"The Security Dilemma Revisited," World Politics, Vol.50 No.1 (Oct. 1997), pp.171-201.
"The Security Dilemma Revisited," World Politics, Vol.50 No.1 (Oct. 1997), pp.171-201.
"The Security Dilemma Revisited," World Politics, Vol.50 No.1 (Oct. 1997), pp.171-201.
http://journals.cambridge.org/WPO
Charles L. Glaser
* For comments on earlier drafts I would like to thank Lynn Eden, Colin Elman, Matt Evangelista,
Jim Fearon, Lloyd Gruber, Chaim Kaufmann, Andy Kydd, Joseph Lepgold, Jim Morrow, and Steve
Walt.
1
Robert Jervis, "Cooperation under the Security Dilemma," World Politics 30 (January 1978), 169.
For earlier discussions of the security dilemma, see John H. Here, "Idealist Internationalism and the
Security Dilemma," World Politics 2 (January 1950), which identifies the basic concept, but does not
develop it extensively, and Herbert Butterfield, History and Human Relations (London: Collins, 1951).
2
Robert Jervis, Perception and Misperception in International Politics (Princeton: Princeton University
Press, 1976), chap. 3, esp. 62-76; these pages provide a more thorough discussion of the basic workings
of the security dilemma than does "Cooperation under the Security Dilemma." For an early discussion
of this type of interaction, see J. David Singer, "Threat-Perception and the Armament-Tension
Dilemma," Journal ofConflict Resolution 2 (March 1958); for arecentdiscussion, see Charles L. Glaser,
"Political Consequences of Military Strategy: Expanding and Refining the Spiral and Deterrence
Models," World Politics 44 (July 1992).
3
Many similar arguments are developed by George Quester, Offense and Defense in the International
System (New York: Wiley, 1977). See also Marion William Boggs, Attempts to Define and Limit "Ag-
gressive"Armament and Strategy (Columbia: University of Missouri, 1941); and MarliesTer Borg, "Re-
ducing Offensive Capabilities: The Attempt of 1932," Journal of Peace Research 29, no. 2 (1992).
24
John J. Mearsheimer, Conventional Deterrence (Ithaca, N.Y.: Cornell University Press, 1983),
chap. 7.
25
This possibility is discussed in Chaim Kaufmann and Charles Glaser, "What Is the Offense-De-
fense Balance and Can We Measure It?" International Security (forthcoming).
26
See, for example, Thomas C. Schelling and Morton H. Halperin, Strategy andArms Control (New
York: Twentieth Century Fund, 1961), esp. 1-2.
SECURITY DILEMMA REVISITED 177
tition can arise. In addition, Jervis lays the groundwork for more recent
work by framing the problem of cooperation in terms of Prisoners'
Dilemma and Stag Hunt game-theory matrices and exploring how the
prospects for cooperation vary with changes in the relative size of coun-
tries' payoffs (pp. 170-83).27 In more recent years, cooperation theorists
have used game theory to establish a more rigorous foundation for the
corefindingsof arms control theory, they frequently use the repeated
Prisoners' Dilemma to model arms competition.28
MAKING ONE'S ADVERSARY LESS SECURE, W H I C H INCREASES THE
VALUE I T PLACES ON EXPANSION
3S
On signaling, see Robert Jems, The Logic ofImages in International Relations (Princeton: Prince-
ton University Press, 1970); and James D. Fearon, "Threats to Use Force: The Role of Costly Signals
in International Crises" (Ph.D. diss., University of California, Berkeley, 1992).
180 WORLD POLITICS
forces that improve its offense.36 The greedy state sees both greater
value in offense and sometimes less value in not provoking others, since
it anticipates conflict anyway.
The question that remains is, can an adversary that makes logically
sound inferences nevertheless reach incorrect conclusions about the
state's goals?37 For the interaction between pure security seekers to pro-
duce the increasing fear and insecurity explained by a rational spiral
model, the answer must be yes. Otherwise, misperceptions would be re-
quired to generate a spiral.
This type of interaction can occur if the states are uncertain about
the size or type of forces required to maintain a given level of security.
