Académique Documents
Professionnel Documents
Culture Documents
Relationship.
Michael Koch
Assistant Professor
Department of Political Science
Texas A&M University
mtkoch@polisci.tamu.edu
Zowie Hay
Graduate Student
Department of Political Science
Texas A&M University
zowiehay@polisci.tamu.edu
Douglas W. Kuberski
Graduate Student
Department of Political Science
Texas A&M University
dkuberski@polisci.tamu.edu
We thank Christopher Sprecher and Randolph Siverson for their comments and
suggestions. However, all mistakes remain solely ours.
Abstract: Quantitative studies of interstate conflict usually include a variable to control
for the effect of alliances on conflict behavior, yet the results of these studies vary quite
considerably. We argue that the reason for these varying results is that the impact of
alliances between states on conflict is conditioned by geographic distance, at least in
terms of how these measures are commonly operationalized in the conflict literature. We
provide an argument for why the influence of alliances is conditioned by geography. We
then demonstrate that inter-acting the two measures provides consistent results by
replicating a number of studies, both monadic and dyadic in nature, which included a
measure controlling for alliances.
1
In “Dangerous Dyads” (1992) Stuart Bremer noted that both the absence of an
alliance and whether states where contiguous or not were the two most important factors
interstate conflict has heeded the words of Bremer and included measures controlling for
both shared alliance membership and spatial proximity. The quantitative conflict
research over the past 10 years provides ample evidence that alliances are still thought to
influence the probability of conflict between states. Arguments range from how alliances
reduce conflict between states to how they might exacerbate conflict between states.
Among the most common reasons cited, alliances are thought to reduce the probability of
differences without resorting to arms. Conversely others suggest that they increase the
probability of conflict because allies may value similar things but their preferences may
diverge, giving allies more to fight over. And while there are different theoretical
alliances are seen as integral to understanding conflict among nations. Reflecting this
importance, much of the quantitative research on the causes of inter-state conflict has
included some measure of alliance commitments. For example of the nine quantitative
in 2006, eight included a measure of whether an alliance was present between dyads
1
For a summaries of the research on whether alliances increase or decrease conflict, see both Sprecher and
Krause (2006) and Kimball (2006).
2
while the other controlled for alliance similarity. Similarly, the Journal of Conflict
between 2005 and 2006. Of these ten, eight included a control measure for alliances.
conflict? If researchers are going to include measures of alliances and proximity, either
to control for alternative explanations or because they might influence the dependent
variable under study, then it is important to properly specify those measures.2 In this
article we suggest that one of the reasons for conflicting findings or even null findings
concerning the influence of alliances on conflict is that the use of alliances as a control
conditioned by geographic proximity. In this paper we briefly review the literature on the
conditions under which alliances between states lead to conflict or peace. We then review
some of the empirical work that incorporates alliances both monadically and dyadically
as “controls” and the results that have emerged from this research. Next we discuss how
geographic distance might influence the relationship between alliances and conflict. We
then present an alternative specification of alliances and geographic proximity and test its
Overall, the results suggest that there is a strong relationship between alliances, the
The most common explanation for controlling for the presence or absence of
alliances is that alliances are a catalyst for peace. By clarifying commitments to defense,
3
belligerents more cautious (Morgenthau, 1967). In other words, alliances are thought to
deter aggression by enhancing the military capability of potential military targets (Levy,
1981; Smith, 1995). Moreover, alliances are seen as enhancing peace when they are
primarily perceived as sending a signal that is defensive in nature (Gibler, 2000; Leeds,
2003). For example, when alliances are signed during times of relative peace between
countries with no clashing interests, or, signed between states that have failed in their
previous war experience, they lead to peace because of the specific signals that they send
to other states about the intentions of the states involved (Raknerund and Hegre, 1997;
Gibler and Vasquez, 1998). Alliances also allow for states to potentially constrain their
allies’ behavior, meaning states with allies are less likely to initiate conflict (Ireland and
promoting peace, whereby states who are signatories of a particular alliance are better
able to send information to each other and eliminate uncertainty with potential
adversaries.
(1981) expected utility model suggests that allies are more likely to engage in bilateral
conflict because alliances between states reduce the likelihood of third party intervention.
Additionally, as states alter their foreign policies this can lead to disagreements over
security issues and increase allies’ incentives to attack one another. Furthermore,
future conflict (Morgenthau, 1967; Levy, 1981; Smith, 1995; Gibler and Vasquez, 1998).
