 years (87%), and 15
years (96%), for example. Dummies for alliances, civil war democracy, joint democracy, the
Cold War, and Africa are all set to zero. Turkey is coded as a former colonizer of Iraq. CINC
scores for Iraq and its neighbors were calculated for 2008 using the most up-to-date information
possible (CIA 2008, International Energy Agency 2008, International Institute for Strategic
Studies 2008, International Iron and Steel Institute 2008; documentation available from the
authors on request). We assume that current trends in regional arms acquisition will continue,
and that Iraq's neighbors' CINC scores will grow by roughly 10 percent per year (though the
results are largely insensitive to this assumption: if CINC scores are held constant over time the
probability of greater than zero interventions falls from 0.73 in 5 years to 0.70; from 0.93 in 10
years to 0.91; and from 0.98 in 15 years to 0.97). Refugee counts were drawn from the UNHCR's
2007 tabulation. The intensity of the Iraq civil war was calculated by taking the Iraq Body
Count's estimate of total violent deaths among the Iraq Security Forces and Iraqi civilians from
the beginning of 2004 through the end of 2007 (roughly 95,000), adding 4,000 for the number of
American troops killed in Iraq, and adding twice that number (8,000) as an estimate of the
number of insurgents killed in Iraq. For Iraq's ratio of primary commodity exports to total GDP,
we used the figure reported by Collier and Hoeffler for 1999.
18
DRAFT – not for citation or distribution without permission
results from more than 4,000 replications of the simulation, broken out for results observed in
five, ten, and fifteen years of simulated post-withdrawal civil warfare.
Table 3: Net probability of intervention
Within
5 yrs
10 yrs
15 yrs
0 states
0.27
0.07
0.02
>0 states
0.73
0.93
0.98
>1 state
0.32
0.69
0.88
>2 states
0.09
0.37
0.64
>3 states
0.01
0.12
0.33
The results suggest that the probability of intervention by at least one neighbor become
extremely high after even five years of post-withdrawal warfare, with over a 70 percent estimated
probability of at least one intervention, and over a 30 percent probability of more than one.
Within ten years of withdrawal, the probabilities rise above 90 percent for at least one
intervention and approach 70 percent for more than one. And within 15 years these probabilities
rise to 98 and 88 percent, respectively. The odds of three or more of Iraq’s neighbors intervening
in the war approach 40 percent within 10 years, and exceed 60 percent if the war continues for 15
years.
Table 4: Country-specific results (probability of intervention for each state)
Bahrain
Iran
Jordan
Kuwait
Qatar
Saudi
Arabia
Syria
(Sunni)
Syria
(Shiite)
Turkey
5 years
.0081
.0193
0.1763
0.1413
0.1363
0.2168
0.2632
0.0271
0.1647
10 years
.0169
0.0377
0.3199
0.2692
0.2731
0.3889
0.4611
0.0538
0.3190
15 years
.0230
0.0599
0.4358
0.3707
0.3903
0.5239
0.5997
0.0837
0.4334
Within
The simulation also enables an examination of the identity of prospective interveners and
the relative magnitude of risk across states in the region. These findings are presented in Table 4,
which reports the estimated probability of intervention for each of Iraq’s neighbors assuming five,
19
DRAFT – not for citation or distribution without permission
ten, and fifteen years of post-withdrawal civil warfare.16 Given Monte Carlo simulation results
based on the statistical findings in Model I, the simulation implies that the greatest threats of
intervention are from Saudi Arabia, Jordan, and Turkey. Syria poses unusual coding complexities
given its Sunni majority population but heterodox Alawite Shiite regime: if coded by reference to
its Sunni majority, it is likelier than Saudi Arabia to intervene; if coded by reference to its Shiite
leadership, it is less likely than Kuwait or Qatar to intervene.
Note that none of these individual-state intervention probabilities exceed 50 percent for 10
years of warfare; only two (Saudi Arabia and Syria coded as Sunni) exceed even a one-in-three
chance of intervention in ten years. But because Iraq has many neighbors, even modest
probabilities of intervention individually cumulate into serious risks over time.
Interestingly, of these neighbors Iran is among the least likely to intervene, with less than a
10 percent probability of intervention after even 15 years of Iraqi civil warfare – it is
overwhelmingly Iraq’s Sunni neighbors who pose the greatest intervention risk. To some extent
this is a function of the absence of contagion or reaction dynamics in Table 1’s statistical
modeling; in the results here, all interventions are considered statistically independent events. In
reality, the odds of counter-intervention following an initial entry by a rival are likely to be much
higher than those for initial interventions (Findley and Teo 2006). In particular, an entry by a
Sunni state into an Iraqi civil war would probably swing the military balance within Iraq
dramatically in the Sunni rebel’s favor, and this would greatly increase Shiite Iran’s incentives to
counter-intervene in order to avoid a Sunni takeover of Iraq. But considered in the broader
context of the empirical record as a whole, these results suggest that Iran is unlikely to be the first
state to cross the border with uniformed military formations, and may pose a smaller danger of
regionalization for the conflict than do Iraq’s Sunni neighbors.
