Top Banner
Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva [email protected] Florida State University Abstract Do alliances prevent preventive war? Preventive war motivations are an important cause of interstate war. Recent research finds that alliances deter less intense forms of militarized conflict. Research is unclear on the relationship between alliances and war, particularly preventive war. I argue that militarily powerful alliances deter states with preventive war motivations from initiating war. Powerful alliances deter war by making war more costly for the attacking state. Absent powerful allies, potential attackers are less likely to believe war will be costlier than the expected loss of influence due to an expected shift in military power. A multivariate model shows some support for the hypotheses in both the pre-Cold War and Cold War eras. Paper Prepared for the Political Institutions and Public Choice Conference, Durham NC May 2015. 1
27

Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva [email protected] Florida State

Aug 30, 2018

Download

Documents

dangkhuong
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

Political Institutions and Political Conflict: Alliances and Preventive War

Mark Souva

[email protected]

Florida State University

Abstract

Do alliances prevent preventive war? Preventive war motivations are an important cause of interstate war. Recent research finds that alliances deter less intense forms of militarized conflict. Research is unclear on the relationship between alliances and war, particularly preventive war. I argue that militarily powerful alliances deter states with preventive war motivations from initiating war. Powerful alliances deter war by making war more costly for the attacking state. Absent powerful allies, potential attackers are less likely to believe war will be costlier than the expected loss of influence due to an expected shift in military power. A multivariate model shows some support for the hypotheses in both the pre-Cold War and Cold War eras.

Paper Prepared for the Political Institutions and Public Choice Conference, Durham NC May 2015.

1

Page 2: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

Do alliances prevent preventive war? A significant body of research contends that

preventive war motivations are central to explaining the outbreak of many wars (see e.g. Fearon,

1995; Powell, 2006; Bell and Johnson, Forthcoming). Perhaps the most famous statement on this

topic is by AJP Taylor (1954: 166), who wrote: “every war between the Great Powers [between

1848 and 1919] started as a preventive war, not a war of conquest.” Another significant body of

research finds that alliances deter conflict (see e.g. Leeds, 2003; Johnson and Leeds, 2011;

Benson, 2011; Fuhrman and Sechser, 2014; Benson et al 2014). When combined, these two

research programs raise a question: Do alliances prevent preventive war?

On the one hand, defensive alliances should deter. If the alliance is credible, then it raises

the cost of war for the attacker. On the other hand, defensive alliances are formed because of

security concerns; thus, they may not deter but be an indicator that war is coming. In this context,

the defensive alliance is endogenous to a hostile situation. Vasquez (1993) also contends that an

alliance may not deter, but for a different reason. For Vasquez, an alliance is an example of

power politics and such actions increase the probability of conflict; they are a step to war.

I argue that most alliances fail to deter preventive war from occurring. They fail to deter

war because they are not sufficiently powerful. Alliances form when there is some expectation of

conflict. While almost any alliance may make war costlier for an attacker, to deter preventive

war it is necessary to significantly increase the cost of war so that a challenger no longer believes

it will prevail. A minor increase in the cost of war is not sufficient for deterrence to succeed. In

the context of preventive war, a would-be attacker has a military advantage and expects to win

should war occur. Alliances that simply make the outcome more difficult to achieve without

altering the expected outcome are unlikely to deter. Yet if the alliance is militarily powerful, then

2

Page 3: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

the alliance may significantly increase the cost of war and change the expected outcome of a

conflict, making preventive war less likely.

Alliances are often viewed as central to understanding the outbreak of major power war

(see e.g. Snyder, 1991), and the argument advanced and findings presented here help explain

variation in major power war in the 20th century. Why did the major powers go to war in 1914

and 1939/1941 but not after that? I argue that World Wars I and II were preventive wars; the

central bargaining problem leading to these wars was a credible commitment problem. Germany

expected to lose its military advantage in the near future. This made war in the present attractive.

In contrast, the major powers have avoided war with each other since the end of the Second

World War. I contend that stems from them not facing significant credible commitment problems

and that alliances are key to understanding why credible commitment problems have been less

common since 1945. In particular, the North Atlantic Treaty Organization alliance, the Southeast

Asian Treaty Organization alliance, and the Warsaw Pact alliance, among others, are central to

understanding the lack of major war since 1945. Each of these alliances features a very strong

ally, an ally with so much military power that credible commitment concerns are greatly reduced

and actors are able to find a peaceful bargain. In the conclusion, I discuss how the alliances prior

to World War I and World War II were not sufficiently powerful to prevent those preventive

wars.

The paper proceeds as follows. After situating this paper in the literature, I develop these

arguments connecting certain types of alliances to deterrence in more detail. Then I present a

research design to test the hypotheses, discuss the empirical results as well as limitations of the

current research and what to do next.

3

Page 4: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

Place in the Literature

Both the core research question—do alliances prevent preventive war—and the primary

hypothesis—very powerful alliances prevent war but less powerful ones do not—are novel.

Research examining the effect of alliances on interstate conflict does not distinguish between

credible commitment and informational mechanisms for bargaining failure. Moon and Souva

(2014) distinguish between credible commitment and informational reasons for bargaining

failure, but they do not examine the role of alliances and they only examine one reason for

credible commitment problems. Early research examining the effect of alliances on conflict did

not distinguish between different types of alliances or how militarily powerful the alliance was.

