ArkPSA

Arkansas Political Science Association

───────────────────────────────────────────────
Offensive Realism, Defensive Realism, and the Role of Constraints
Author(s): Peter Rudloff
Source: The Midsouth Political Science Review, Volume 14 (December 2013), pp.
45-77
ISSN: 2330-6882 [print]; 2330-6890 [online]
Published by: Arkansas Political Science Association
Website: http://uca.edu/politicalscience/midsouth-political-science-
review-mpsr/
───────────────────────────────────────────────
 Offensive Realism, Defensive Realism, and the Role of Constraints1

Peter Rudloff
Oklahoma State University

Despite the prominence of realist theory in international relations, there

is much disagreement among realists as to how states make decisions,
and few studies attempt to compare different realist theories with formal
models and large-N empirical analysis. This article applies optimization
models from the economic literature to two variants of realist theory:
offensive and defensive realism. The key constraints in these economic
models are shown to be one way to develop clear alternative hypotheses
that distinguish between these two realist theories. These two
hypotheses are tested with data on major powers from 1820 to 2000.
The results of this analysis indicate that while there is little support for
either offensive or defensive realist theory across the entire period of
study, there is more support for defensive realism when examining the
pre-World War I and post-World War II time periods separately.

Introduction

Do major powers aggressively seek power over other states, or do states

only seek enough power in order to achieve an acceptable level of security?
This is an ongoing question in international relations research (Labs 1997),
and the answer to such a question has profound implications for the
international system. In the case of aggressive states, the international
system is a never-ending competition for power, and the losers of this
competition face insecurity and possible extinction (see, for example,
Mearsheimer 2001). On the other hand, major powers may fear what other
states may do, but this fear does not drive states to necessarily aggressively
dominate other states, but only to protect themselves (see, for example,
Waltz 1979). Whether one model of major power behavior is more
appropriate than the other is an ongoing debate in the academic literature,
and leads to vastly different arguments for how major powers should
approach foreign policy.

One strand of realism, known as “offensive realism” (Mearsheimer 2001,

5), argues “that the international system forces great power states to

1I thank Paul Diehl, Michael Findley, Robert Pahre, James Scott, Tze Kwang Teo, Brandon
Valeriano, and Mark Wolfgram for helpful comments and suggestions, and Lynsey Parsons for
research assistance. I also wish to thank the Program in Arms Control, Disarmament, and
International Security at the University of Illinois at Urbana-Champaign
(http://www.acdis.uiuc.edu) for fellowship support during this project. An earlier version of
this paper was presented at the 2005 International Studies Association-Midwest meeting in St.
Louis, MO.
46 | Rudloff

maximize their relative power because that is the optimal way to maximize
their security” (Mearsheimer 2001, 21).2 The alternative theory, known as
“defensive realism” (Lynn-Jones 1998, 157) derives primarily from Waltz
(1979), and is a theoretical perspective proposing “that the international
system does not necessarily generate intense conflict and war, and that
defensive strategies are often the best route to security” (Lynn-Jones 1998,
157–158).3 Note that while there are similarities between these two realist
theories, Snyder (2002) points out that “Where Mearsheimer departs from
Waltz is in his assertion that the search for power and security is insatiable”
(151).

The problem with distinguishing between these two theories (as well as
between realist and non-realist theory) is that both realist theories assume
that states seek some amount of power, and both theories assume that states
seek power through similar means (Snyder 2002, 151–152). In other words,
both theories assume that states engage in similar foreign policy behaviors,
but the key difference is in the degree to which states engage in these foreign
policy strategies. Ultimately, whether a strategy should be considered
offensive or defensive in character is difficult to ascertain, particularly given
that both theories argue the underlying motivation is “security”
(Mearsheimer 2001, 31; Waltz 1979, 126). For example, the United States
invasion of Iraq in 2003 can certainly be interpreted as an aggressive act, but
at the time of the invasion, the arguments of United States leaders had partly
to do with defense against weapons of mass destruction, implying at least
the possibility that decision makers were motivated partly by defensive
purposes. Recently, Russia initiated conflict against Georgia (Chivers 2008),
but does this hostility amount to aggressive, offensively oriented behavior,
or is Russia attempting to defend itself from what Russia perceives as the
potential aggression of major powers such as the United States (Finn 2008)?
Not only do these cases illustrate the difficulty of empirically distinguishing
between offensive and defensive motivations, but assuming one type of
motivation over the other may lead to vastly different policy
recommendations for the states involved.4

2The terminology “offensive realism” (Mearsheimer 2001, 5) and “maximize” (Mearsheimer

2001, 21) as well as variations of these terms are used throughout to reference this theory.
3The term “defensive realism” (Lynn-Jones 1998, 157) is adopted throughout the rest of the

paper.
4See, for example, Mearsheimer’s (2001) analysis and predictions concerning future international

relations (384–400).
Offensive Realism, Defensive Realism, and the Role of Constraints | 47

This paper proceeds by further exploring offensive and defensive

realism, as well as debates between these two perspectives. Two formal
models are formulated, one for offensive realism and another for defensive
realism. These models are then used to derive theoretical implications that
focus on the constraints facing these states (see McGinnins 1994, 71–72 for
the importance of constraints). By focusing on the fundamental constraints
that drive realist foreign policy, separate hypotheses can be derived for each
realist theory. These hypotheses allow for the testing of both whether either
realist theory provides an adequate explanation of major power foreign
policy, and which of the two realist theories provides a better explanation.
The paper concludes with an analysis of major power foreign policy
behavior between 1820 and 2000 in an attempt to use these differing
theoretical implications to empirically compare offensive and defensive
realism and determine whether one or either of the theories provides a better
explanation of major power foreign policy behavior.

Distinguishing Between Realist Theories

Realism remains an important theoretical perspective in international

relations, but over time realist theory has fragmented into a variety of
theories united under the title of realism (Lynn-Jones 1998, 157). For
example, one division is between traditional realist theories (Carr 1964;
Morgenthau 1978) and “neorealist” theory (Mearsheimer 2001; Waltz 1979).
Another division is between systemic forms of realist theory (Mearsheimer
2001; Waltz 1979) and realist theories that acknowledge the potential impact
of domestic factors and institutions (Lobell 2002/2003; Rose 1998; Snyder
1991). Yet another division is among systemic-focused realist theorists, who
disagree over whether major powers seek to maximize power (Elman 2004;
Labs 1997; Mearsheimer 1994; 2001) or whether a particular amount of
power is sufficient to ensure security (Waltz 1979). Lynn-Jones (1998)
summarizes this division concisely:
One of the most significant divides within realism is
between offensive realism and defensive realism.
Offensive realists generally argue that the international
system fosters conflict and aggression. Security is scarce,
making international competition and war
likely…Defensive realists, on the other hand, argue that
the international system does not necessarily generate
conflict and war, and defensive strategies are often the
best route to security (157–158; footnotes from original
omitted)
48 | Rudloff

This ongoing debate among realists over the virtues of offensive and
defensive realism is the topic of this paper, which attempts to model and
empirically compare these two forms of realist theory.

Many of the studies in the offensive-defensive realist debate, and the

challenge of offensive realism in particular, focus on the theoretical aspects
of the two theories (Elman 2004; Labs 1997; Lee 2002/2003; Lobell 2002/2003;
Lynn-Jones 1998; Montgomery 2006; Snyder 2002; Tang and Montgomery
2007). On the other hand, attempts to empirically distinguish between
offensive and defensive realism, or to test the implications of offensive
realism, tend to focus on small sets of case studies (Elman 2004; Feng 2005;
Labs 1997; Lobell 2002/2003; Mearsheimer 2001; Montgomery 2006).
Mearsheimer (2001) focuses on a number of major powers and whether “the
history of great-power politics involves primarily the clashing of revisionist
states” (168). A potential pitfall of such analysis is that many actions states
take can be seen as either “revisionist” or “status quo” oriented
(Mearsheimer 2001, 168–169) depending on one’s perspective. Snyder (2002),
for example, points out that
Although it is coherent and without obvious
inconsistency, it is sometimes pushed to extremes.
Exhibit A in this respect is the claim that all great
powers all the time are primarily concerned with
maximizing power so as to maximize their security. Can
it really be true that the world is condemned to a future
of constant conflict and power struggles (171).
The debates surrounding particular cases, and the motivations of states are
not easily resolved. For example, World War I remains one of the most
important wars in recent history, and yet disagreement still exists over
whether the war was ultimately caused by an overly-aggressive Germany
(Tuchman 1962), or by a breakdown in the “balance of power” (Kagan 1995,
81–231).5 Given the disagreement over such an important case, is the
generation of general empirical findings regarding the underlying
motivations by examining major power foreign policies possible?