Given this uncertainty, consider the simple case in which some pure se-
curity seekers would be satisfied with a lower level of forces and some
with a higher level, but all greedy states would require a higher level. A
pure security seeker that builds to the higher level will then convince its
adversary that it is more likely to be greedy, since only some security
seekers but all greedy states would build to this level. Similar interac-
tions can occur if there is uncertainty about whether security seekers re-
quire offensive capabilities and if there is uncertainty about the level of
security that security seekers believe is adequate.38
A spiral can also result from an adversary's uncertainty about the
state's understanding of its motives. For example, when the adversary
believes that the state believes there is only a small probability that the
adversary is greedy and, therefore, that the state does not fear it, the ad-
versary will conclude that the state's buildup is largely unnecessary for
security and therefore that the state is likely to be motivated by greed. If
the adversary's initial estimate of the probability that the state is fearful
is too low, then this increase in its assessment that the state is greedy
will be too large, resulting in a spiral. Kydd's formal analysis of this in-
teraction shows that under a wide range of conditions a rational adver-
sary will find the state's buildup to be provocative and that updating of
beliefs is sensitive to prior expectations about the state's motives.39
36
This is an oversimplification, however, since a pure security seeker might want the capability to
take territory for a variety of reasons. See fn. 57 below for qualifications.
37
Although Jervis identifies the types of buildups, he does not really explain how rational states can
spiral; Jervis (fn. 1), 199—201. He notes that states often cannot accurately infer motives from others'
military forces and therefore they tend to assume the worst; Jervis (fn. 2), 64-65. This would clearly
generate increased hostility, but the judgment is logically flawed, since without additional information
states' assessments should remain unchanged. Moreover, states should not assume the worst in the face
of uncertainty, since this can support policies that are too competitive/provocative.
38
For discussion of related issues concerning subjective security requirements, see Jervis (fn. 1),
174-76.
39
Andrew Kydd, "Game Theory and the Spiral Model," World Politics 49 (April 1997). Kydd notes
(p. 373) that Jervis explores this dynamic for the extreme cases in which bias leads the adversary to
SECURITY DILEMMA REVISITED 181
The logic of these signaling arguments also works in the opposite di-
rection—a state can sometimes use restraint in building military forces
to reduce the adversary's concern about its greediness.40 A greedy state
wants to mislead its adversaries into believing that it is interested only
in security, since its adversaries would then be more likely to pursue
policies that leave them vulnerable. Given these incentives for a greedy
state to misrepresent its motives, a security seeker can communicate in-
formation about its motives only by adopting a policy that would be
less costly for it than for a greedy state.
Depending on the conditions they face, states can try to communi-
cate their benign intentions via three types of military policies.41 First,
arms control agreements that limit both countries' current or future abil-
ity to perform offensive missions communicate a lack of greed, since a
greedy state sees greater value in offensive missions than does a pure secu-
rity seeker. Second, a state may be able to adopt unilateral defense, choos-
ing to protect its country with a defensive doctrine, even if its adversary
continues to pursue an offensive one. When offense has the advantage,
maintaining its security via unilateral defense will require the state to
outspend its adversary, which reinforces the message that its motives are
benign. Finally, a state can exercise unilateral restraint, that is, reduce its
military capability below what it would choose for adequate deterrence
and defense were it not considering the effects of signaling. One use of
unilateral restraint is primarily tactical—to set in motion a process of
reciprocated restraint.42 Even if not reciprocated, however, unilateral re-
straint can succeed by communicating that the state is not greedy and is
committed to improving relations. Thus, although the state's ability to
defend is reduced, the net effect can be an increase in its security. There
is, however, the danger that the adversary will misinterpret the state's re-
straint, seeing a lack of resolve instead of a lack greed; in this case, re-
straint encourages the adversary to challenge the state. The dual dangers
of military shortfalls and underestimates of their resolve make states re-
luctant to pursue ambitious policies of unilateral restraint.
assume the state is definitely secure. For pure security seekers, however, all that is necessary to gener-
ate a spiral is for the adversary to believe there is some possibility that the state is secure. See also
George W. Downs and David M. Rocke, Tacit Bargaining, Arms Races, and Arms Control (Ann Arbor
University of Michigan Press, 1990), chap. 4.