16
Values in Table 4 are the fraction of all (10,000) simulation replications in which the given state
intervened in an Iraqi civil war assumed to be ongoing as of the given time.
20
DRAFT – not for citation or distribution without permission
Of course, none of these values reach unity, and for the odds of multiple interventions to
reach dangerous levels requires multiple years of post-US warfare in Iraq. Intervention is far from
a certainty, whether for any given neighbor or across the region as a whole. And it cannot be
known how long the Iraqi civil war would continue after a US withdrawal – it could last less than
five years or more than fifteen.
But given the potential consequences – both strategic and humanitarian – of regional
warfare in the Persian Gulf, the results in Tables 3 and 4 are grounds for concern. Certainly these
findings give no basis for dismissing the danger of regional intervention in the Iraq war if the
United States withdraws. This is a nontrivial risk which must be considered carefully in the
design of US policy for Iraq and in any planning for troop reductions there – it cannot safely be
ruled out on the basis of a belief that Iraq’s unique conditions make the war unlikely to spread.
And the results also suggest some priorities for US diplomatic efforts to contain any postwithdrawal warfare within Iraq’s borders in the event that the conflict re-escalates after a US
departure. In particular, the ethno-sectarian affinity between the Sunni insurgency in Iraq and its
Sunni neighbors demands special attention – and greater concern than does Shiite Iran to the east,
at least for the problem of containing an Iraqi civil war.
Perhaps most broadly of all, the results above suggest that important policy debates need
not be conducted in isolation from empirical scholarship in international relations. The theoretical
and empirical literature has much to offer if framed appropriately and extended where necessary
to account for the particular issues at stake. Yet such decisions are often made in the absence of
any systematic consideration of the range of evidence and experience that empirical scholarship
can consider. Knowledge is important in its own right. But where the stakes in public decision are
as grave as those in war and peace, opportunities to apply knowledge to inform public debate can
be – and should be – exploited more often.
21
DRAFT – not for citation or distribution without permission
References
Austvoll, Martin. 2006. “Transnational Ethnic Dimensions of Third-Party Interventions in Civil
Wars.” Presented at the Annual Meeting of the American Political Science Association,
Philadelphia, PA.
Bennett, D. 2006. “Exploring Operationalizations of Political Relevance.” Conflict Management
and Peace Science 23 (3): 245-261.
Biddle, Stephen, Michael O’Hanlon and Kenneth Pollack, 2008. “How to Leave a Stable Iraq:
Building on Progress,” Foreign Affairs 87 (5): 40-58.
Buhaug, Halvard, Kristian Skrede Gleditsch. 2008. “Contagion or Confusion? Why Conflicts
Cluster in Space. International Studies Quarterly 52 (2): 215-233.
Boot, Max 2008. “How Not to End the War,” Washington Post, March 31.
Carment, David, Patrick James, Dane Rowlands. 1997. “Ethnic Conflict and Third Party
Intervention: Riskiness, Rationality and Commitment.” In Enforcing Cooperation: ‘Risky’
States and Intergovernmental Management of Conflict, ed. Gerald Schneider and Patricia
Weitsman eds. London: Macmillan.
Cederman, Lars-Erik and Luc Girardin. 2007. “Beyond Fractionalization: Mapping Ethnicity onto
Nationalist Insurgencies. American Political Science Review 101 (1): 173-185.
Cetinyan, Rupen. 2002. “Ethnic Bargaining in the Shadow of Third-Party Intervention.”
International Organization 56 (3): 645-677.
Clodfelter, Micheal. 2002. Warfare and Armed Conflicts: A Statistical Reference to Casualty and
Other Figures, 1500-2000. 2nd ed. Jefferson, NC: McFarland & Company, Inc.
Collier, Paul and Anke Hoeffler. 2004. “Greed and Grievance in Civil War.” Oxford Economic
Papers 56 (4): 563-595.
Davis, David R. and Will H. Moore. 1997. “Ethnicity Matters: Transnational Ethnic Alliances
and Foreign Policy Behavior.” International Studies Quarterly 41 (1): 171-184.
Davis, David R., Keith Jaggers, and Will H. Moore. 1997. “Ethnicity, Minorities, and
International Conflict.” In Wars in the Midst of Peace: The International Politics of
Ethnic Conflict, ed. David W. Carment and Patrick James eds. Pittsburgh: University of
Pittsburgh Press.
Fearon, James D., Kimuli Kasara, and David D. Laitin. 2007. “Ethnic Minority Rule and Civil
War Onset.” American Political Science Review 101 (1): 187-193.
Findley, Michael G., Tze Kwang Teo. 2006. “Rethinking Third-Party Interventions into Civil
Wars: An Actor-Centric Approach.” Journal of Politics 68 (4): 828-837.
Freedom House. 2008. Freedom in the World 2008. New York: Freedom House.
22
DRAFT – not for citation or distribution without permission
Gause, F. Gregory III. 2008. “Iraq 2012: The Regional Context. Testimony before the Committee
on Foreign Relations, US Senate.”