Leeds (2003), Johnson and Leeds (2011), Benson (2011), and Wright and Rider (2014) all

distinguish between different types of alliances, yet this body of work does not examine wars,

only militarized interstate disputes. Perhaps the study closest to this research is by Johnson,

Leeds, and Wu (Forthcoming). They measure the military capabilities in the alliance. This is

critical for distinguishing between powerful and weak alliances. Yet they do not examine wars,

which also means they do not focus on the preventive war context. Both of these are significant

for it appears that alliances have one effect on MIDS and a different one on wars. Moreover, the

bargaining context matters and their research does not take that into account.

Most recent research examining the effect of alliances on conflict measures conflict as

the occurrence, or initiation, of a militarized interstate dispute (MID) (Leeds, 2003; Johnson and

Leeds, 2011; Benson, 2011; Fuhrman and Sechser, 2014; Wright and Rider, 2014). This body of

research finds that some alliances (defensive or conditional deterrent) reduce the probability of a

MID. Research in the steps-to-war program, however, measures conflict as the occurrence of a

war, and this body of research tends to find that alliances are positively associated with war

4

Page 5: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

(Colaresi and Thompson, 2005; Senese and Vasquez, 2005). But this body of research does not

distinguish between types of alliances (e.g. offensive and defensive; conditional and

unconditional deterrent), the strength of the alliance, or whether the central mechanism leading

to war was informational or a credible commitment problem. These same limitations apply to

other relevant research programs. Some work on power transition theory takes into account

alliances, but this work does not distinguish between types of alliances (Kim, 1989, 1991).

(While power transition theory does not explicitly distinguish between information and credible

commitment problems, a charitable interpretation of that research program suggests that all

major wars are caused by credible commitment problems, thereby obviating the need for a

distinction between informational and credible commitment problems.) It is worth noting that

this research finds that alliances do not deter.

Finally, the small amount of empirical work on preventive war does not take into account

alliances (Lemke, 2003; Bell and Johnson, Forthcoming). However, this research does show that

preventive war expectations are statistically associated with war breaking out.

Theory

In the bargaining model of war, there are two general causes of bargaining failure leading

to war, information problems and credible commitment problems (Fearon, 1995; Wagner, 2007;

Powell, 2006). Since the publication of Fearon’s essay, most theoretical (see e.g. Fearon, 1994,

1997; Schultz, 1998, 1999; Smith, 1998; Slantchev, 2005, 2006) and especially most empirical

work has focused on information problems (see e.g. Schultz, 1999; Weeks, 2008; Uzonyi et al,

2012; Downes and Sechser, 2012). Yet increasingly research points to the importance of credible

commitment problems for explaining war (see e.g. Slantchev, 2012). A credible commitment

problem may occur for a variety of reasons: expectations of an immediate attack (i.e. pre-

5

Page 6: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

emptive war), a first-strike advantage, an expected long-term and costly defense, expectations of

a change in the other side’s political leadership, issue indivisibility, bargaining over goods that

affect future power, and expectations of a shift in power. I motivate the argument by focusing on

the preventive war context but the primary conclusions hold regardless of the reason for the

credible commitment problem.

Preventive war occurs when a declining state believes it is better off attacking a rising

state now than waiting for a worse deal, given an expected change in power, in the near future.

Fearon (1995) summarizes the basic preventive war calculus as follows:

δ(p2-p1) > Ca(1-δ) (Eq 1)

Here delta (δ) is the expectation of future interaction, p2 is State B’s (the rising state’s)

probability of winning in period 2, p1 is State B’s probability of winning in period 1, and Ca is

the cost of war for the declining state. If the left-hand side of the inequality is greater than the

right-hand side, then preventive war is likely. Typically the major insight gleaned from this

equation is that if the rising state’s probability of winning in period 2 is much greater than its

probability of winning in period 1, owing to an expected increase in military power, then

preventive war may occur. The central question of this research is the following: How do

alliances affect the preventive war calculus? To answer this question we need to distinguish

between types of alliances, alliances with the declining state and those with the rising state, and

the strength of the alliance.

There are a variety of types of alliances. Both the Correlates of War (COW) and the

Alliance Treaty Obligations Project (ATOP) distinguish between offensive, defensive, non-

aggression, and neutrality pacts. ATOP also includes consultation pacts and whether the alliance

has various conditions attached to it. Benson (2011) contends and finds that whether an alliance

6

Page 7: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

is conditional or unconditional affects the decision-making calculus of attackers and defenders.

In particular, only conditional defense pacts deter. I argue that only some defense pacts will deter

preventive war. There is little reason to believe that a non-aggression or neutrality pact is likely

to deter or embolden an attacker as they do not affect any of the parameters in the calculus.

Offensive alliances with the declining state increase the incentive for preventive war. If the

declining state with offensive allies does not attack today, then it is even more vulnerable in the

future when it is relatively weaker and the alliance does not bind the allies to aid in defense.

Offensive alliances with a rising state are also unlikely to deter. Knowing that a rising state has

an offensive alliance makes a declining state value the present even more than the future. In the

present, the offensive allies are not bound to aid the rising state if it is attacked; thus, the alliance

has no immediate deterrence value. In the future, the rising state’s bargaining advantage is extra

strong. It will be stronger and with offensive allies it can credibly threaten to punish its opponent.

In brief, offensive alliances are not likely to reduce the probability of preventive war, though

such alliances with a declining state may further increase the likelihood of war.

Defensive alliances with a declining state are unlikely to reduce the probability of war.