* Kydd (fn. 39) provides a formal treatment.
41
Glaser (fh. 5), 67-70; and idem (fh. 2), 526-33; for related points, see Downs, Rocke, and Siver-
son (fh. 28). On reassurance more generally, see Stein (fn. 6); on the inhibiting effect of appearing
weak, see James D. Morrow, "Signaling Difficulties with Linkage in Crisis Bargaining," International
Studies Quarterly 36 (June 1996).
"This approach, often referred to as GRIT, was developed by Charles E. Osgood, An Alternative to
War or Surrender (Urbana: University of Illinois, 1962).
182 WORLD POLITICS
MISPERCEPTIONS
In contrast to the rational updating we have considered so far, Jervis
also discusses the role of psychological biases in contributing to states'
overly hostile assessments of others' actions.43 More recent work has fo-
cused on bureaucratic and domestic political processes as alternative
sources of bias.
Statesmen who do not understand the security dilemma and there-
fore do a poor job of appreciating the choices faced by their adversary
will infer incorrectly that the adversary's buildup reflects greedy mo-
tives. A state is likely to make errors that build on each other: a key ini-
tial mistake is for a state to assume that others know it is interested only
in security; the state is then likely to assume that others will not be
threatened by its buildup. Consequently, the state is inclined to see the
adversary's arms buildup as a sign of greed, when in fact the adversary
is building in response to the state's buildup. This is a distorted form of
the rational spiral, described above, which is driven entirely by uncer-
tainty about motives; when this bias prevails, states will be more inse-
cure and competition will be more intense than is predicted by a
rational security dilemma.
Attribution theory offers a psychological explanation for this type of
flawed reasoning.44 Leaders commonly make the mistake of interpret-
ing the behavior of other countries in terms of their goals/motives, even
though they understand their own behavior differently, in terms of the
situation they face. As a result, leaders fail to appreciate that other
countries face a security dilemma and therefore interpret the actions of
others as reflecting greed, even though a pure security seeker might
have acted the same way.
Recent work on misperceptions provides alternative explanations, lo-
cating these analyticflawsat the level of the state instead of at the level
of the individual. This work uses theories of organizational behavior
and domestic political dynamics to explain why states often exaggerate
an adversary's hostility.45 Militaries are inclined to exaggerate the of-
fensive potential of the adversary's forces and to impute malign inten-
tions, even when the purposes of the adversary's forces are ambiguous.
43
Misperception receives far less discussion in "Cooperation under the Security Dilemma" (but see
pp. 181-83) than in Jervis (fh. 2), 67-76.
44
On attribution theory, see Jervis (fn. 2), esp. 35—48; Deborah Welch Larson, Origins of Contain-
ment: A Psychological Explanation (Princeton: Princeton University Press, 1985), esp. 34-42; and
Jonathan Mercer, Refutation and International Politics (Ithaca, N.Y.: Cornell University Press, 1996).
45
For an emphasis on organizational perspectives, see Stephen Van Evera, "Causes of War" (Ph.D.
diss., University of California, Berkeley, 1984), pt. 2; and Snyder (fn. 10). Snyder (fn. 11) emphasizes
domestic political dynamics.
SECURITY DILEMMA REVISITED 183
Powerful interest groups that would benefit from military competition
or expansion often advance self-serving strategic arguments that exag-
gerate the threat posed by the adversary's capabilities or motives. Fur-
thermore, states are inclined to create myths about the unthreatening
nature of their own behavior, which makes adversaries appear still more
dangerous.46
In sum, through rational updating a state's military buildup (or re-
straint) can lead an adversary to alter its assessment of the state's mo-
tives. Research on misperceptions cautions that states are inclined to do
a poor job of updating, with a bias toward exaggerating the hostility of
others. When they suffer from these biases, states will act as though the
security dilemma is more severe than it actually is.