By arousing suspicion in other states, alliances have the potential to incite conflict
4
because they spark a security dilemma and have the potential to entrap states who are
concerned about national prestige and the ramifications of not fulfilling their alliance
Given the importance that international relations scholars have placed on the role
of alliances and the probability of conflict, it is not surprising that alliances have taken on
such a prominent role in studies of IR. This importance has been reflected in much of the
“Dangerous Dyads” (1992) his bivariate analysis revealed that war is more likely to occur
between allies. However, results of his multivariate analysis revealed that the absence of
an alliance between states leads to an increased probability of war. Even though these
mixed results suggested that the presumed relationship between alliances and war was
spurious, or perhaps conditioned by some other variable, Bremer argued that both the
absence of an alliance and whether states where contiguous or not were the two most
conflict. In their dyadic assessment of the “liberal” peace Oneal and Russett (2001)
include a control for alliances, and find a negative relationship between alliance
commitments and conflict. More recently, Clark and Regan (2003) provide an alternative
statistical method to model the role of opportunity and willingness to fight in the study of
conflict. They also find that an alliance significantly reduces the willingness and hence
the probability that two states will fight. Gartzke (2007), in testing a theory of liberal
5
peace based on capitalism and common interstate interests rather than regime type, also
However, others have found that alliances often have no effect on the probability
of conflict (Barbieri, 1996; Benson 2005; Peceny, Beer and Sanchez-Terry, 2002). For
conflict suggests that alliances have no significant impact on the probability of a dispute
dyads finds no statistically significant relationship between alliances and conflict while
Peceny, Beer and Sanchez-Terry (2002) also find no statistically significant effect of
between alliances and conflict, others have taken the monadic approach, with conflicting
results. Ireland and Gartner (2001) examine how executive constraints in democratic
states, as determined by parliamentary government type, affect the hazard of that state
entering into a conflict. Controlling for the presence of a defense alliance, the authors
find that alliances decrease the probability of conflict initiation. Lai and Slater (2006),
predict that the more allies states have, the more opportunities for conflict. However,
their results show that no statistical relationship exists between the number of alliances a
state has and conflict. Finally, Siverson and Starr (1990) show a positive connection
between conflict and alliances as alliances create opportunities for conflict diffusion.
In sum, research at both the monadic and dyadic level has failed to clearly
6
different studies have asserted that alliances can lead to peace, have no effect on conflict,
or lead to war. We feel that these mixed empirical findings are a misspecification of the
relationship between alliances and conflict, as they do not consider the interactive effect
of alliance proximity. .
Re-specifying Alliances:
As noted above, Bremer concluded that the two best predictors of conflict
between states were proximity and alliances. We agree with Bremer’s initial assessment
that both allies and proximity are important predictors of conflict but suggest that a
missing link is their interaction. Rather it is not only whether you have an alliance but
Some of the more common reasons given to control for the presence of alliances
are that they reduce conflict; either by signaling intentions or by providing alternative
mechanisms for conflict resolution. Alternatively, others control for alliances because
they may increase conflict as more allies represent more conflict opportunities. The more
alliances a state is involved in the greater the potential for being drawn into a conflict.
We posit that alliances that are geographically proximate have the intended effect of
reducing conflict. Alliances with neighboring states are likely to reduce the conflict
opportunities that occur via geographic contiguity (e.g. Bremer 1992; Vasquez 1993).3
However, alliances with geographically distant states do not mitigate the influence of
geography. In fact the distance between them may embolden either the distant ally or
3
For an overview of the role of spatial proximity and conflict see Starr (2005).
7
opponents of the ally. Below we articulate a reason why we should see differential
We agree with Morrow (1991) that one of the motivations for states to engage in
alliances with other states is the trade-off of security for autonomy. Morrow argues states
seek alliances when they are already facing some threat and/or cannot internally balance
through the build-up of arms, therefore trading some policy autonomy for security or vice
versa. While alliances provide gains of trade for each party, those seeking greater security
versus those seeking greater autonomy, depending on the alliance there may be times in
which states that have engaged in this trade off do not see eye to eye over the same
issues. Under certain circumstances, the ties that bind in terms of security also create
conditions for disputes to emerge among allies Bueno De Mesquita and Lalman (1992).