On the one hand, it might seem that a declining state with defensive alliances has significantly

less incentive to engage in preventive war than it otherwise would. A declining state is concerned

about the future, but if it has strong allies to aid it in the event of an attack, then it should be less

concerned. While this effect should be present, two other considerations are relevant. First, if a

declining state’s allies are very strong, then it is likely that the declining state itself is not a major

power as most alliances tend to be militarily asymmetric. Countries that are relatively weak are

not likely to be declining in power with respect to potential foes; they are likely to already be

weaker. One may fear becoming even weaker, but there should be a small number of cases that

7

Page 8: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

meet this condition. Second, if a declining state is strong, relative to its allies, then it not only

fears the adverse shift in military power with its foe but also a decrease in autonomy benefits

from its alliance. Relatively strong states receive some, if not more, autonomy benefits than

security benefits from their alliances (Morrow, 1991). Thus, a declining state with defensive

allies may have less reason to fear a future attack than a state without such allies but the future is

costly to it both in terms of its decline vis-à-vis the rising state and with respect to its own allies.

Since war is costly, these costs from relative decline may well be less making preventive war

unlikely. Yet given what we know about the endowment effect (Tversky and Kahneman, 1981) it

seems as likely that leaders will believe the costs of decline are greater than the cost of war,

making preventive war more likely.

Some defensive alliances with a rising state do reduce the incentive for war. The

motivation for preventive war is usually thought of only in terms of the expected shift in power.

With this focus in mind, whether the rising state has defensive allies or not is irrelevant. But the

preventive war calculus also depends on the declining state’s expected cost of war today. This is

where the rising state’s defense pacts alter the probability of preventive war. The cost of war for

a declining state increases when a rising state has strong defensive allies. A declining state would

rather fight a rising state with no allies or weak allies than one with very strong allies. Strong

allies, for the rising state, decrease the incentive for preventive war.

Under what conditions is one actor likely to believe another’s defensive alliances will

increase its cost of war? The allies have to be strong. A rising state with weak allies may not be

able to deter a declining state for they will not alter the cost of war enough. Given the tendency

of states to bandwagon with the system leader (see e.g. Organski and Kugler, 1981), it may be

that many rising states are not able to attract strong allies to prevent preventive war. Stated

8

Page 9: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

differently, but with the same implication, a rising state may have difficulty attracting allies

when it is commonly known that war with a strong country is a serious consideration. As

discussed in more detail below, this helps account for the outbreak of World Wars I and II.

The above arguments lead to the following hypothesis.

Hypothesis 1: When preventive war motivations are high, the probability of preventive war decreases with the military strength of the rising state’s allies.

Research Design

Central to this research is the distinction between potential attackers and potential

defenders. The independent variable of interest, powerful defensive alliances, is only expected to

reduce the probability of war for rising states. With this in mind, the unit of analysis for this

research is the directed-dyad year.

Empirical research on international conflict often analyzes the initiation of a militarized

interstate dispute (MID). MIDs are threats, shows, displays, or uses of force. The vast majority of

MIDs are not wars, as conventionally defined (1000 battle deaths). A primary benefit, then, of

analyzing MIDs is that there are more to analyze. Yet our theories typically focus on war as the

outcome variable of interest. This is the case with the bargaining model of war. Accordingly, this

research examines wars as the outcome variable and uses data from the Correlates of War project

(Sarkees and Wayman, 2010). Most empirical models of war include joiners, for both the

initiating and target sides. Since the focus here is on alliances, this is especially appropriate. For

example, if State A is contemplating war with State B and State B has a defense pact, then State

A knows that it is likely going to war against both State B and its allies.1

1 In sensitivity analyses, I examine only original participants and joiners for the target side but not the initiating side.

9

Page 10: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

One objection to a focus on wars is that they are the outcome of a process and by looking

at the end of the process one may not properly assess the effect of an independent variable on

that process. The balance of power, for example, may have one effect on the occurrence of a

conflict but another effect on the escalation of that conflict to war (Fearon, 1994). However, a

multistage research design requires a clear theory about the effect of an independent variable on

different stages of conflict. In this case, the theory does not make predictions of that sort. If

anything, the predictions are about the outcome of war, not MIDs. In addition, empirical research

modeling the process leading to war still arbitrarily chooses stages. Why should the outbreak of a

MID be the first stage modeled? Why not the outbreak of an issue claim or diplomatic dispute?

This body of research is also not clear about precisely how to model the process. Should ongoing

disputes be included or not? Together the lack of a clear theory for focusing on MIDs instead of

wars and the empirical concerns with how to empirically model the selection process suggest that

using wars as the outcome measure is appropriate.2 Finally, it is worth noting that research

examining preventive war has employed a research design similar to the one here (Lemke, 2003;

Bell and Johnson, forthcoming).

The primary independent variable of interest is capable defensive alliances. A capable

defensive alliance is defined as a defensive alliance that is militarily strong. This requires

identifying all defensive alliances that State B, the potential target, has with obligations to aid B

in the event of an attack by State A. In other words, many alliances identify specific foes and

allies are not bound to enter a conflict if the attacker is someone different than identified in the

alliance. Practically speaking, this means State B may have some allies that are not relevant for

2 Braithwaite and Lemke (2011) provide an excellent discussion of the sensitivity of Heckman selection models in the context of international conflict.

10

Page 11: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

understanding a potential conflict between A and B. Johnson and Leeds (2014) and Johnson,

Leeds, and Wu (Forthcoming) have identified all relevant alliances for each directed-dyad.