Gent, Stephen E. 2007. “Strange Bedfellows: The Strategic Dynamics of Major Power Military
Interventions.” Journal of Politics 69 (4): 1089-1102.
Gibler, Douglas M., and Meredith Reid Sarkees. 2004. “Measuring Alliances: The Correlates of
War Formal Interstate Alliance Dataset, 1816-2000.” Journal of Peace Research 41 (2): 211222.
Hensel, Paul. 2006. Issue Correlates of War (ICOW) Colonial History Data Set version 0.4.
Heraclides, Alexis. 1990. “Secessionist Minorities and External Involvement.” International
Organization 44 (3): 341-378.
Kathman, Jacob D. Unpublished. “Civil War, Contagion, and Intervention.” University of North
Carolina, Chapel Hill.
Korb, Lawrence J., Brian Katulis, Sean Duggan, Peter Juul. 2008. How Does this End? Strategic
Failures Overshadow Tactical Gains in Iraq. Washington, DC: Center for American Progress.
Khosla, Deepa. 1999. “Third World States as Intervenors in Ethnic Conflicts: Implications for
Regional and International Security.” Third World Quarterly 20 (6): 1143-1156.
Lacina, Bethany, and Nils Petter Gleditsch. 2005. “Monitoring Trends in Global Combat: A New
Dataset of Battle Deaths.” European Journal of Population 21 (2-3): 145-166.
Lemke, Douglas, and William Reed. 2001. “The Relevance of Politically Relevant Dyads.”
Journal of Conflict Resolution 45 (1): 126-144.
Lemke, Douglas and Patrick M. Regan. 2004. “Interventions as Influence.” In The Scourge of
War, ed. Paul F. Diehl. Ann Arbor: University of Michigan Press.
Marshall, Monty G., and Keith Jaggers. 2007. “Polity IV Project: Political Regime Characteristics
and Transitions, 1800-2006.”
Mullenbach, Mark, Gerard P. Matthews. 2008. “Deciding to Intervene: An Analysis of
International and Domestic Influences on United States Interventions in Intrastate Disputes.”
International Interactions 34 (1): 25-52.
Nasr, Vali. 2006. The Shia Revival: How Conflicts within Islam Will Shape the Future. New
York. W.W. Norton & Company, Inc.
National Intelligence Council 2007. Prospects for Iraq’s Stability: A Challenging Road Ahead.
National Intelligence Estimate, January
Posen, Barry R. 2007. “The Risks of Staying vs. Leaving Iraq.” Boston Globe, April 19.
Quackenbush, Stephen L. 2006. “Identifying Opportunity for Conflict: Politically Active Dyads.”
Conflict Management and Peace Science 23 (1): 37-51.
23
DRAFT – not for citation or distribution without permission
Regan, Patrick M. 1998. “Choosing to Intervene: Outside Intervention in Internal Conflicts.”
Journal of Politics 60 (3): 754-779.
Regan, Patrick M. 2002. “Third-Party Intervention and the Duration of Intrastate Conflicts.”
Journal of Conflict Resolution 46 (1): 55-73.
Sambanis, Nicholas. 2001. “Do Ethnic and Nonethnic Civil Wars Have the Same Causes?: A
Theoretical and Empirical Inquiry (Part 1).” Journal of Conflict Resolution 45 (3) 259-282.
Sambanis, Nicholas. 2005. “Using Case Studies to Refine and Expand the Theory of Civil War.”
InUnderstanding Civil War: Evidence and Analysis, ed. Paul Collier and Nicholas Sambanis,
eds. Washington. The World Bank.
Saideman, Stephen M. 2001. The Ties that Divide: Ethnic Politics, Foreign Policy, and
International Conflict. New York: Columbia University Press.
Sarkees, Meredith Reid. 2000. “The Correlates of War Data on War: An Update to 1997.”
Conflict Management and Peace Science 18 (1): 123-144.
Simon, Steven N. 2007. After the Surge: The Case for U.S. Military Disengagement from Iraq.
New York: Council on Foreign Relations Special Report No. 23.
Singer, J. David, Stuart Bremer, and John Stuckey. 1972. "Capability Distribution, Uncertainty,
and Major Power War, 1820-1965." In Peace, War, and Numbers ed. Bruce Russett. Beverly
Hills: Sage.
Stinnett, Douglas M., Jaroslav Tir, Philip Schafer, Paul F. Diehl, and Charles Gochman. 2002.
"The Correlates of War Project Direct Contiguity Data, Version 3." Conflict Management and
Peace Science 19 (2): 58-66.
Takeyh, Ray, John Podesta, and Lawrence J. Korb. 2008. “A War We Must End.” Washington
Post, February 26.
Werner, Suzanne. 2000. “Deterring Intervention: The Stakes of War and Third-Party
Involvement.” American Journal of Political Science 44 (4): 720-732.
Woodwell, Douglas. 2004. “Unwelcome Neighbors: Shared Ethnicity and International Conflict
During the Cold War.” International Studies Quarterly 48 (1): 197-223.
24