This may explain the relative lack of large-N empirical studies

attempting to distinguish between offensive and defensive realism.
Although a number of empirical tests of realism have been conducted (see
Palmer and Morgan 2007 for such a test, as well as 335-336 for a concise

5Kagan (1995) uses the term “balance of power” periodically throughout this section (that is,
pages 81–231), for example on page 93.
Offensive Realism, Defensive Realism, and the Role of Constraints | 49

review of other tests), few focus on the distinctions between offensive and
defensive realism, or the implications of offensive realism specifically
(Lemke 2004). One possible exception is Valeriano (2009), which attempts to
test offensive realism by examining the militarized conflict behavior of
states, although this study does not directly compare offensive and defensive
variations of realism. There is a lack of large-N empirical analysis that
directly compares the empirical effectiveness of offensive and defensive
realism. This may be due to the underlying problem of motivations, and the
difficulty of discerning these motivations from actions. Without testable
implications, such a large-N comparison of offensive and defensive realism
is impossible.

One method for generating testable implications is to adopt a formal

modeling approach, which has been shown to be successful in generating
implications from realist theory (Powell 1999). Powell (1999), formalizes
realist theory although he notes that “a natural next step in the modeling
would be to allow the states to respond to a threat in more than one way”
(213). Incorporating multiple realist foreign policies addresses concerns
related to “substitutability” (Most and Starr 1989, 102–110). Some models
incorporate these concerns by adopting a modeling framework that
incorporates a number of foreign policy strategies (Morgan and Palmer 2000;
Palmer and Morgan 2006). Such studies offer a method for distinguishing
between offensive and defensive realism through the role of constraints, and
this general approach is adopted in the following pages. McGinnis (1994)
notes that:
Since preferences are not directly measurable, it is
impossible to test any assertion that an actor’s behavior
has changed because its preferences have changed...
Economists typically relate behavioral changes to
observable changes in constraints (71).
By examining the role of constraints in realist foreign policy, this study
hopes to theoretically distinguish between offensive and defensive realism in
a manner that can be empirically tested.

By focusing on foreign policy, this study may be criticized by realists as

“reductionist” (Waltz 1979, 18–37). Offensive and defensive realists both
argue that aspects of the international system are key factors in international
relations (Mearsheimer 2001, 19–22; Waltz 1979, 73–78). However, realists
themselves tend to focus on foreign policy, such that even though
“structure” (Waltz 1979, 73) may be an important causal factor, state
behavior is often the focus of theoretical and empirical analysis
50 | Rudloff

(Mearsheimer 2001, 334–359). Even so, aspects of the international system are
not abandoned in the following pages.

Two Models of Realist Foreign Policy Decision Making

The formal models in this paper must accomplish two goals. First, the
models must allow for theoretical implications that are both testable, and
distinguishable between different models of realist foreign policy. Second, the
models must easily incorporate a number of foreign policy strategies within
a single framework, such that the implications of the formal models will not
be confounded by the possibility of substitutability (Most and Starr 1989). To
fulfill both of these goals, models commonly used in the field of economics
known as “constrained optimization” models (Kreps 1990, 775–789; Simon
and Blume 1994, 411–482) are used. A number of other studies of foreign
policy decision making use these types of models in their analysis (Morgan
and Palmer 2000; Palmer and Morgan 2006). Variations in these models
match up well with realist theories, and the theoretical implications of these
models, via the role of constraints on decision making (McGinnis 1994),
allow us to more readily empirically distinguish between offensive and
defensive realism.

One common factor uniting the following models is the importance of

“relative power” (Mearsheimer 2001, 34).6 According to realists, acquiring
relative power is the first priority of states, and the difference between
offensive and defensive realism has to do with “amount” (Snyder 2002, 152).
How do states acquire power? Following realist theory, there are three
primary ways that states may achieve relative power over other states. Waltz
(1979) identifies “internal balancing” (168) and “external balancing” (168) as
two foreign policy strategies of states. These two strategies roughly
correspond to military spending and alliances (respectively), two means
through which states may increase their power according to both defensive
(see Waltz 1979 above) and offensive realists (Mearsheimer 2001, 156–157). A
third strategy available to states is military conflict, which allows states to
strengthen themselves or weaken potential major power opponents
(Mearsheimer 2001, 147-152). These three strategies form the basis of relative
power acquisition in the following two sections outlining formal models of
offensive and defensive realism. By including each of these three strategies,
the models reflect how realist theorists view foreign policy. The models,
however, also provide the ability to test whether the realist formulation of

6The term “relative power” (Mearsheimer 2001, 34) is used throughout the rest of the paper.
Offensive Realism, Defensive Realism, and the Role of Constraints | 51

foreign policy is effective, by allowing for the empirical examination of each

of the three strategies separately.

Offensive Realism

As noted earlier, the offensive realism view of major power foreign

policy is that states attempt to maximize relative power (Labs 1997;
Mearsheimer 1994; 2001). States, however, are not able to simply acquire an
infinite amount of power due to limited resources (Mearsheimer 2001, 60–
75). These resources serve as a significant constraint, in that states must deal
with these limitations when attempting to maximize their relative power.

In the offensive realism model, relative power is represented by , and

relative power is assumed to be a function of each of the three realist foreign
policies outlined above: military spending ( ), alliances ( ), and conflict ( ).
Each variable represents the investment in a particular foreign policy
strategy, and are the key variables in the model that can be manipulated by
states. That is to say, states do not produce relative power directly, but
instead invest in foreign policy strategies, which in turn lead to relative
power. The relative power equation takes the form of a Cobb-Douglas
function (Cobb and Douglas 1928), such that:
(1)
where , , and represent parameters that translate investment in each of
the realist foreign policies into power (each of these three parameters are
assumed to be greater than 0 and less than 1 (Sydsaeter, Strom, and Berck
2000, 156)). This equation, , represents the key function in the
production of power for major powers. Note that an increase in investment
in each of these strategies leads to an increase in power, so higher levels of a
strategy such as military spending (i.e. ) leads to higher levels of power
( ). The Cobb-Douglas function (Cobb and Douglas 1928) is useful because
of its mathematical clarity, and is used in many formal models of foreign
policy (Morgan and Palmer 2000; Palmer and Morgan 2006). One of the
further assumptions of the model is that states invest in a positive and non-
zero amount of each foreign policy strategy (i.e. , , ). In other words,
states invest in a mix of strategies (this will be further explored in terms of
the equilibria below).

Equation 1 is only part of the story. If this were the only consideration of
states seeking to maximize their power, then states would simply produce
infinite amounts of power by investing in infinite amounts of each of the
52 | Rudloff

realist foreign policies. Of course, states cannot invest in infinite amounts of

any foreign policy because states are limited in their capabilities of
investment (see Mearsheimer 2001, 60–75). Due to this limitation, states take
into consideration their level of capabilities ( ). A state cannot invest more
in foreign policy strategies than their capabilities allow, but a state that
wishes to maximize relative power will spend all of their capabilities on
producing relative power. Therefore, a state maximizes equation 1 subject to
the following capability constraint:
(2)
where , , and are parameters that translate investment in each of the
strategies into capabilities, . The capabilities function is represented as an
equality, because states will not spend less than their capabilities. By
investing less than , a state would produce less than the maximum amount
of power, which would not occur with states attempting to maximize
relative power.

These two equations may be a simplification, however, this

formalization attempts to capture only the essential aspects of major power
decision making as described by offensive realist theory. For example, one
might argue that states do not spend all of their capabilities seeking relative
power. Mearsheimer (2001), after all, argues that “States care about relative
wealth, because economic might is the foundation of military might” (143),
and as some have pointed out, investing in strategies such as military
spending may ultimately decrease a state’s economic capability (Heo and
Eger 2005). Even if this is the case, does not necessarily represent the
entirety of a state’s economic resources. For example, may represent the
amount of resources available to be spent on realist foreign policies that will
not compromise the state’s future economic ability to invest in realist foreign
policy strategies.