Further, Johnson, Leeds, and Wu aggregate the military power of the allies where military power

is defined as the Correlates of War Composite Capabilities Index. I use their data for the Alliance

Power variable. Alliance power is the ratio of State B’s power plus its relevant allies’ power to

the power of States A, B, and all relevant defensive allies for B. The variable ranges from zero to

one with higher values indicating that State B and its allies have a greater probability of

prevailing in a conflict. Now it is not straightforward exactly how much power the target and

allies need to prevent preventive war. Equal power or a slight power advantage for the target

may not be sufficient. Based on the theory, however, we would expect to observe support for

Hypothesis 1 when the target and its allies are significantly stronger than the challenger. In other

words, it is not entirely clear if the relationship between alliance power and war is linear or holds

only at high values of alliance power. To guard against non-linearity in the data and examine the

relationship for the most relevant set of cases, I employ a binary variable. Specifically, Target’s

Allies’ Power is equal to one if the target and allies’ power is five times or more greater than the

challenger’s power.3

In addition to the presence of militarily powerful defensive alliances, other factors affect

the probability that one state will initiate a war against another state. I control for the following

potential confounders: the expected reliability of the alliance, the regime type of each state and

regime similarity, foreign policy affinity, geographical proximity, and conflict history.4

3 The findings in the pre-1942 period hold when the target and its allies’ power is three times as great as the challenger. This is not the case in the post-World War II era; five times the challenger’s power seems to be the cutpoint. 4 Unless otherwise noted, data for the control variables was generated from EUGene (Bennett and Stam, 2000) v3.204.

11

Page 12: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

The reliability of an alliance may be central to explaining whether the alliance has an

effect on the outcome. It may be the case, for example, that alliances expected to be unreliable

will not deter but alliances expected to be reliable will deter. Johnson, Leeds, and Wu

(Forthcoming) suggest an alternative relationship between reliability and deterrence.5 Credibility

is more important for deterrence when the alliance is relatively weak. Capability and credibility

are not complements but substitutes. For the primary analyses, I do not interact Target’s Allies’

Power and Alliance Reliability, but this relationship is explored in sensitivity analyses. Alliance

Reliability is measured as the similarity in foreign policy portfolios between State B and its

allies. Data come from Johnson, Leeds, and Wu.

To measure each country’s political institutions and the relationship between each state’s

political institutions, I draw on work by Bennett (2006). In recent years, a body of research has

established that there is (a) a democratic peace, (b) a weaker but still relevant autocratic peace,

and (c) a democratic initiator peace. Bennett refers to the first two as the joint coherence peace,

that is, two countries that are institutionally coherent and similar (either a positive or a negative

ten on the Polity democracy index) share a separate peace. To account for all of these known

empirical regularities, I include the following variables: State A Democracy, State B Democracy,

Joint Democracy, Joint Democracy Squared, and State A Democracy*Political Similarity. All

measures are based on the rules described in Bennett (2006). Data come from the Polity IV

project (Marshall and Jaggers, 2002).

Affinity is a measure of the alliance portfolio similarity between the two states in the

dyad, weighted by the Correlates of War capability index. I use the S version of this variable

5 In sensitivity analysis, I substitute military institutionalization for portfolio similarity as the measure of expected reliability. The more institutionalized the alliance, the greater the expected reliability. This measure is based on Leeds and Anac (2005) and is also used by Johnson, Leeds, and Wu (Forthcoming).

12

Page 13: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

(Signorino and Ritter, 1999). Higher values indicate more similar foreign policies. I include two

measures of geographic proximity, contiguity and distance between capital cities. Contiguity

equals one if the states are contiguous by land or separated by less than 400 miles of war, zero

otherwise. Distance is the natural log of the great circle distance between capital cities. Finally, I

include three variables to address the conflict history of the dyad and temporal dependencies in

the data. The variable Peace Years is a count of the number of years since the last COW war.

Peace Years Squared and Cubed are straightforward (see Carter and Signorino, 2010).

The empirical domain of this study spans all directed-dyads in which there is (a) a high

level of preventive war motivations and (b) a relevant defense pact for the period 1816-2000.6

The focus on dyads with a preventive war motivation is unique. The primary reason for limiting

the analysis to this set of dyads is theoretical. It is not clear, for example, that a potential initiator

with defensive allies will be more likely to initiate a war when preventive war motivations are

low. In this context, instead of credible commitment problems being central to the bargaining

dynamic, information problems take center stage. Deterrence is still relevant to understanding

bargaining failure as a result of an information problem, but generally such failures stem from

asymmetric information on resolve. Thus, the expected reliability of the alliance probably

matters more when preventive war motivations are not high. In addition to these theoretical

considerations, an empirical focus on preventive war contexts is important. As noted earlier,

credible commitment problems, and preventive war in particular, seem to be very important

causes of war, yet they have received little empirical attention. If nothing else, this is an

interesting subset of cases.

6 In sensitivity analyses, the models are run only on politically relevant dyads. Results are similar.

13

Page 14: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

When are preventive war motivations high? When should State A expect an adverse

relative power shift with respect to State B in the near future? Lemke (2003) examines changes

in relative power over the last twenty years. The more one state is declining in power relative to

another the greater the preventive war motive. Lemke’s analysis shows little empirical support

for preventive war motivations being positively related to war. Bell and Johnson (Forthcoming)

note that preventive war motivations are based on expectations about the future. Changes in

power in recent years should already be incorporated into the bargaining status quo and therefore

are unlikely to trigger a preventive war. Accordingly, they calculate each country’s future power

(one, three, and five years into the future) and then the expected change in relative power in the

dyad. Bell and Johnson find robust support for their measure of expected shifts in power and the

outbreak of war. Notwithstanding the merits of the Bell and Johnson approach, their measure

depends heavily on a sound model of military spending. With a variety of models of military

spending available, one would want to examine the sensitivity of their results to different models.