SIMPLY WASTES M O N E Y
The third way in which efforts to make oneself more secure can be self-
defeating is by simply wasting money. Unlike the first two cases, the
state would not end up less secure for having pursued a policy that ini-
tially made its adversary less secure. In this case, the action-reaction
process does not reduce the state's military capability; as we have seen,
action-reaction processes could leave the state's ability to defend itself
unchanged. Nor does the action-reaction generate a spiral of increas-
ingly negative views of the adversary's motives, which is possible since
not all military buildups provide information about motives. Neverthe-
less, the security dilemma results in self-defeating efforts that leave the
state less prosperous, yet no more secure.47
* Van Evera (fn. 45), chap. 8; and idem, "Why States Believe Foolish Ideas: Non-Self-Evaluation
by Government and Society" (Paper presented at the annual meeting of the American Political Sci-
ence Association, Washington, D.C., 1988).
47
Snyder (fn. 33), 461, emphasizes wasted resources; he argues (p. 462) that alliance formation is
similar to arming, in that all states would be better off remaining outside an alliance; action and reac-
tion nevertheless generate alliance blocks that are costly but fail to increase security.
48
On Stag Hunt and related games, see Kenneth A. Oye, "Explaining Cooperation under Anar-
chy: Hypotheses and Strategies," in Oye (fn. 28).
184 WORLD POLITICS
The core of the answer lies in appreciating that states are uncertain
about their adversaries' motives, lacking confidence that others are pure
security seekers. Uncertainty about motives means that states cannot be
confident that they are in a Stag Hunt, even if they are. This uncer-
tainty about the type of game can make competition/arming the states
best option. Jervis takes a different view, identifying different sources of
competition when in a Stag Hunt—misperceptions and irrational be-
havior (p. 168). Important as misperceptions and irrational behavior
can be, however, these are additions to the basic story.49 From the per-
spective of structural realism and the emphasis it places on the implica-
tions of anarchy, uncertainty about the adversary—both its motives and
its understanding of one's own motives—is the fundamental ingredient
for understanding competition.50 This subsection describes a simple
game-theory model of how uncertainty about the adversary could re-
duce the prospects for productive cooperation.
It is useful to begin by considering why two security seekers might
be in a Stag Hunt.51 On the one hand, a pure security seeker that is
confident of not being attacked has no reason to expand: it is satisfied
with the territory it controls and does not see instrumental value in ad-
ditional territory, because it does not fear attack. It could prefer the ter-
ritorial status quo (CC) to unopposed expansion (DC) for a variety of
reasons, including how costly it is to govern the additional territory or
because expansion violates an international norm that the state values.
Because the state values its own territory, it could prefer fighting to pro-
tect its territory (DD) to allowing its adversary to expand unopposed
(CD). If two pure security seekers have these preferences and this is
common knowledge, then they are in a Stag Hunt and both countries
should optimally choose to cooperate.
On the other hand, a state that is unsure whether its adversary is a
pure security seeker faces a very different situation. A greedy adversary
will prefer unopposed expansion to the territorial status quo; it has
Prisoners' Dilemma preferences. Even if the state's own preference or-
dering remains unchanged,52 its preferred option in light of this adver-
sary's expected behavior is competition.
49
For discussion of their impact on a Stag Hunt, see Downs, Rocke, and Siverson (fn. 28), 134-37.
50
In addition, a type of uncertainty that is not explored in the text is also important—uncertainty
about whether an adversary will become greedier. This uncertainty cannot be eliminated because lead-
ers cannot bind themselves and their successors to current goals; see Jervis (fn. 1), 168.
51
Although my discussion focuses on competition over territory, a parallel analysis can be devel-
oped for arms competition.
52
If facing a greedy adversary, however, a pure security seeker would now see instrumental value
in expansion, if this would increase its security. The state would then have Prisoners' Dilemma
SECURITY DILEMMA REVISITED 185
If the state is unsure about the type of adversary it faces, then it is
uncertain about which game it is in. The state should therefore weigh
its payoffs for cooperation and competition by its estimate of whether
the adversary is greedy or a pure security seeker.53 Competition is more
attractive the higher the state's estimate that the adversary is greedy.