How much do states invest in each of the foreign policies? This depends
on a variety of factors, namely the level of a state’s capabilities ( ), and the
translation parameters described above ( , , , , , and ). The optimal
investment in each of the foreign policies is determined by a process
described in Simon and Blume (1994, 411–482).7 The optimal levels of each
strategy are represented by the following three equations (the “ ” represents

7Given space limitations, a full step-by-step derivation of equilibria is not provided, but instead
the reader can refer to Simon and Blume (1994, 411-482).
Offensive Realism, Defensive Realism, and the Role of Constraints | 53

the fact that these are optimal levels of foreign policy investments for
offensive realists):
(3)

(4)

(5)

Note that what each of these equations has in common is the role of
capabilities ( ). In each of these equations, an increase in capabilities ( )
leads to an increase in each of the realist foreign policies ( , , and ).
Furthermore, note that as long as each of the parameters , , and are
between 0 and 1 and there is a positive level of economic resources (i.e.
), a state will invest in at least some of each strategy. This does not
mean, however, that a state will invest in equal amounts of each strategy.
Depending on how much each individual strategy is translated into power
(through , , and ), a state may invest much more in one or a set of
strategies compared to another. Even in these cases, however, an increase in
economic resources will lead to increases in each of the strategies, even
though some of these increases in strategies are quite small.

The above equilibria indicate that states will increase each of these
power seeking strategies given an increase in available economic resources
because more resources means that the states can potentially gain more
relative power through increasing investment. In other words, the constraint
is the key driver of state investment in realist foreign policies (see McGinnis
1994, 71–72). This finding leads to the offensive realism hypotheses:

Offensive Realism Hypothesis: Military Spending As the economic

capabilities of a major power state increase, the major power will
subsequently increase its military spending.
Offensive Realism Hypothesis: Alliances As the economic capabilities of a
major power state increase, the major power will subsequently increase its
membership in alliances with other states.
Offensive Realism Hypothesis: Conflict As the economic capabilities of a
major power state increase, the major power will subsequently increase its
use of conflict against other states.

These hypotheses represent the key theoretical implication of the

offensive realism model. Note that this relationship holds, despite the levels
54 | Rudloff

of the parameters , , and , which alter the attractiveness of an individual

strategy relative to the others. For example, assume that a major power is
maximizing a version of equation 1, where is much greater than either or
, thus meaning that investments in conflict ( ) result in larger contributions
to power than investments in other power strategies (see Mearsheimer 2001,
344 for such an argument). Even in this case, the expectation is that increases
in capabilities ( ) lead to an increase in each strategy rather than an increase
in only conflict. A further benefit of this approach is the testing of each
individual foreign policy strategy, as well as offensive realism as a whole.
Analysis may indicate that realist theory provides an explanation of only
certain strategies. As detailed in the next section, the relationship between
capabilities and each power strategy provides a method for distinguishing
between offensive and defensive realism forms of foreign policy.

Defensive Realism

Under the defensive realism decision making framework, relative power

is important, but “the first concern of states is not to maximize power but to
maintain their position in the system” (Waltz 1979, 126). Furthermore,
assuming that states rely on capabilities for a variety of ends, including
power (Waltz 1979, 126), then capabilities can no longer be seen as the
primary constraint on power seeking behavior. An alternative approach to
this problem assumes that states seek both power ( ) and “other goals”
(Waltz 1979, 126)8 (these non-power goals are referred to as ), but that
ultimately, states seek to maximize these other goals, despite the fact that
states still need power, and power and other goals both require capabilities.
In other words, the decision making problem major powers face involves the
maximization of utility in the following manner 9:
(6)
Under this framework, states increase utility ( ) by investing capabilities
into achieving its other goals ( ), while investing capabilities into achieving
power ( ) decreases the utility of the state. If this were the only calculation
that a state was required to make, then the state would invest all of its

8The term “other goals” (Waltz 1979, 126) is adopted throughout. Although Waltz (1979)
describes some of these “other goals” (126), one can still solve the model without specifying
these policy ends.
9Note that the model outlined in this section is the model used to explain the security policy

behaviors of minor powers in my dissertation (Rudloff 2007).

Offensive Realism, Defensive Realism, and the Role of Constraints | 55

capabilities into achieving its other goals ( ), and would not invest any
capabilities in power seeking foreign policies (such as , , and ).

Unfortunately for major powers, they must invest some amount of

capabilities into power, in order to achieve security (Waltz 1979, 126). As
suggested by some realists, however, states do not need to maximize power,
but rather invest in enough power to achieve security (Snyder 2002, 151–
155). Waltz (1979) argues that “Only if survival is assured can states safely
seek such other goals as tranquility, profits, and power” (126). In other
words, defensive realists seek enough relative power to achieve security, but
do not necessarily seek more power, because seeking additional power
would require the state to compromise other goals ( ) that represent the true
goals of the state.10 This implies that instead of seeking to maximize power
(as in equation 1), states seek to minimize the costs of security (see Simon
and Blume 1994, 436–439).

If the primary motivation of states is to minimize the capabilities

invested in realist foreign policies, namely military spending ( ), alliances
( ), and militarized conflict ( ), then this focus on costs can be represented
in the following equation:
(7)
Note that this is the same as equation 2, except in the offensive realism
section this represents the key constrain, whereas equation 7 represents what
is to be minimized for defensive realists. This model is consistent with the
statement by Lobell (2002/2003) that “A defensive security strategy involves
fiscal and monetary restraint, enhancing security through low cost defensive
arrangements” (168), although Lobell (2002/2003) considers a different set of
strategies than considered here (168). As above, the parameters , , and
are used to translate the investment of each of the realist foreign policies into
capabilities. Also, note that unlike the offensive realism model, where major
powers are maximizing power, in this model major powers minimize the
amount of capabilities invested in power. Fortunately, both of these models
can be addressed within a common “optimization” framework (Kreps 1990,
775–789; Simon and Blume 1994, 411–482).

As above, the constraint is quite important (McGinnis 1994, 71–72). If

major powers wish to minimize without constraints, then major powers

10Such a framework is similar to Maslow (1943).

56 | Rudloff

simply invest no capabilities in realist foreign policies, would equal zero,

and equation 6 would be minimized. Major powers, however, must produce
a finite amount of power in order to achieve security, and this amount of
power ( ) represents the key constraint for the defensive realist:
(8)
Notice again, that equations 8 and 1 are identical. As before, the parameters
, , and translate investment in each strategy into power. In the defensive
realism case, however, the power equation represents the constraint. As
above, the constraint becomes a primary consideration for defensive realist
states, and this constraint allows us to easily distinguish between offensive
and defensive realism.

Ultimately, the focus is on the foreign policies of states, because this is

what states manipulate within these decision making frameworks. In other
words, the investment in military spending ( ), alliances ( ), and conflict
( ) must be found that minimizes equation 7 while following the constraint
in equation 8. The defensive realist decision making equations are solved in
much the same way that the offensive realist equations are solved (see Simon
and Blume 1994, 411–482), and are presented below, where “ ” represents
that these are optimal levels of foreign policy investments for the defensive
realist state:

(9)

(10)

(11)
Note that the constraint ( ) appears in each of the three optimal investment
equations (McGinnis 1994, 71–72). In the case of these equations, an increase
in results in an increase in each of the realist foreign policies ( , , and
).

How do states know what amount of power is needed in order to

achieve security? The amount of power needed to achieve security varies
across states, and is a function of the amount of threat facing a state. For
realists, the amount of threat facing a state is directly related to the amount
of power possessed by other states (see Waltz 1979, 126). Given that power
represents threat, which in turn determines the amount of power a state
Offensive Realism, Defensive Realism, and the Role of Constraints | 57

needs to achieve security, the following hypotheses summarize the

theoretical implications of the defensive realist model:

Defensive Realism Hypothesis: Military Spending As the threat facing a

major power state increases, the major power will subsequently increase its
military spending.
Defensive Realism Hypothesis: Alliances As the threat facing a major
power state increases, the major power will subsequently increase its
membership in alliances with other states.
Defensive Realism Hypothesis: Conflict As the threat facing a major power
state increases, the major power will subsequently increase its use of
conflict against other states.