Also, they do not predict military spending per se, but a state’s share of military power (the two

military components of the COW capabilities index). It is not clear that models of military

spending are appropriate for predicting a country’s share of these components. Finally, it is also

unclear that such a measure, or thought process, is used by decision-makers. It seems likely that

decision-makers use a simpler rule of thumb. I turn to one such possibility.

Preventive war motivations are a function of relative militarization levels, where a

militarization advantage indicates a greater motivation for preventive war. I use the COW

capability data to measure militarization. A country’s militarization is the difference between its

military share and the average of its economic and demographic share. A high value on this

variable means a country is more militarized than its basic economic and demographic attributes

14

Page 15: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

would suggest. I define preventive war motivations as the difference between each country’s

militarization score, where higher values indicate a greater motivation for preventive war. The

logic works as follows. It is relatively easy for State A to know if it is more militarized than State

B. If State A is more militarized than it has reason to believe that State B will increase its level of

militarization, for State A has militarized more based on the same inputs. If State B increases its

level of militarization, then State A either has to increase its level, and it is already more

militarized, or suffer a decrease. To avoid the extra expenditure, State A may choose preventive

war.7 Next, I classify a dyad as having a high preventive war motivation if the relative

militarization measure is in the top quartile.

Empirical Analysis

First, I examine whether the preventive war measure is correlated with war. If there is

little empirical support for the preventive war measure, then splitting the sample for the

subsequent analyses is less persuasive. I do, however, find a statistically significant and positive

relationship between preventive war motivations and the initiation of war (see Table 2 in

Appendix A). About 38% of dyads in the pre-1942 period have high preventive war motivations,

but they account for about 60% of all war initiations. After World War II, dyads with high

preventive war motivations account for about 25% of the sample and 44% of war initiations.

Table 2 shows coefficient estimates for a multivariate model of war initiation. The addition of

likely confounders does not change the statistical relationship between preventive war

expectations and war initiation. Given the known lack of major power war in the latter era and

7 In sensitivity analyses, I explore a variety of variations on this basic measure. For example, do preventive war motivations hold for all dyads or only when one state is militarily stronger? Do preventive war motivations hold for all dyads or only dyads in which there is some expectation of hostility or future conflict? Does militarization depend on both economic and demographic attributes? Should these attributes be weighted equally? I also examine different cutoffs for the ‘high preventive war motivation.’ None of the main findings change.

15

Page 16: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

the existence of nuclear weapons, there is some reason to believe that preventive war motivations

may only be relevant for earlier historical periods. The analysis shows that preventive war is still

a relevant concern today. Substantively, a dyad with preventive war motivations is about 88%

more likely to experience war than a dyad without such motivations prior to the Cold War and

about 110% more likely to experience war after 1945.

The main empirical results are presented in Table 1. To guard against temporal treatment

effects (Clark and Nordstrom, 2003), particularly whether the post-World War II era is different

than the earlier period, I estimated separate models on the 1816-1945 and 1946-1991 time

periods. For Hypothesis 1, the key parameter estimate of interest is on the term, Target’s Allies’

Power. It is negative and statistically significant, as expected by the hypothesis. Although the

major focus here is on the statistical significance of the parameter, it is worth discussing the

substantive effect. The variable Target’s Allies’ Power reduces the probability of war by about

71%.8 By comparison changing the domestic political institutions in the two states from mixed

autocratic-democratic to both fully democratic reduces the probability of war by about 96%. In

the 1816-1941 period, about 56% of the cases, potential targets, have an ally that makes the

target and allies’ power at least five times the power of the potential initiator, yet only 25% of all

war initiations in the sample occur in this context.

The results are similar for the post-World War II era (see Model 2 in Table 1). Target’s

Allies’ Power is again statistically significant and negative. In this sample, when the potential

target and its allies are five times stronger than the challenger, the probability of war decreases

by about 62%. Moreover, an examination of the number of cases with a strong ally (or allies)

8 Changes in predicted probability were calculated using Clarify (King, Tomz, and Wittenberg, 2000). All variables were set at their mean or median, except for the variable of interest which was changed from zero to one.

16

Page 17: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

suggests why the post-World War II period has seen a decrease in interstate war. About 83% of

all potential targets have an alliance such that the power of the target and its allies is at least five

times as great as the potential initiator’s power. Since more countries now have powerful allies

(e.g. the NATO, SEATO, Warsaw Pact, etc), it is perhaps not surprising that war has decreased.

The ability of powerful alliances to reduce the probability of war is all the more

noteworthy considering the context. The context is dyads with high preventive war motivations.

Powerful alliances, then, are reducing the probability of war among dyads that are relatively

more likely to go to war than other dyads.

Further Analysis

Further analysis of the post-World War II era suggests that the central factor of interest is

not Target’s Allies’ Power per se but alliances with nuclear weapons. I re-ran Model 2 replacing

Target’s Allies’ Power with a binary variable equal to one if at least one of the alliance partners

possessed nuclear weapons. This variable, Nuclear Capable Ally, is statistically significant and

negative. When a state has an ally with nuclear weapons, preventive war is unlikely to occur.