Bueno de Mequita and Lalman contend that under many circumstance allies are
more likely to have disputes with one another than they are with non-allied states. They
posit that the stronger side in an alliance may have somewhat of a free hand to bully the
weaker member given the reduced expectation of third party intervention and the
provision of increased security. Conversely, the weaker party has an incentive to act
foster prisoner’s-dilemma conditions that encourage wars (167).” Thus, if states within an
alliance begin to disagree with one another the conditions exist which can lead to
conflict.
proximate allies are less likely to go to war than geographically distant allies. Bueno de
4
We recognize that this is only one a number of plausible arguments to explain the following results. Our
main goal here is to demonstrate the relationship so that future research employ better specified models.
8
Mesquita and Lalman (1992) suggest that the incentive for conflict can be removed if
each ally faces a substantial political cost for using force and that this information can be
conveyed to the other. We argue that geographically proximate allies are better able to
convey this information to one another than geographically distant allies. As Kimball
facilitates increased opportunities to interact not only negatively but positively as well
(Siverson and Starr 1991; Diehl 1992). Thus we suggest that proximate states that have
reached an alliance have a greater potential to share information with one another
reducing conflict between them. Similarly, geographically proximate neighbors are more
likely to create security communities further reducing conflict (Adler and Barnett 1996
Deutsch et al. 1957, Lake 1997). Thus given similar security concerns and the ability to
of a dispute occurring given the prisoner-dilemma like conditions created by the alliance,
the likelihood of disagreements over changing preferences, and the reduced ability to
Rather than creating a new data set to examine the connection between alliances
and geography, we replicate three relatively recent studies but include the interactive
term between allies and proximity. Specifically, we replicate three relatively well known
alliances and proximity (Oneal and Russett 2001; Peceny Beer and Sanchez-Terry 2002;
9
Lai and Slater 2006).5 In each of the dyadic models we follow the researchers’ coding
schemes. We then replicate the models using the same regression techniques but with an
interactive term between alliance and proximity. For the monadic study we create a
measure of the number of proximate alliances and include this with the already present
measure of total allies. We assess whether the interactive term improves the model by
examining changes in the individual coefficients as well as using goodness of fit tests to
gauge whether the interactive term improves the overall model fit.
Replication:
In one of the seminal pieces on the democratic peace proposition, Oneal and
Russett (2001) examine the relationship of both regime similarity and trade levels to the
onset of inter-state disputes. Included in their model are controls for the influence of
alliances and contiguity on dispute onset. Alliance is coded as one if the dyad shared a
defense pact, non-aggression pact or entente and zero otherwise. Instead of coding
whether states are contiguous they create a measure of non-contiguity. States are said to
be non-contiguous if they do not share a border and are separated by more than 150 miles
of water. Non-contiguous states are coded as one with contiguous states coded as zero.
They use Generalized Estimating Equation using an AR1 specification to control for
temporal dependence. Model 1 of table 1 replicates the results of their Model 5.1 (p316).
Both the presence of an alliance and being non-contiguous has statistically significant
5
We also replicated the other models from the articles above where data was available. Our main result
remained robust.
10
Model 2 represents the replication of their model interacting their measures of
contiguity and alliance commitment. While the contiguity measure is still negative and
significant the alliance measure is no longer significant. More importantly the interactive
term is positive and significant. The results suggest that alliances by themselves do not
inhibit conflict within the dyad. Rather, they suggest that states that have an alliance and
are not contiguous are more likely to engage in militarized disputes, while alliances with
contiguous states have no effect. Thus Oneal and Russett’s result, in terms of their
alliance measure, is due to the number of contiguous states with alliances. To test for
whether the interactive term provides a better fit of the data, we employ a Hosmer and
Lemeshow (HL) Goodness of Fit Test given that using the GEE model prevents us from
ratios. In general, a lower Chi Square value indicates a better fit between the observed
and predicted outcomes. The HL statistic for the interactive model is 57.38 with a p value
of .000. In comparison, the HL statistic for the Oneal and Russett model is 90.93 with a p
value of .000.6
Our second replication examines the “Dictatorial Peace.” Peceny, Beer and
Sanchez-Terry (2002) suggest that certain non-democratic dyads have a lower probability
democratic states. In their models they code the presence of an alliance within the dyad
as one, and no alliance as zero. Their contiguity measure identifies states that share a
6
The Hosmer and Lemeshow goodness of fit test is performed by dividing the predicted probabilities into
deciles (10 groups based on percentile ranks) and then computing a Pearson chi-square that compares the
predicted to the observed frequencies (in a 2 X 10 table). Lower values indicate a good fit to the data and,
therefore, good overall model fit.