Note that the constraint drives the defensive realism hypothesis as the
constraint drives the offensive realism hypothesis (see McGinnis 1994, 71–
72), but that the constraint is different in each model. Furthermore, the
relationship between threat and each strategy is expected despite variation
in those factors (represented by , , and ) that make particular strategies
more effective. The difference in constraints distinguishes between the two
realist models, and serves as the basis of the empirical comparison between
offensive and defensive realist theories in the following sections.

Testing Realist Foreign Policy Behavior

The focus of the analysis is on the periods between 1816-1916 and 1946-
2000, because a variety of different potential major powers exist during these
periods, and the period spans a variety of different phases in international
relations, including the Cold War. This means that the hypotheses can be
tested in a variety of international contexts, and more importantly for realist
theory, in a variety of “polarity” (James 1995, 183) contexts. Realists argue
that the concentration and “distribution of power” (Mearsheimer 2001, 348)
in the international system represent important determinants of how states
behave (Mearsheimer 2001, 334–359). This argument can be examined in the
following analysis by examining the effects of defensive and offensive realist
variables in different time periods.

Realists typically argue that the only important states in the international
system are major powers (Mearsheimer 2001, 17), so in order to test realist
theories on their own terms, the focus here is also on major powers. The
operationalization of “major power” is potentially controversial, because
power (a continuous characteristic) must be used to create a dichotomous
58 | Rudloff

distinction (i.e. major power versus other types of states). The list of major
powers developed by the Correlates of War Project (2005b)11 is used to
establish which states are major powers, leading to nine major powers
between 1816-1916 and 1946-2000: the United Kingdom, the United States,
France, Germany, Austria-Hungary, Italy, Russia, China and Japan
(Correlates of War 2005b). Table 1 lists the major powers and the dates these
states are considered to qualify as major powers, according to the Correlates
of War Project (Correlates of War 2005b). In the nineteenth century, there are
significant gaps in the economic data (as will be discussed later), and so
many of these states (such as Germany before 1870 (Maddison 2009)) must
be dropped from the analysis due to missing data. See the “Years in
Analysis” column of Table 1.

Table 1: List of Major Power and Years in Data Set

Major Power Years According to COW Years in Analysis

United Kingdom 1816-1916 1830-1916

1946-2000 1946-2000
France 1816-1916 1820-1916
1946-2000 1946-2000
Russia 1816-1916 1946-2000
1946-2000
Germany 1816-1916 1850-1916
1991-2000 1991-2000
Austria-Hungary 1816-1916 1870-1916

Italy 1860-1916 1861-1916

Japan 1895-1916 1895-1916

1991-2000 1991-2000
United States 1898-1916 1898-1916
1946-2000 1946-2000
China 1950-2000 1950-2000

Source: Correlates of War (COW) Project (2005b); Maddison (2009)

The following sections outline the dependent variables, independent

variables, and statistical estimators to be used. The analysis is complicated
by the fact that the above formal models result in three different equilibria,
meaning that three sets of analysis must be conducted to fully test the

11More information on the Correlates of War Project can be found at

http://www.correlatesofwar.org.
Offensive Realism, Defensive Realism, and the Role of Constraints | 59

theoretical implications of offensive and defensive realism. This requires

different statistical methods to accommodate the different types of
dependent variables.

Dependent Variables

According to the above analysis, each realist foreign policy is a potential

dependent variable. Given the theoretical implications of the model, separate
dependent variables for each strategy are constructed: military spending,
alliances, and conflict.

Military Spending. The military spending variable is constructed from

the Correlates of War Project’s National Military Capabilities data set,
version 3.02 (Singer, Bremer and Stuckey 1972).12 Several complications with
this data must be addressed before analysis can be completed. First, the data
are presented in two different currencies. Before 1914, the data is in British
pounds, while after 1913 the data is United States dollars (Correlates of War
2005a, 17). In order to transform the data into a single currency, the currency
figures prior to 1914 are converted from British pounds to United States
dollars using Officer’s (2008) currency conversion data. 13 As a result, all of
the data are in a single currency, United States dollars.

The second issue is that the data is in “current” values for each year
(Correlates of War 2005a, 20), meaning that the data is not controlled for
inflation. Comparing currency values at different times is problematic
because a United States dollar in 1856 is not the same as a United States
dollar in 1956. Sahr’s (2009) data on inflation is used to convert the military
spending data, such that the data is in year 2000 United States dollars, 14
making the comparison of data possible between the nineteenth and
twentieth century.

The military spending variable measures the year to year change in the
military spending of a state by subtracting the previous year’s military

12The Correlates of War Project’s National Military Capabilities data set can be downloaded
from http://www.correlatesofwar.org.
13Officer’s (2008) currency conversion data was downloaded from

http://www.measuringworth.com on March 8, 2008. This website contains a variety of

economic data in addition to currency conversion data.
14Sahr’s (2009) data was downloaded on October 25, 2009 from

http://oregonstate.edu/cla/polisci/faculty-research/sahr/sahr.htm. This website contains a

variety of data on inflation.
60 | Rudloff

spending from the current year. This results in a continuous variable that
varies from negative values (indicating a decrease in military spending) to
positive values (indicating an increase in military spending).

Alliances. The alliance data is constructed from the Correlates of War

Project’s Formal Alliance data set, version 3.03 (Gibler and Sarkees 2004). For
the purposes of constructing the alliance data, all alliances in the Correlates
of War data set are included: “defense pacts”, “neutrality or non-aggression
pact”, and “entente” (Gibler and Sarkees 2004, 212). Two variables were
constructed from this data to measure the change in alliance strategies. The
first is a variable indicating whether a state entered into an alliance in a
given year, and the second is a variable indicating whether a state exited an
alliance in a given year. In this way, the analysis can measure both increases
and decreases in alliance behavior over time.

Conflict. The conflict variable is constructed using the Correlates of War

Project’s Militarized Interstate Dispute data, version 3.10 (Ghosn, Palmer and
Bremer 2004).15 As with the alliance data, the conflict data is an indicator
variable, which measures whether an event (i.e. “militarized interstate
disputes” (Ghosn, Palmer and Bremer 2004)) occurs in a given year. Conflicts
only in the year in which the dispute begins or the “start year of
participation in dispute” (Ghosn and Palmer 2003, 4) are included. Conflict
initiation is a rare event. Furthermore, many MIDs last for only short periods
of time, and are thus contained within the same year. Both these facts make
measuring change over time difficult, which is why the analysis relies on the
indicator of conflict initiation. Alternative measures of the conflict variable
are a count of the number of conflict initiations, and the year to year
difference in the number of disputes initiated. Results for these alternative
variables are noted later in the paper.

Independent Variables

The statistical analysis centers on two key independent variables,

economic resources and threat, corresponding to the two realist hypotheses
(offensive realism and defensive realism, respectively). The economic
resources variable corresponds to in the above formal model of offensive
realism, or the general capabilities of a state as indicated in equations 3, 4,
and 5. The threat variable corresponds to in the above defensive realism

15The
Correlates of War Project’s Militarized Interstate Dispute data set (Ghosn, Palmer and
Bremer 2004) can be downloaded from http://www.correlatesofwar.org.
Offensive Realism, Defensive Realism, and the Role of Constraints | 61

formal model, or the relative power needed to ensure security in equations 9,

10, and 11. Several other variables are included as controls. Where
appropriate, the year-to-year change in these variables are measured to
analyze whether changes in the independent variables lead to appropriate
changes in the dependent variables. All of the independent variables are
lagged a year.

Economic Resources. The economic constraint variable developed in the

above formal models ( ) are measured with the economic capability of each
major power in a given year. As noted above, realists acknowledge
(Mearsheimer 2001, 143), and other theorists argue (Heo and Eger 2005) that
one cannot simply ignore economic capability considerations. Instead, the
constraint represents some portion of a state’s economic capability that will
be used to invest in each power strategy. Realists such as Mearsheimer
(2001), however, do not provide clear empirical guidance as to what portion
of economic capability will be used. Assuming that overall economic
capability is a proxy for the overall amount that will be used to invest in the
strategies is, however, reasonable. As the overall economic capabilities of a
state increase, so too will the amount of resources that can be used without
compromising the future ability of the state to invest in power strategies.