54
For related analysis, see Malcolm W. Hoag, "On Stability in Deterrent Races," World Politics 13
(July 1961).
s5
On the relationship between first-strike advantages, preemption, and accidents, see Schelling and
Halperin (fn. 26), 14-16.
56
See also Quester (fh. 3), 105-6.
57
As Jervis (fn. 1) notes, this is an overstatement (pp. 201—2). A pure security seeker might buy of-
fense for a number of reasons: (1) if offense has a great advantage over defense; (2) if the state has ex-
tended deterrence commitments; (3) because offense may be necessary to regain territory lost at the
beginning of a war; and (4) because the threat of counterofFense can enhance deterrence. On extended
deterrence, see Stephen W. Van Evera, "Offense, Defense and Strategy: When Is Offense Best?"
(Paper presented at the annual meeting of the American Political Science Association, Chicago, Sep-
tember 1987). On counteroffense, see Samuel R Huntington, "Convention Deterrence and Conven-
tional Retaliation in Europe," International Security 8 (Winter 1983-84); and Barry R. Posen, "Crisis
Stability and Conventional Arms Control," Daedulus 120 (Winter 1991).
58
The concept of differentiation is implicit in the distinction between qualitative and quantitative
arms control, with the former relying on differentiation. This concept lies at the core of modern arms
control theory; see, for example, Thomas C. Schelling, "What Went Wrong with Arms Control," For-
eign Affairs 64 (Winter 1985-86).
SECURITY DILEMMA REVISITED 187
In contrast, when offense and defense are not differentiated, a coun-
try buying forces to protect its territory can only choose forces that re-
duce its adversary's ability to defend. Signaling becomes much harder
and riskier, because pure security seekers and greedy states will buy the
same types of forces.
This stock of hypotheses has been supplemented by Van Evera, who
adds hypotheses on how offense advantage fuels preventive war and en-
courages styles of diplomacy that increase the probability of war. Of-
fense advantage heightens the significance of shifts in power, which
increases incentives for preventive war. States are more likely to use fait
accompli tactics when offense has the advantage because winning dis-
putes is more important when security is scarce, which encourages
states to overlook negotiable compromises. States negotiate less and
less successfully, because offense advantage makes it more important
that agreements be carefully balanced, which makes negotiations more
difficult, and because offense advantage increases the advantages of vi-
olating agreements, which makes agreements riskier. In addition, of-
fense advantage makes states more secretive, since information about
military forces can increase their vulnerability. Secrecy in turn increases
the probability of war by fueling miscalculations of military capabilities
and of states' interests.59
The deductive strength of this body of offense-defense hypotheses
has gone largely unchallenged, but recent work has questioned the rela-
tionship between offense advantage and the frequency of war. Because
the risks of war could be greater for the attacker when offense has the
advantage, potential attackers should face countervailing pressures that
make them more cautious, especially when considering large wars,
which could sometimes make war less likely.60
Van Evera has performed the most extensive tests of offense-defense
hypotheses, although even these are preliminary.61 Focusing on Europe
since 1798, he finds strong support for the theory's basic hypothesis:
war is more likely when offense has (or is perceived to have) the advan-
tage. Examining the First World War in detail, Van Evera finds sup-
" Van Evera (fh. 5), chap. 5.
40
James D. Fearon, "Rationalist Explanations for War," International Organization 49 (Summer
1995), 402-3. And see the following papers prepared for the annual meeting of the International Stud-
ies Association, Chicago, February 1995: Fearon, "The Offense-Defense Balance and War since 1648";
ind Charles Glaser and Chaim Kaufmann, "Inspecting the Foundations of Offense-Defense Theory:
Can They Bear the Weight?"
" Van Evera (fh. 5), chaps. 5,6; and idem, "The Cult of the Offensive and the Origins of the First
World War," International Security 9 (Summer 1984). See also Ted Hopf, "Polarity, the Offense-De-
fense Balance, and War," American Political Science Review 85 (June 1991); and Fearon (fn. 60, Febru-
ary 1995), who argues that the frequency of war in broad historical periods runs counter to standard
offense-defense predictions.