11
land border or are separated by no more than 400 miles of water as contiguous. They
employ a logit specification with cubic splines to control for temporal dependence.
variable is whether the dyad experienced a militarized interstate dispute involving the use
of force or greater. The results show that contiguity is associated with an increased
probability of a dispute occurring within the dyad, while alliance has no effect. Model 2
of table 2 shows the interactive affect between alliances and contiguity. Both alliance and
contiguity are positive indicating that contiguous states without an alliance, and non-
militarized dispute. However, the combined effect is that contiguous allied states are less
likely to engage in a MID involving the use of force. What happens if we change the
criteria for states being contiguous? Models 3 and 4 present the same models as models 1
and 2 but with the more conventional 150 mile cut-off used to determine contiguity. The
magnitudes of the coefficients increase in both models. But is the model a better fit as
well? Comparing Bayesian Information Criterion statistic we find the difference between
them is 37.061, strongly suggesting that the interactive model is the correct model even
with the addition of another parameter.7 Finally, the proportional reduction in error
term.
7
The difference in BIC between Models 1 and 2 of Table 2 gave some positive support to the interactive
model.
12
Our third replication is based on the monadic study of Lai and Slater
(2006) who examine differences among autocratic regimes. Focusing on not only the size
of the authoritarian regimes winning coalition but the also on type of regime and support
provided, they find that military regimes are much more conflict prone that other types of
autocratic regimes. To control for the effects alliances on conflict they create a measure
of the total number of alliances a state has, using the common dichotomous coding
scheme. For contiguity they control for the number of borders that a state shares with
other states. They employ a negative bi-nomial regression with robust standard errors on
the state and a lagged dependent variable to account for heteroskedasticity and
autocorrelation, respectively.
Model 1 of table three replicates their results. While they find a positive
relationship between the number of bordering states and the number of initiated disputes
in a year, they find no statistical relationship between the total number of allies a state has
and the number of conflicts. Model 2 of table 2 includes a variable we created that counts
the number of geographically proximate allies. As the results show, the coefficient for
the number of borders is positive and significant but the measure of total allies is also
positive and significant, which is what Lai and Slater expected. More importantly our
Again the difference in BIC points strongly towards the model including the proximate
13
Conclusion:
improves statistical significance, model fit and most importantly fits with the existing
expectations about the conditions with which alliances will lead to peace or will
exacerbate conflict among states. The results are robust across a variety of statistical
methods as well as a variety of dependent variables. In all but one of the models, the
expected alliance-proximity relationship was found except for the effect of contiguous
alliances in replicating Oneal and Russett’s results. The results of our replications suggest
that the null findings produced in many studies in regards to alliances is not due to the
“key explanatory” measures somehow better accounting for the relationships between or
among states but rather due to mis-specification. Our results also suggest that
the models in table 2 demonstrate using the 150 mile cut off provided better estimates
than the 400 mile cut off initially employed. However, it is also worth noting that while
Our examination of the relationship between alliances and proximity fits into a
small but growing body of research that suggests contextual factors need to be thought
through more carefully when specifying models.8 As Ray (2003) notes “the construction
that the key explanatory factor, as well as the control variables, have essentially identical
impacts on interstate conflict across space, and over time. Substantial evidence …
suggests that this assumption is unwarranted (1).” Of course we do not suggest that
8
For example see Ward, Siverson and Cao (2007), Ray (2003), and Clark and Regan (2003)
14
researchers should always specify alliances, conditional on proximity. Specification
should be driven both by the questions being asked and the theoretical expectations
However, we do hope that this study will motivate others to avoid ritualistically including
the “Usual Suspects” as controls. Indeed, future research should seriously consider how
15
References
ADLER, E., and M Barnett. (1996) Governing Anarchy: A Research Agenda for the
Study of Security Communities. Ethics and International Affairs 10 (1): 63-98.
DIEHL, P. F. (1992) What Are They Fighting For? The Importance of Issues in
International Conflict Research Journal of Peace Research 29 (3): 333-344.
BUENO DE MESQUITA, B. (1981) The War Trap. New Haven : Yale University Press.
BUENO DE MESQUITA, B., and D. Lalman (1992) War and Reason. New Haven :
Yale University Press.
GIBLER, D. M. (2000) "Alliances: Why Some Cause War and Why Others Cause
Peace." In What Do We Know about War? edited by J. A. Vasquez.Lanham, MD :
Rowman & Littlefield.