The economic resources variable is constructed from Maddison’s (2009)

“Statistics on World Population, GDP, and Per Capita GDP, 1 - 2006 AD”,
which was downloaded on October 25, 2009.16 Maddison’s (2009) data
includes a variety of states from 1820 onward, although some data is missing
for various states.17 Due to missing data, the number of state-years included
in the final analysis is smaller than the Correlates of War Project’s (2005b) list
of major powers suggests. Table 1 contains a column entitled “Years in
Analysis” which lists the years each state is included in the analysis as a
result of the Maddison (2009) data. Although this limits the number of cases,

16Maddison’s (2009) gross domestic product data was downloaded from

http://www.ggdc.net/maddison/, and the website contains a variety of information on
economic data collected for various states. Although “Maddison (2009)” is cited throughout in
reference to the data, note that this data is associated with a variety of written works (Maddison
1995; 2001; 2003). Throughout the rest of the paper, however, references to “Maddison (2009)” as
the source of the data are made.
17A variety of states do not have economic data for a variety of years in Maddison’s (2009) data

set. In addition to these missing values, several changes to the data are made, because there are
a number of states in the nineteenth century that have a observation for a given year, but not
proceeding and subsequent years (Maddison 2009). First, the Austria-Hungary observations
begin in 1861. Second, the Germany observations begin in 1850. Third, the Russia observations
begin in 1946. Fourth, the Italy observations begin in 1861.
62 | Rudloff

the list can be viewed as a more conservative estimation of which states are
major powers in a particular year.18 In order to measure change, the
economic resources variable measures year-to-year changes in gross
domestic product (GDP). The variable is constructed by subtracting the
previous year’s GDP from a given year’s GDP.

Threat. The threat variable is constructed from the Correlates of War

Project’s National Military Capabilities data set, version 3.02 (Singer, Bremer
and Stuckey 1972).19 This is the same data used above to estimate a state’s
military spending. In order to estimate the threat facing a state, the sum is
taken of military spending for every other major power (Correlates of War
2005b). The military spending data is converted into United States dollars for
the years before 1914 by using Officer’s (2008) data on currency conversion. 20
Second, the data is controlled for inflation using Sahr’s (2009) data for 2000
United States dollars.21 Finally, the data was transformed into a change
variable by subtracting the previous year’s threat from a given year’s threat.

Control Variables. Two variables are constructed to measure the

systemic power distribution. First, a polarity variable is intended to measure
the manner in which power is distributed in the international system for a
given year (see James 1995). As described above, realists argue that polarity
is an important causal factor in explaining state behavior (for example,
Mearsheimer 2001, 334–359), and including a variable for polarity will help
to control for power characteristics at the system level that may potentially
affect the foreign policies of major powers (see Valeriano 2009). The polarity
distinction of “bipolarity” and “multipolarity” from Mearsheimer (2001, 348)
is adopted, dividing the time period into three categories: “multipolarity”
from 1816–1916, “bipolarity” from 1946–1990, and “multipolarity” from
1991–2000 (see Mearsheimer 2001, 348 for the original measure on which this

18The Maddison (2009) data already appear to be controlled for inflation. To help ascertain the
validity of Maddison’s (2009) data, a correlation of Maddison’s (2009) data with Gleditsch’s
(2002) Expanded Trade and GDP Data, version 4.1, which contains gross domestic product data
(this data is also controlled for inflation) was created. Although the Gleditsch (2002) data only
begins in 1950, the correlation between the Maddison (2009) and Gleditsch (2002) data for the
1950–2000 period is quite high (i.e. 0.9). The Maddison (2009) data is therefore consistent
with another gross domestic product data set.
19The Correlates of War Project’s National Military Capabilities data set can be downloaded

from http://www.correlatesofwar.org.
20Officer’s (2008) currency conversion data was downloaded from

http://www.measuringworth.com on March 8, 2008.

21Sahr’s (2009) inflation data was downloaded from

http://oregonstate.edu/cla/polisci/faculty-research/sahr/sahr.htm on October 25, 2009.

Offensive Realism, Defensive Realism, and the Role of Constraints | 63

is adapted). Second, a variable providing a continuous measure of the

concentration of power in the international system is created by summing
the power of all the major powers. This is done by adding together the major
powers’ Correlates of War “Composite Index of National Capability”
(Correlates of War 2010, 1) in each year. This provides a rough measure of
how much power the major powers possess in the international system.

An additional indicator variable is added to measure whether or not a

state is involved in an ongoing war according to the Correlates of War Inter-
state War data (Sarkees and Wayman 2010).22 States involved in a war may
be more likely to increase strategies such as military spending, regardless of
the threat from other major powers. Finally, two non-realist variables are
also included to provide additional controls. The first is a variable measuring
the year-to-year change in the total imports and exports of a major power
(Barbieri, Keshk, and Pollins 2009; Barbieri and Keshk 2012). 23 The second is
the change in the PolityIV (Marshall, Jaggers, and Gurr 2011) regime type
measurement.24 These latter two variables are included in a separate analysis
as each is not a realist variable, and the trade data is limited to the time
period after 1870 (Barbieri, Keshk, and Pollins 2009; Barbieri and Keshk
2012).

Statistical Estimation

Given the variety of dependent variables, utilizing a single statistical

estimator in subsequent analysis is impossible. Therefore, an appropriate
estimator for the distribution of each of the dependent variables must be
used. To analyze the military spending of major powers, a feasible
generalized least squares model is presented, which is appropriate given the
continuous distribution of the military spending variable (Greene 2003, 322).
The alliance and conflict variables are analyzed with a logit model, because
each of these variables represents whether a particular event takes place
(Long and Freese 2006, 129-140). Random effects are included to account for
the time series cross-sectional data, helping to control for unit-level effects
(Greene 2003, 293-295). Finally, for each of these random effects models, the
year is included as an independent variable. Given that some variables are
generally increasing over time (particularly military spending), inclusion of

22The Correlates of War Inter-State data, version 4.0 can be downloaded from
http://www.correlatesofwar.org.
23The Correlates of War Bilateral Trade data can be downloaded at

http://www.correlatesofwar.org.
24The Polity IV dataset can be downloaded from http://www.systemicpeace.org.
64 | Rudloff

year will help control for this year-to-year dependence in the data (see Drury
and Li 2006, 319 for a similar use of a temporal variable).

There are significant changes across the time period of study, including
significant changes in the states that are considered major powers. Due to
this, not only is the overall model included for all of the years in the data, but
also a separate analysis for the time period before World War I and after
World War II. Such a division in the data is justified for both theoretical and
empirical reasons. First, both offensive and defensive realists believe that the
structure of the international system is important (Mearsheimer 2001; Waltz
1979), and there are important changes in this structure that occur before and
after the World Wars (see Mearsheimer 2001, 348). Second, other empirical
studies demonstrate that important changes may occur between different
time periods (Senese and Vasquez 2005).25 This division allows for the
examination of this possibility.

Empirical Analysis of Offensive and Defensive Realism

Table 2 summarizes the results of the military spending analysis. There

is little evidence indicating that either of the realist models provide an
adequate explanation for the year-to-year change in military spending when
examining the entire time period. Models 1 and 2 illustrate that both key
realist variables, while positive as one would expect, are not statistically
significant. A further check was conducted that split each of the two key
variables (economic resources and threat) into two separate models with the
same controls as in Model 1, as suggested by Clarke (2001). In both models,
each of these variables remained statistically insignificant.

A more interesting story emerges with a breakdown of the years in the

analysis into those prior to World War I and those after World War II. While
offensive realism receives no support in either case, there is some evidence to
suggest that defensive realist considerations may be operating prior to
World War I. In Model 3, the change in threat variable is positive and
statistically significant, indicating that as the total military spending of other
major powers increase year-to-year, a major power is likely to increase its
own military expenditures. The effect of this relationship is also sizable, with
an increase of one billion United States dollars in the change of military

25Thetwo time periods adopted for this paper (1820–1917 and 1946–2000) are similar to those
found in Senese and Vasquez (2005).
Offensive Realism, Defensive Realism, and the Role of Constraints | 65

spending of other states leading to an increase of 28 million United States

dollars in the change of a state’s own military spending.