Moreover, the model with the Nuclear Capable Ally model outperforms (i.e. has a lower log-

likelihood, AIC and BIC) the model with Target’s Allies’ Power. If both variables are included,

then Target’s Allies’ Power variable becomes statistically insignificant while Nuclear Capable

Ally is still significant and negative. Nevertheless, the two variables agree in about ninety

percent of cases.

The sample for Models 1 and 2 only includes directed-dyads for which there is a relevant

defensive alliance. With respect to the independent variable, then, we are comparing alliances

with a lot of power versus alliances with only some power. In all cases, there is an alliance. This

17

Page 18: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

is a difficult test for the hypothesis for anytime an alliance is present there is some expectation

that it will contribute to deterrence. Moreover, because alliances form when security concerns

are present, one might expect that military conflict is more likely in this sample. This selection

process makes it less likely to find support for the hypothesis advanced in this research.

Nevertheless, some countries may fail to form an alliance because conflict is probable and

potential allies do not want to go to war. Do strong allies reduce the probability of war when we

examine all dyads with preventive war motivations? Table 3 shows model results that address

this question. For both the pre-1942 and post 1945 periods, the data is consistent with the

hypothesis: a strong alliance reduces the probability of preventive war.

An examination of the countries with powerful alliances that deter others with preventive

war motivations sheds light on the multivariate analysis. In the post-World War II period, some

of the targets meeting these criteria that are not attacked, despite being under-militarized, are

Japan, Turkey for many years but not all, and a number of Western and Eastern European

countries. Japan, Turkey, and the Western European countries were allied with the United States;

the Eastern European countries that were under-militarized were allied with the Soviet Union.

Powerful allies, particularly nuclear capable ones, appear to enhance deterrence. This carries an

important policy implication. To reduce the probability of war, countries do not necessarily need

to develop nuclear weapons. All they need is an ally with nuclear weapons.

Based on the measure presented here, Germany had preventive war motivations in 1914

and 1939. Further, the alliances arrayed against Germany were not sufficiently strong to deter

Germany from attacking. The combined CINC scores for the Germany & Austria-Hungary

alliance is .226, while the combined CINC scores for the Triple Alliance of France, Russia, and

the United Kingdom is .321. The latter alliance was only a little stronger based on these

18

Page 19: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

numbers, and based on the empirical results presented here not nearly strong enough to deter

Germany’s preventive war motivations. If the United States would have allied with Great Britain

and the others, then it is possible that World War I could have been prevented as the alliance

would have been more than twice as strong as the Entente. Instead of blaming alliance politics

for the expansion of World War I, it may be more accurate to blame politics for preventing

certain alliances that could have prevented that preventive war. Alternatively, if these countries

would have armed more in peacetime, then these wars may have been avoided.

In the post-World War II period, wars that could potentially have been deterred include

the Korean War, the Iraq-Iran war in the 1980s, and the Iraq attack on Kuwait in 1990. The Iraq-

Iran War is sometimes described as a war of opportunity. Saddam Hussein saw that the

revolution in Iran weakened the country temporarily so he attack while conditions were ripe.

What this research adds is that the primary factor motivating Saddam’s thinking was not what

happened inside Iran but the loss of American support for Iran. The Iraq-Kuwait war of 1990

may have been easier to prevent. Ambassador Glaspie’s words would have been less ambiguous

if the U.S. had a defense pact with Kuwait. A similar story describes the Korean War. Secretary

of State Acheson inadvertently signaled to North Korea that South Korea was not critical to the

U.S. geopolitical strategy. One implication of the Kuwait and Korean wars is that institutions are

better at deterring than diplomacy.

Conclusion

This research examines whether alliances can prevent preventive war. Preventive war, or

war motivated by expectations of a change in military power, are a statistically significant

predictor of war initiation. At the same time, a large body of research indicates that alliances, at

least certain types of alliances, deter others from attacking one’s allies. I argue that militarily

19

Page 20: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

powerful alliances will deter preventive war. I find support for the primary hypothesis of this

research in multiple time periods and samples. In the pre-Cold War era, powerful alliances

decrease the likelihood of a war. Further, it appears that one of the reasons for the outbreak of

World Wars I and II is that the alliances arrayed against Germany were not sufficiently strong.

This research does not address whether the alliances contributed to the expansion of the war, but

it does suggest that the alliances in place in 1914 and 1939/1941 were not strong enough to deter

Germany from launching a preventive war. Since the end of the Second World War, there has

been an increase in powerful alliances and, I argue, this has contributed to the relative

peacefulness in this era. Powerful alliances since 1945 are associated with countries possessing

nuclear weapons, but as the pre-1942 analysis makes clear, nuclear weapons are not the only way

to make preventive war too costly to undertake.

There is still much research to do on this topic. Further work is needed on the theory. Is

an examination of MIDs and escalation to war the appropriate way to assess the deterrent effect

of alliances? Should one expect an interactive relationship between alliance power and alliance

credibility? Are these factors complements or substitutes? Is the theoretical motivation for

distinguishing between dyads with a high preventive war motivation and those with a low

preventive war motivation a sufficient reason for separating sample? Would different results

obtain if the measure of preventive war was different? These are only a few of the questions to

investigate in further work. Nevertheless, the current research suggests two interesting aspects of

the causes of war. Preventive war motivations are often relevant to explaining war initiation, and

political institutions are sometimes a force for peace. A country with very strong allies is much

less likely to be the target of preventive war than a country lacking powerful allies.