GIBLER, D. M., and J. A. Vasquez (1998) Uncovering the Dangerous Alliances, 1495–
1980. International Studies Quarterly 42: 785–807.
IRELAND, M. J., and S.S. Gartner (2001) Time to Fight: Government Type and Conflict
Initiations in Parliamentary Systems. Journal of Conflict Resolution 45 (5): 547-568.
KIMBALL, A. L. (2006) Alliance Formation and Conflict Initiation: The Missing Link.
Journal of Peace Research 43 (4): 371-389.
LAI, B., and D. Slater (2006) Institutions of the Offensive: Domestic Sources of Dispute
Initiation in Authoritarian Regimes, 1950–1992. American Journal of Political Science
50 (1): 113-126.
16
LAKE, D. A. (1997) Regional Security Complexes: A Systems Approach.. In Regional
Orders. Building Security in a New World, edited by D. A. Lake and P. M. Morgan.
University Park, PA: The Pennsylvania State University Press. pp. 45-67
LEVY, J. S. (1981) Alliance formation and war behavior. Journal of Conflict Revolution
25: 581-613.
MORGENTHAU, H. (1967) Politics among nations; the struggle for power and peace.
New York: Alfred A Knopf.
ONEAL, J., and B. Russett. (2001) Clear and Clean: The Fixed Effects of the Liberal
Peace. International Organization 55 (2): 469-485.
OREN, I. (1990) The War Proneness of Alliances. Journal of Conflict Resolution 34 (2):
208-233.
RAKNERUD, A., and H. Hegre (1997) The Hazard of War: Reassessing the Evidence
for the Democratic Peace. Journal of Peace Research 34 (4): 385-404.
RAY, J. L. (2003) Explaining Interstate Conflict and War: What Should Be Controlled
For? Conflict Management and Peace Science 20: 1-31.
SIVERSON, R. M., and H. Starr (1990) Opportunity, Willingness and the Diffusion of
War. American Political Science Review 84 (1): 47-67.
SMITH, A. (1995) Alliances and War. International Studies Quarterly 39: 405-425.
SPRECHER, C., and V. KRAUSE (2006) Alliances, Armed Conflict and Cooperation:
Theoretical Approaches and Empirical Evidence. Journal of Peace Research 43 (4): 363-
369.
17
STARR, H. (2005) Territory, Proximity, and Spatiality: The Geography of International
Conflict The International Studies Review 7 (3) :387-406.
VASQUEZ, J. A. (1993) The War Puzzle Cambridge, New York: Cambridge University
Press.
18
Table 1. Replication of Triangulating Peace: Model 5.1
Dependent Variable MID Onset
Original Interaction
Model 1 Model 2
Allies -.539*** .011
(.159) (.322)
Non-Contiguity -.989*** -1.05***
(.168) (.174)
Allies X Non Contiguity .642*
(.354)
Lower Democracy -.061*** -.062
(.009) (.009)
Lower Dependence 52.92*** -54.41
(13.41) (13.75)
International Organizations -.013*** -.013
(.004) (.004)
Log Distance -.376*** -.374
(.064) (.064)
Only Minor powers -.647*** -.626
(.178) (.178)
Constant -.128 -.171
(.536) (.532)
N 39988 39988
Groups 1178 1178
Chi Square 288.11*** 244.48***
*p < .05, **p < .01,***p < .001
19
Table 2. Replication of the Dictatorial Peace: Table 4 Model 2
Dependent Variable Use of Force
20
Table 3. Replication of Institutions of the Offensive: Table 1 Model 1
Dependent Variable # of MIDS Initiated
Original Interaction
Model 1 Model 2
Total Allies .007 .021***
(.006) (.006)
Total Borders .070*** .113***
(.018) (.018)
Near Allies -.160***
(.039)
Single Party Regime .125 .205
(.207) (.188)
Military Regime .516* .602**
(.224) (.216)
Capability 4.68** 4.06***
(1.61) (1.22)
Openness -.947*** -.961***
(.283) (.283)
Lag DV .549*** -.514***
(.097) (.099)
Constant -2.17*** -2.34**
(.280) (.243)
N 4960 4960
Log Likelihood -2612.5 -2591.06
Chi Square 135.1*** 285.78***
Democracy is the excluded reference category for the regime type variables.
*p < .05, **p < .01,***p < .001 All tests are two-tailed. Robust Standard Errors (clustered on each
state) are in parentheses.
21