Table 2: Military Spending Analysis

Model 1 Model 2 Model 3 Model 4
All Years All Years 1820-1917 1946–2000

GDPt-1 6.65 10.24 -115.15** 8.86

(10.53) (10.90) (36.94) (15.94)
Threatt-1 0.01 0.008 0.03** -0.01
(0.01) (0.01) (0.01) (0.02)
Polarityt-1 -25.11** -4.70 - -32.96**
(7.65) (6.19) - (12.59)
Power Concentrationt-1 60.16** 78.87** 3.00 235.48**
(27.39) (23.27) (9.32) (80.66)
Wart-1 4.15** 4.00** 2.25** 3.21
(1.97) (1.72) (0.80) (4.41)
Polityt-1 - -1.47 - -
- (1.13) - -
Total Tradet-1 - -37.52* - -
- (20.53) - -
Year -0.002 -0.001 0.03** -0.24**
(0.01) (0.01) (0.01) (0.10)
Constant 4.46 3.11 -64.57** 473.81**
(25.80) (28.49) (19.53) (196.90)
N 658 446 381 277

0.002 < 0.001 < 0.001 0.003

Dependent variable: change in military spending (spendingt – spendingt-1)

Parentheses contain standard errors; * = p < 0.1, ** = p < 0.05, *** = p < 0.01

A similar story emerges when one considers changes in alliances. As

Table 3 demonstrates, there remains little overall support for either of the
realist theories if one considers the entire time period. In both Model 1 and
Model 2, neither of the key realist variables is in the direction one would
expect given the realist models presented above. An increase in either
constraint (economic resources or threat) should result in new alliance
formation, as major powers seek new alliance partners. Instead, both
variables are negative.26

26Ifone separates the model in Model 1 into two models, each including only one of the key
realist variables (see Clarke 2001), the results do not change substantively.
66 | Rudloff

Table 3: Alliance Entrance Analysis

Model 1 Model 2 Model 3 Model 4
All Years All Years 1820-1917 1946–2000
GDPt-1 -0.51 -0.64 -13.31 -0.33
(1.83) (3.38) (20.82) (2.22)
Threatt-1 -0.006** -0.006** 0.01** -0.01***
(0.002) (0.002) (0.005) (0.003)
Polarityt-1 1.76* 0.63 - 2.70**
(1.04) (1.31) - (1.22)
Power Concentrationt-1 -4.92 -2.41 -13.56* 1.65
(5.20) (6.11) (7.80) (7.68)
Wart-1 1.10*** 1.28*** 0.58 1.17**
(0.30) (0.37) (0.40) (0.53)
Polityt-1 - -0.15 - -
- (0.28) - -
Total Tradet-1 - -6.41 - -
- (5.91) - -
Year -0.007** -0.007* -0.005 -0.05**
(0.003) (0.004) (0.006) (0.01)
Constant 10.82** 11.54 8.62 93.67**
(4.89) (7.49) (10.57) (27.90)
N 658 446 381 277

< 0.001 0.016 0.021 < 0.001

Dependent variable: indicator of alliance entrance in given year

Parentheses contain standard errors; * = p < 0.1, ** = p < 0.05, *** = p < 0.01

A more interesting picture emerges when the time period is separated

into pre-World War I and post-World War II models, however. As with the
model of changes in military spending, the pre-World War I model indicates
that defensive realism appears to be operating as expected. An increase in
the military spending of other major powers leads a major power to enter
into alliances, as the threat variable in Model 3 is both positive and
statistically significant at 0.05 level. In the pre-World War I period, the
predicted probability of entering an alliance increases by 416% (from a
probability of 14.7% to a probability of 75.8%) if one increases the change in
threat variable from its minimum (yearly decrease of 11.6 billion United
States dollars) to its maximum (yearly increase of 220.9 billion United States
dollars) (Long and Freese 2006, 160-162).27 Once again, defensive realism is

27To calculate predicted probabilities, the model was run without random effects (i.e. as a
regular logit). The threat variable remained positive and statistically significant even without
the inclusion of random effects. Using SPost, the economic resources and power concentration
Offensive Realism, Defensive Realism, and the Role of Constraints | 67

empirically supported more than offensive realism, but this support is

limited to the pre-World War I years.

Defensive realism also receives more support in the case of exiting

alliances. Unlike with entering an alliance, the expectation is that as the key
realist constraint variable increase, there is a decrease in the alliances a state
will be a party to. In the defensive realism context, as threat decreases, a
states is more likely to exit existing alliances, as the security provided is no
longer needed. Table 4 summarizes the analysis for the exiting of alliances.
Although the key realist constraint variables are negative in Model 1 and
Model 2, neither is statistically significant at the 0.05 level (although both are
statistically significant at the 0.1 level).28 Furthermore, the overall fit of the

Table 4: Alliance Exiting Analysis

Model 1 Model 2 Model 3 Model 4
All Years All Years 1820-1917 1946–2000
GDPt-1 -3.78* -2.74 8.62 -3.92*
(2.07) (3.75) (25.33) (2.12)
Threatt-1 -0.005* -0.003 0.005 -0.005**
(0.002) (0.003) (0.005) (0.003)
Polarityt-1 -22.67 -18.77 - -22.50
(136469) (18704) - (168076)
Power Concentrationt-1 -2.35 2.00 -3.50 3.26
(5.61) (6.04) (7.15) (9.77)
Wart-1 0.19 0.24 0.51 -1.64
(0.37) (0.49) (0.45) (1.06)
Polityt-1 - 0.04 - -
- (0.31) - -
Total Tradet-1 - 4.43 - -
- (6.20) - -
Year -0.004 -0.002 -0.01** -0.03**
(0.003) (0.004) (0.006) (0.01)
Constant 5.95 1.82 22.15* 53.62**
(4.78) (8.41) (12.03) (26.78)
N 658 446 381 277
0.098 0.984 0.272 0.028
Dependent variable: indicator of alliance exit in given year
Parentheses contain standard errors; * = p < 0.1, ** = p < 0.05, *** = p < 0.01

variables are held at their mean, the year and ongoing war variables are held at their medians
(Long and Freese 2006, 160-162).
28This remains true, even if one runs separate models on each of the realist constraint variables

with the same controls as in Model 1 (see Clarke 2001).

68 | Rudloff

model is problematic, making it difficult to draw solid conclusions from the

analysis.

Once again, if one splits the model into pre-World War I and post-World
War II years, a more complicated picture emerges. Once again, the defensive
realist model appears to be more supported than the offensive realist model.
Unlike previous analysis, however, this is the case in the post-World War II
period rather than the pre-World War I period. Still, in the case of the post-
World War II model, the variable is negative and statistically significant, as
the defensive realist model suggests. The predicted probability of exiting an
alliance decreases by 90% (from a probability of 47.7% to a probability of
4.9%) as the change in threat variable increases from its minimum (yearly
decrease of 391.7 billion United States dollars) to its maximum (yearly
increase of 148.7 billion United States dollars) in the Model 4 model (Long
and Freese 2006, 160-162).29

Table 5 summarizes the results of the conflict analysis, and there appears
to be no support for either model. The key realist constraints do not
demonstrate support for either of the realist theories across all four of the
models. Increases in either realist constraint does not lead to a clear increase
in the conflict involvement of major powers. 30 Even in the pre-World War I
and post-World War II models, there is little evidence that either realist
constraint is driving conflict involvement.

Given the complication of any conflict variable noted in the research

design, a number of other models were run with various specifications of the
conflict dependent variable. First, a count of the number of conflicts initiated
(Ghosn, Palmer, and Bremer 2004) in a given year was constructed (rather
than a simple indication of conflict involvement). A negative binomial
regression was run with this variable as the dependent variable, and
although the results were generally consistent with the results in Table 5, the
variable measuring threat was both positive and statistically significant in
the model of pre-World War I years. This finding in the alternative model is

29As with the calculation of predicted probabilities for entering alliances, the model from which
the predicted probabilities were calculated was a regular logit (absent random effects). The
threat variable remained negative and statistically significant without the inclusion of random
effects. With SPost, the year and ongoing war variables are held constant at their medians, while
the economic resources and power concentration variables are held constant at their means
(Long and Freese 2006, 160-162). The polarity variable was dropped from the regular logit
model due to a lack of variation when associated with the dependent variable.
30This is also the case when one splits the model in Model 1 to account for the potentially

separate effects of the two realist constraints (see Clarke 2001).

Offensive Realism, Defensive Realism, and the Role of Constraints | 69

consistent with previous models that indicate that defensive realism receives
more support when one breaks down the years by time period, and similar
to the alliance entrance and military spending model in its indication that
defensive realism operates in pre-World War I years.