188 WORLD POLITICS
port for the broad range of hypotheses about the dangers generated by
offense advantage.62 Many of these findings about the First World War
have been challenged by Trachtenberg, who is especially critical of as-
sertions about the role of offense dominance in fueling German expan-
sion, and also casts doubt on the importance of first-strike and
preventive incentives in bringing about the war.63
73
Restraint may not be ill-advised, however, since for a greedy insecure adversary the reduction in its
insecurity could still offset the reduction in the state's military capabilities. Assessing the net effect re-
quires more specific assumptions and a detailed model of interaction. By contrast, unilateral concessions
are always misguided when one is facing a secure greedy state, as described by the deterrence model.
74
On the deterrent value of counteroffensive capabilities, see fn. 57.
192 WORLD POLITICS
rity concerns. If the adversary were certain that it faced a pure security
seeker, then the state would face a greatly reduced security dilemma.75
The adversary would understand that military buildups reflected the
state's insecurity rather than its greed. And military advantages would be
far less threatening, since the adversary would know that the state did not
want to attack. As a result, in contrast to the predictions of the security
dilemma, arming and gaining military advantages would usually increase
the state's security, not reduce it, and at worst would waste money.
A key assumption of structural realism guarantees, in combination
with the security dilemma, that states will be at least somewhat uncer-
tain about others' motives. The theory posits a world in which states do
not rely on the internal characteristics of other states—for example,
their type of political or economic system—to divine their motives. In-
stead, states interpret their adversaries' actions to infer their motives.
The security dilemma, however, can prevent adversaries from acting in
ways that would entirely clarify their motives. As discussed above, when
states face a security dilemma, many of the policies that would provide
necessary military capabilities will provide ambiguous information
about their motives. For example, when offense and defense are not en-
tirely differentiated, the military policies adopted by a pure security
seeker might also be taken by an adversary that was motivated partly by
greed; and, when offense has a large advantage, pure security seekers
may be unable to afford defensive forces and strategies. Under these
conditions, reducing uncertainty would require states to deploy forces
that are militarily inadequate, since this is their only option for signal-
ing benign motives. States will often conclude that these policies are
more dangerous than alternatives that do not reduce uncertainty about
their motives but that do provide better military capabilities.
If, however, states can rely on sources of information beyond those
that structural realism allows, they may be able to reduce uncertainty
further and thereby mitigate the security dilemma. The result can be a
shift to more cooperative policies. Consider the argument that democ-
racies are believed not to have greedy motives. If this were the case,
then a military buildup launched by a democracy would be less alarm-
ing to its adversaries than a similar buildup launched by an authoritar-
7S
Although at first glance this might seem to eliminate the security dilemma, this need not be the
case. A state motivated entirely by security might choose war to increase its security; consequently, the
state's insecurity should lead the adversary to fear it. Thus, eliminating the security dilemma would re-
quire that the adversary know not only that the state was a pure security seeker but also that the state
did not fear it. Consequently, some points in this paragraph are overstated.
SECURITY DILEMMA REVISITED 193
76
ian regime. As a result, the democracy faces a less severe security
dilemma;77 and interactions between democracies could result in a de-
mocratic peace instead of intense competition, even when structural
conditions create a severe security dilemma.
Unit-level information that mitigates the security dilemma can also
support more competitive policies. This occurs when a state remains
uncertain of an adversary's motives but believes unit-level information
enables its adversary to appreciate that it is a pure security seeker. Com-
petitive policies now become more desirable, since they do not suffer
from a key shortcoming identified by the security dilemma; that is, they
do not signal malign motives.
This line of argument plays a central role in the "deterrence model,"
which rejects the security dilemma completely, albeit implicitly, by as-
suming that the adversary knows the state is a pure security seeker.78
Combining this with the assumption that the adversary is greedy, the
deterrence model calls for highly competitive policies and warns against
the dangers of restraint and concessions. For example, in describing the
cold war competition between the United States and the Soviet Union,
the deterrence model held that the Soviets were bent on expansion for
entirely greedy reasons and knew that they had nothing to fear from the
United States.