20

Page 21: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

Bibliography

Bell, S. R. and J.C. Johnson. Forthcoming. Shifting Power, Commitment Problems, and Preventive War. International Studies Quarterly. Bennett, D. Scott, and Allan C. Stam. 2000. EUGene: A conceptual manual. International Interactions 26.2: 179-204. Bennett, D. Scott. 2006. "Toward a continuous specification of the democracy–autocracy connection." International Studies Quarterly 50 (2): 313-338. Benson, B. V. (2011). Unpacking alliances: Deterrent and compellent alliances and their relationship with conflict, 1816–2000. The Journal of Politics, 73(04), 1111-1127. Benson, Brett V., Adam Meirowitz, and Kristopher W. Ramsay. "Inducing Deterrence through Moral Hazard in Alliance Contracts." Journal of Conflict Resolution 58.2 (2014): 307-335. Braithwaite, Alex, and Douglas Lemke. 2011. Unpacking Escalation. Conflict Management and Peace Science 28 (2): 111-123. Carter, David B., and Curtis S. Signorino. 2010. Back to the future: Modeling time dependencein binary data. Political Analysis 18 (3): 271-292. Clark, David and Timothy Nordstrom. 2003. Risky inference: Unobserved treatment effects in conflict studies. International Studies Quarterly 47 (3): 417-429. Colaresi, Michael and William Thompson. 2005. Alliances, Arms Buildups and Recurrent Conflict: Testing a Steps-to-War Model. Journal of Politics 67: 345-364. Downes, A. B., & Sechser, T. S. (2012). The illusion of democratic credibility.International Organization, 66(03), 457-489. Fang, Songying, Jesse C. Johnson, and Brett Ashley Leeds. 2014. To Concede or to Resist? The Restraining Effect of Military Alliances." International Organization 68 (4): 775-809. Fearon, James D. 1994. Domestic Political Audiences and the Escalation of International Disputes. American Political Science Review 88: 577-592. Fearon, James D. 1995. Rationalist explanations for war. International Organization 49(03): 379–414. Fearon, J. D. (1997). Signaling foreign policy interests tying hands versus sinking costs. Journal of Conflict Resolution, 41(1), 68-90.

21

Page 22: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

Fuhrmann, M., & Sechser, T. S. (2014). Signaling Alliance Commitments: Hand-Tying and Sunk Costs in Extended Nuclear Deterrence. American Journal of Political Science, 58(4), 919-935. Johnson, J. C., & Leeds, B. A. (2011). Defense Pacts: A Prescription for Peace? 1. Foreign Policy Analysis, 7(1), 45-65. Johnson, J.C., B. A. Leeds, and A. Wu. Forthcoming. Capability, Credibility, and Extended General Deterrence. International Interactions. Kim, Woosang. "Alliance transitions and great power war." American Journal of Political Science (1991): 833-850. Kim, Woosang. "Power, alliance, and major wars, 1816-1975." Journal of Conflict Resolution 33.2 (1989): 255-273. Leeds, B. A. (2003). Do alliances deter aggression? The influence of military alliances on the initiation of militarized interstate disputes. American Journal of Political Science, 47(3), 427-439. Leeds, B.A., & Anac, S. (2005). Alliance institutionalization and alliance performance. International Interactions, 31(3), 183-202. Leeds, B., Ritter, J., Mitchell, S., & Long, A. (2002). Alliance Treaty Obligations and Provisions, 1815-1944. International Interactions, 28(3), 237-260. Lemke, D. (2003). Investigating the preventive motive for war. International Interactions, 29(4), 273-292. Marshall, Monty and Keith Jaggers. 2002. Polity IV Project: Political Regime Characteristics and Transitions, 1800-1999. http://www.bsos.umd.edu/cidcm/polity/ Moon, C., & Souva, M. (2014). Audience Costs, Information, and Credible Commitment Problems. Journal of Conflict Resolution. Morrow, J. D. (1991). Alliances and asymmetry: An alternative to the capability aggregation model of alliances. American Journal of Political Science, 904-933. Organski, A. F. and Jacek Kugler. (1981). The war ledger. University of Chicago Press. Powell, Robert. 2006. War as a Commitment Problem. International Organization 60 (Winter): 169-203. Sarkees, Meredith Reid and Frank Wayman (2010). Resort to War: 1816 - 2007. CQ Press.

22

Page 23: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

Schultz, Kenneth. 1998. Domestic Opposition and Signaling in International Crises. American Political Science Review. 92 (4): 829-844. Schultz, K. A. (1999). Do democratic institutions constrain or inform? Contrasting two institutional perspectives on democracy and war. International Organization, 53(02), 233-266. Senese, Paul D. and John A. Vasquez. 2005. Assessing the steps to war. British Journal of Political Science 35 (October): 607-633. Signorino, Curtis S., and Jeffrey M. Ritter. 1999. Tau-b or Not Tau-b: Measuring the Similarity of Foreign Policy Positions. International Studies Quarterly 43.1: 115-144. Singer, J. David, Stuart Bremer, and John Stuckey. 1972. “Capability Distribution, Uncertainty, and Major Power War, 1820-1965.” in Bruce Russett (ed) Peace, War, and Numbers, Beverly Hills: Sage, 19-48. Slantchev, B. L. (2006). Politicians, the media, and domestic audience costs.International Studies Quarterly, 50(2), 445-477. Slantchev, B. L. (2005). Military coercion in interstate crises. American Political Science Review, 99(04), 533-547. Slantchev, B. L. (2012). Borrowed power: debt finance and the resort to arms.American Political Science Review, 106(04), 787-809. Smith, A. (1998). International crises and domestic politics. American Political Science Review, 92(03), 623-638. Taylor, A. J. P. (1954). The struggle for mastery in Europe, 1848-1918. Oxford University Press. King, Gary, Michael Tomz, and Jason Wittenberg. 2000. Making the most of statistical analyses: Improving interpretation and presentation. American journal of political science (2000): 347-361. Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211(4481), 453-458. Uzonyi, G., Souva, M., & Golder, S. N. (2012). Domestic Institutions and Credible Signals. International Studies Quarterly, 56(4), 765-776. Vasquez, John. A. 1993. The war puzzle. Cambridge University Press. Wagner, R. Harrison. 2007. War and the State. Ann Arbor: University of Michigan Press. Weeks, J. L. (2008). Autocratic audience costs: Regime type and signaling resolve. International Organization, 62(01), 35-64.