Table 5: Conflict Analysis

Model 1 Model 2 Model 3 Model 4
All Years All Years 1820-1917 1946–2000
GDPt-1 2.17 1.11 -5.51 3.01*
(1.62) (2.48) (18.88) (1.81)
Threatt-1 -0.00007 -0.001 0.006 -0.002
(0.002) (0.002) (0.005) (0.002)
Polarityt-1 1.29 1.02 - 2.19
(1.21) (1.27) - (1.34)
Power Concentrationt-1 -2.82 -0.31 -1.64 -12.91
(3.82) (4.37) (4.31) (8.03)
Wart-1 0.22 0.28 -0.29 0.62
(0.26) (0.33) (0.38) (0.44)
Polityt-1 - -0.51 - -
- (0.31) - -
Total Tradet-1 - -0.38 - -
- (4.03) - -
Year -0.001 -0.004 0.01* -0.01
(0.002) (0.003) (0.01) (0.01)
Constant 2.52 6.81 -17.21* 20.57
(3.96) (6.12) (9.74) (19.86)
N 658 446 381 277
0.685 0.744 0.322 0.132
Dependent variable: indicator of conflict involvement in given year
Parentheses contain standard errors; * = p < 0.1, ** = p < 0.05, *** = p < 0.01

In the case of a count variable, however, this still only measures the
number of conflicts initiated in a year, and does not give any indication of
the change in conflict from year to year. In order to better measure the change
in conflict behavior, the alternative count variable was used to create a new
variable indicating the year-to-year change in the number of conflicts
initiated (by subtracting the previous year’s number of conflicts from the
current year’s number of conflicts). As with the conflict involvement variable
demonstrated in Table 5, however, there is no support for either the
defensive or offensive realism models. Given the weak support for either
realist theory in terms of conflict, neither theory provides a robust
explanation of conflict compared to the other two foreign policy strategies.
70 | Rudloff

This may suggest a general weakness in realist theory for explaining conflict
strategies (see Mearsheimer 2001, 147-152).

What conclusions can be drawn from the analysis? First, there is no

evidence that either offensive realism or defensive realism is at work across
the entire time period and across all states. The analysis here does not allow
us to distinguish between the two realist models. That is, there is no
evidence to suggest whether offensive realism or defensive realism is more
appropriate than the other across the previous two centuries. Indeed, the
analysis indicates that the key constraints that theoretically drive both realist
models are ineffective as a general explanation of the foreign policy behavior
of major powers.

Second, the division of the sample into pre-World War I and post-World
War II observations leads to the clear finding that defensive realism operates
more consistently within different time periods. The threat variable is
associated with military spending and entering into alliances in the way
defensive realists would expect, prior to World War I. There is also some
limited evidence in terms of the number of conflicts per year, that defensive
realism is at work in the pre-World War I period. In the post-World War II
period, defensive realism also appears to be a better explanation for the
exiting of alliances by major powers. Although these cases of defensive
realism’s support are limited to specific time periods, there is far more
support in these cases for defensive realism than offensive realism. This
demonstrates the utility of utilizing constraints to distinguish between the
two realist theories.

Third, the inconsistent results between the pre-World War and the post-
World War time periods point to an important shift in the decision making
of major powers. Specifically, defensive realist considerations seem to be less
important in the post-World War II time period. Only in the case of alliance
exits is there evidence to suggest that defensive realism is a better
explanation after World War II. There is similarity between this finding and
that of Senese and Vasquez (2005) regarding conflict, further supporting the
assertion of Senese and Vasquez (2005) that one should decompose long time
series data sets into different periods of time. The analysis presented here
suggests that in the post-World War II period, major powers often behave in
the opposite manner than both defensive and offensive realists expect.

There are several possible explanations for the decreasing importance of

defensive realist decision making in the later time period. The inconsistency
Offensive Realism, Defensive Realism, and the Role of Constraints | 71

in the conflict results across the two time periods may be particularly
important (see Senese and Vasquez 2005). There are important economic
changes taking place in the international system across both periods,
suggesting that major powers may be more likely to behave in ways
suggested by liberal theories of decision making (e.g. Gartzke 2007; Oneal
and Russett 1997). Likewise, major powers are becoming increasingly (and
predominantly) democratic in the nearly two centuries after the Napoleonic
Wars, suggesting once again that liberal considerations may be at work (e.g.
Oneal and Russett 1997; Russett 1993). Of course, these are merely
suggestions at this stage, and without further theoretical and empirical
comparisons between these two realist models and liberal alternatives,
determining with certainty that liberal theories explain the inconsistencies
between realist expectations and the empirical findings across time is
impossible.

Conclusion

This paper attempts to contribute to the ongoing debate between

offensive and defensive realists (and other theoretical perspectives) by
applying both formal modeling and statistical methods. Through the use of
formal models, simple expectations regarding defensive and offensive realist
theory are derived, while still incorporating a number of realist strategies
within the same theoretical framework (Powell 1999, 213). Empirical tests of
these expectations reveal that although both theories receive empirical
support, only defensive realism receives some support across all three realist
strategies (military spending, alliances, and conflict), and only when specific
time periods (i.e. pre-World War I and post-World War II) are considered.
Although the period between the Napoleonic Wars and the World Wars
demonstrates that defensive realist motivations are at work in major power
decision making (although this finding is weakest in terms of conflict), after
World War II there is little evidence that either defensive or offensive realism
explain these major power strategies (except in the case of defensive realism
and the exiting of alliances). This article demonstrate a weakening of realist
explanations over time, as well as the utility of applying such formal
methods to realist debates.

Of course, there are a number of important questions left to be

answered, both because of the limited scope, and the results of this study.
First, the empirical tests address a major debate in realism, but do not
attempt to find empirical evidence comparing either of these realist theories
to possible alternative non-realist explanations (e.g. Gartzke 2007; Oneal and
72 | Rudloff

Russett 1997; Russett 1993). Second, the finding that the defensive realist
logic offers possible explanations for major power alliance entrance and
military spending decisions in the pre-World War I period, but less for the
post-World War II period, introduces a number of questions as to why this
change occurs. Although answering such a question is outside the goals of
this study, such findings indicate not only the importance of comparing
empirical results across time periods (Senese and Vasquez 2005), but the
further need for theoretical efforts to better explain such changes.
Offensive Realism, Defensive Realism, and the Role of Constraints | 73

References

Barbieri, Katherine, Omar M.G. Keshk, and Brian Pollins. 2009. “Trading Data:
Evaluating our Assumptions and Coding Rules,” Conflict Management and Peace
Science 26 (5): 471-491.

Barbieri, Katherine and Omar Keshk. 2012. “Correlates of War Project Trade Data Set
Codebook, Version 3.0.” Available at: http://www.correlatesofwar.org.
Accessed on: April 15, 2013.

Carr, Edward Hallett. 1964. The Twenty Years’ Crisis: 1919 - 1939. New York: Harper
and Row.

Chivers, C. J. 2008. “Russia Warns It May Back Breakaway Republics in Georgia,” The
New York Times, February 16, 2008, section A, page 3. Available through
LexisNexis Academic. Accessed on March 10, 2008.

Clarke, Kevin A. 2001. “Testing Nonnested Models of International Relations:

Reevaluating Realism,” American Journal of Political Science 45 (3): 724–744.

Cobb, Charles W. and Paul H. Douglas. 1928. “A Theory of Production,” The

American Economic Review 38 (Supplement): 139–165.

Correlates of War Project. 2005a. “National Military Capabilities Data

Documentation, Version 3.0.” Available at http://www.correlatesofwar.org.
Accessed on March 10, 2008. Accessed on: April 15, 2013.

Correlates of War Project. 2005b. “State System Membership List, v2004.1." Online,
http://correlatesofwar.org.

Correlates of War Project. 2010. “National Material Capability Data Documentation,

Version 4.0.” Available at: http://www.correlatesofwar.org.

Drury, A. Cooper and Yitan Li. 2006. “U.S. Economic Sanctions Threats Against
China: Failing to Leverage Better Human Rights,” Foreign Policy Analysis 2 (4):
307-324.

Elman, Colin. 2004. “Extending Offensive Realism: The Louisiana Purchase and
America’s Rise to Regional Hegemony,” American Political Science Review 98 (4):
563- 576.

Feng, Huiyun. 2005. “The Operational Code of Mao Zedong: Defensive or Offensive
Realist?” Security Studies 14 (4): 637–662.
74 | Rudloff

Finn, Peter. 2008. “Putin Threatens Ukraine on NATO; Russian Raises Issue of U.S.
Missile Shield,” The Washington Post, February 13, 2008, page A08. Available
through LexisNexis Academic. Accessed on March 10, 2008.

Gartzke, Erik. 2007. “The Capitalist Peace,” American Journal of Political Science 51 (1):
166–191.

Ghosn, Faten and Glenn Palmer. 2003. “Codebook for the Militarized Interstate
Dispute Data, Version 3.0.” Available at http://cow2.la.psu.edu. Accessed on
March 10, 2008.

Ghosn, Faten, Glenn Palmer, and Stuart Bremer. 2004. “The MID3 Data Set, 1993 -
2001: Procedures, Coding Rules, and Description,” Conflict Management and Peace
Science 21 (2): 133–154.

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): 211–222.

Gleditsch, Kristian D. 2002. “Expanded Trade and GDP Data,” Journal of Conflict
Resolution 46 (5): 712–724.

Greene, William H. 2004. Econometric Analysis. Fifth edition. Upper Saddle River, NJ:
Prentice Hall.

Heo, Uk and Robert J. Eger. 2005. “Paying for Security: The Security-Prosperity
Dilemma in the United States,” Journal of Conflict Resolution 49 (5): 792–817.

Kagan, Donald. 1995. On the Origins of War and the Preservation of Peace. New York:
Anchor Books.

James, Patrick. 1995. “Structural Realism and the Causes of War,” Mershon
International Studies Review 39 (2): 181–208.

Kreps, David M. 1990. A Course in Microeconomic Theory. Princeton, NJ: Princeton

University Press.

Labs, Eric J. 1997. “Beyond Victory: Offensive Realism and The Expansion of War
Aims,” Security Studies 6 (4): 1–49.

Lee, Gerald Geunwook. 2002/2003. “To Be Long or Not To Be Long - That is the
Question: The Contradiction of Time-Horizon in Offensive Realism,” Security
Studies 12 (2): 196–217.
Offensive Realism, Defensive Realism, and the Role of Constraints | 75

Lemke, Douglas. 2004. “Great Powers in the Post-Cold War World: A Power
Transition Perspective,” in Balance of Power: Theory and Practice in the 21st
Century, edited by T. V. Paul, James J. Wirtz, and Michel Fortmann. Stanford,
CA: Stanford University Press, 52–75.

Lobell, Steven E. 2002/2003. “War is Politics: Offensive Realism, Domestic Politics,

and Security Strategies,” Security Studies 12 (2): 165–195.

Long, J. Scott and Jeremy Freese. 2006. Regression Models for Categorical Dependent
Variables Using Stata. Second edition. College Station, TX: Stata Press.

Lynn-Jones, Sean M. 1998. “Realism and America’s Rise: A Review Essay,”

International Security 23 (2): 157–182.

Maddison, Angus. 1995. Monitoring the World Economy 1820–1992. Paris: OECD.

Maddison, Angus. 2001. The World Economy: A Millennial Perspective. Paris: OECD
Development Center.

Maddison, Angus. 2003. The World Economy: Historical Statistics. Paris: OECD
Development Center.

Maddison, Angus. 2009. “Statistics on World Population, GDP and Per Capita GDP,
1–2006 AD.” Available at: http://www.ggdc.net/maddison/. Accessed on:
October 25, 2009.

Marshall, Monty G., Keith Jaggers, and Ted Robert Gurr. 2011. “Polity IV Project:
Political Regime Characteristics and Transitions, 1800-2010, Data Users Manual.”
Available at: http://www.systemicpeace.org/inscr/p4manualv2010.pdf.
Accessed on: April 15, 2013.

Maslow, A. H. 1943. “A Theory of Human Motivation,” Psychological Review 50 (4):

370–396.

McGinnis, Michael D. 1994. “Rational Choice and Foreign Policy Change: The Arms
and Alignments or Regional Powers,” in Foreign Policy Restructuring: How
Governments Respond to Global Change, edited by Jerel A. Rosati, Joe D. Hagan,
and Martin W. Sampson III. Columbia, SC: University of South Carolina Press,
65- 87.

Mearsheimer, John J. 1994. “The False Promise of International Institutions,”

International Security 19 (3): 5–49.

Mearsheimer, John J. 2001. The Tragedy of Great Power Politics. New York: W. W.
Norton.
76 | Rudloff

Morgan, T. Clifton and Glenn Palmer. 2000. “A Model of Foreign Policy

Substitutability: Selecting the Right Tools for the Job(s),” Journal of Conflict
Resolution 44 (1): 11–32.

Montgomery, Evan Braden. 2006. “Breaking Out of the Security Dilemma: Realism,
Reassurance, and the Problems of Uncertainty,” International Security 31 (2): 151-
185.

Morgenthau, Hans J. 1978. Politics Among Nations: The Struggle for Power and Peace.
Fifth edition. New York: Knopf.

Most, Benjamin A. and Harvey Starr. 1989. Inquiry, Logic and International Politics.
Columbia, SC: University of South Carolina.

Officer, Lawrence H. “Dollar-Pound Exchange Rate From 1791.”

MeasuringWorth.com, 2008. Data accessed on March 8, 2008.

Oneal, John R. and Bruce M. Russett. 1997. “The Classical Liberals Were Right:
Democracy, Interdependence, and Conflict, 1950–1985,” International Studies
Quarterly 41 (2): 267–293.

Palmer, Glenn and T. Clifton Morgan. 2006. A Theory of Foreign Policy. Princeton, NJ:
Princeton University Press.

Palmer, Glenn and T. Clifton Morgan. 2007. “Power Transition, the Two-Good
Theory, and Neorealism: A Comparison with Comments on Recent U.S. Foreign
Policy,” International Interactions 33 (3): 329–346.

Powell, Robert. 1999. In the Shadow of Power: States and Strategies in International
Politics. Princeton, NJ: Princeton University Press.

Rose, Gideon. 1998. “Neoclassical Realism and Theories of Foreign Policy,” World
Politics 51 (1): 144–172.

Rudloff, Peter. 2007. Substituting Security: Economic Resources, Strategies Resources, and
the Security Policy Choices of Minor Powers. Dissertation, University of Illinois at
Urbana-Champaign.

Russett, Bruce. 1993. Grasping the Democratic Peace: Principles for a Post-Cold War World.
Princeton, NJ: Princeton University Press.

Sahr, Robert. 2009. “Conversion Factors in 2000 Dollars for 1800 to estimated 2016.”
Available at http://oregonstate.edu/cla/polisci/faculty-
research/sahr/sahr.htm. Accessed on October 25, 2009.
Offensive Realism, Defensive Realism, and the Role of Constraints | 77

Sarkees, Meredith Reid and Frank Whelon Wayman. 2010. Resort to War: A Data Guide
to Inter-state, Extra-state, Intra-state, and Non-state Wars, 1816-2006. Washington,
DC: CQ Press.

Senese, Paul D. and John A. Vasquez. 2005. “Assessing the Steps to War,” British
Journal of Political Science 35 (4): 607–633.

Simon, Carl P. and Lawrence Blume. 1994. Mathematics for Economists. New York: W.
W. Norton.

Singer, J. David, Stuart Bremer, and John Stuckey. 1972. “Capability Distribution,
Uncertainty, and Major Power War, 1820-1965,” in Peace, War, and Numbers.
Edited by Bruce Russett. Beverly Hills: Sage, 19–48.

Snyder, Glenn H. 2002. “Mearsheimer’s World - Offensive Realism and the Struggle
for Security: A Review Essay,” International Security 27 (1): 149–173.

Snyder, Jack. 1991. Myths of Empire: Domestic Politics and International Ambition. Ithaca,
NY: Cornell University Press.

Sydsaeter, Knut, Arne Strom, and Peter Berck. 2000. Economist’s Mathematical Manual.
Third edition. Berlin: Springer.

Tang, Shiping and Evan Braden Montgomery. 2007. “Uncertainty and Reassurance in
International Politics,” International Security 32 (1): 193-200.

Tuchman, Barbara W. 1962. The Guns of August. New York: Macmillan.

Valeriano, Brandon. 2009. “The Tragedy of Offensive Realism: Testing Aggressive

Power Politics Models,” International Interactions 35 (2): 179–206.

Waltz, Kenneth N. 1979. Theory of International Politics. New York: McGraw Hill.