23

Page 24: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

Wright, T. M., & Rider, T. J. (2014). Disputed territory, defensive alliances and conflict initiation. Conflict Management and Peace Science, 31(2), 119-144.

24

Page 25: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

TABLE 1 MULTIVARIATE ANALYSIS OF ALLIANCE POWER AND WAR INITIATION IN DIRECTED-

DYADS WITH A RELEVANT DEFENSE PACT

Model 1: 1816-1941 Model 2: 1946-2000 b/se b/se

Target’s Allies’ Power -1.364** -1.224** (0.515) (0.677) Alliance Reliability -1.332 -1.319 (1.172) (1.051) Initiator’s Alliance Power 6.372*** -1.943 (1.455) (2.316) Democracy A -0.217* 0.014 (0.093) (0.080) Democracy B -0.014 -0.013 (0.063) (0.052) Joint Democracy -0.024* -0.010 (0.010) (0.016) Joint Democracy2 -0.000 0.000 (0.000) (0.000) Democracy A x Similarity 0.013 -0.012 (0.007) (0.008) F.P. Affinity -0.161 -1.420 (0.630) (1.310) Ln (Distance) -0.133 -0.925*** (0.349) (0.209) Contiguity 0.847 0.018 (0.677) (1.077) Peace Years 0.098 -0.137* (0.074) (0.055) Peace Years2 -0.002 0.003** (0.001) (0.001) Peace Years3 0.000 -0.000** (0.000) (0.000) Constant -6.361* 2.134 (2.648) (2.656)

N 6270 135,034 BIC 387 465 * p<0.05, ** p<0.01, *** p<0.001, one-tail significance test. Standard errors clustered on the directed-dyad.

25

Page 26: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

Appendix A: Sensitivity Analysis

TABLE 2

MODELS WITH PREVENTIVE WAR MOTIVATIONS AS AN INDEPENDENT VARIABLE, ALL DIRECTED DYADS

Model 1: 1816-1941 Model 2: 1946-2000 b/se b/se

Preventive Motivations 0.615** 0.698** (0.192) (0.268) Democracy A -0.022 -0.000 (0.034) (0.052) Democracy B 0.027 0.004 (0.023) (0.039) Joint Democracy -0.006 -0.006 (0.004) (0.006) Joint Democracy2 -0.000 -0.000 (0.000) (0.000) Democracy A x Similarity 0.000 -0.007 (0.003) (0.004) F.P. Affinity -1.829*** -0.548 (0.257) (0.404) Contiguity 2.141*** 2.019*** (0.345) (0.396) Ln (Distance) 0.060 -0.727*** (0.152) (0.123) Peace Years 0.050* -0.102*** (0.022) (0.020) Peace Years2 -0.001** 0.002*** (0.000) (0.000) Peace Years3 0.000** -0.000*** (0.000) (0.000) Constant -7.535*** -2.166 (1.254) (1.184)

N 186,231 797,560 BIC 2403 1650 * p<0.05, ** p<0.01, *** p<0.001, two-tail significance test. Standard errors clustered on the directed-dyad.

26

Page 27: Political Institutions and Political Conflict: Alliances ... · Political Institutions and Political Conflict: Alliances and Preventive War Mark Souva msouva@fsu.edu Florida State

TABLE 3 MULTIVARIATE ANALYSIS OF ALLIANCE POWER AND WAR INITIATION IN DIRECTED-

DYADS (ALL DYADS WITH HIGH PREVENTIVE WAR MOTIVATIONS)

Model 1: 1816-1941 Model 2: 1946-2000 Target’s Allies’ Power -0.683 -0.922 (0.329)** (0.322)** Alliance Reliability 2.288 -0.442 (0.968)* (0.871) Initiator’s Alliance Power 6.241 1.909 (1.120)** (1.285) Democracy A -0.012 0.031 (0.047) (0.089) Democracy B 0.039 0.003 (0.030) (0.071) Joint Democracy -0.005 -0.008 (0.005) (0.011) Joint Democracy2 -0.000 -0.000 (0.000) (0.000) Democracy A x Similarity 0.000 -0.011 (0.004) (0.007) Ln (Distance) 0.168 -0.789 (0.168) (0.170)** Contiguity 2.114 1.326 (0.389)** (0.596)* Peace Years 0.116 -0.109 (0.032)** (0.028)** Peace Years2 -0.003 0.002 (0.001)** (0.001)** Peace Years3 0.000 -0.000 (0.000)** (0.000)** Constant -11.523 0.026 (1.775)** (1.946)

Log Likelihood -634.67 -309.80 N 66,815 201,082 * p<0.05, ** p<0.01, *** p<0.001, one-tail significance test. Standard errors clustered on the directed-dyad.

27