In short, examining a couple of key nonstructural variables high-
lights the fact that the role of the security dilemma depends on certain
basic theoretical assumptions. Structural realism is built on assumptions
that guarantee a central role for the security dilemma. If, however, we
start from different assumptions about states' motives or the sources of
information about state's motives, the importance and severity of the
security dilemma can decrease.
76
In this spirit, see Bruce Bueno de Mesquita and David Lalman, War and Reason: Domestic and In-
ternational Imperatives (New Haven: Yale University Press, 1992), chap. 5. A different argument is that
the openness that characterizes mature democracies results in domestic debate that cannot be manip-
ulated simply to deceive an adversary, and therefore provides valuable information about motives. See
Andrew Kydd, "Signaling and Structural Realism" (Manuscript 1996); and Kenneth A. Schultz, "Do-
mestic Political Competition and Bargaining in International Crises" (Ph.D. diss., Stanford University,
1996). Schultz develops this type of argument for crisis interactions.
77
Arguing along these lines is James D. Fearon, "Domestic Political Audiences and the Escalation
of International Disputes," American Political Science Review 88 (September 1994), 587. If, however, a
democracy believes this to be the case and it is not, then we get the kind of misperception described
above: the democracy believes that it doe not face a security dilemma and therefore does not moderate
its building; its adversary, not confident that the democracy is a pure security seeker, then responds to
the buildup; and the democracy then increases its estimate of the adversary's greediness because it be-
lieves that its adversary knew that a response was unnecessary for maintaining its security.
78
See Jervis (fn. 2), chap. 3; and Glaser (fn. 2).
194 WORLD POLITICS
CRITICISMS O F T H E SECURITY D I L E M M A AND
OFFENSE-DEFENSE THEORY
86
John J. Mearsheimer, "The False Promise of International Institutions," International Security 19
(Winter 1994-95), 11-12. Other prominent structural realists reject this position; see Waltz (fn. 4,
1979), 118,126,127.
87
John J. Mearsheimer, "Back to the Future: Instability in Europe after the Cold War," International
Security 15 (Summer 1990), 12. See also Fareed Zakaria, "Realism and Domestic Politics: A Review
Essay," International Security 17 (Summer 1992), 190-96.
88
Alexander Wendt, "Anarchy Is What States Make of It: The Social Construction of Power Poli-
tics," International Organization 46 (Spring 1992), quote at 407, and also 401-2; and idem, "Con-
structing International Politics," International Security 20 (Summer 1995), 77. For a very different
constructivist critique, which focuses on the offense-defense balance, instead of the security dilemma,
see Kier (fn. 10).
89
Wendt (fn. 88,1995), 73.
SECURITY DILEMMA REVISITED 197
security dilemma is created by states' interactions, states can choose
policies—for example, reassurance—that will avoid creating it. Accord-
ing to Wendt, realists overlook these possibilities for avoiding security
dilemmas because they conceive of states' interests and security in ways
that guarantee competition. For example, they assume states identify
negatively with others' security and should act on the basis of worst-
case assumptions. Because his objective is to clarify the logic of anar-
chy, Wendt frames his critique in terms of third-image structural
theory; thus, his disagreement with structural realists is not rooted in
disagreements over which level of analysis is most important.90
This disagreement between Wendt and realists may appear to be
simply one of terminology: Wendt is using "security dilemma" to de-
scribe the results of states' interaction, whereas Jervis and the literature
he has spawned use "security dilemma" to refer to a situation created by
the material conditions facing states, such as geography and prevailing
technology. By redefining well-established terminology, then, Wendt
has created confusion.
More important are the significant substantive disagreements. First,
Wendt exaggerates the extent to which structural realism calls for com-
petitive policies and, therefore, the extent to which it leads to security
dilemmas, as he defines them. As discussed above, although offensive
realists believe the international system requires states to pursue com-
petitive policies, the structural realists who place the greatest impor-
tance on the security dilemma—defensive/contingent realists—believe
that under a range of conditions states should cooperate. Part of the
problem is that the assumptions that Wendt assigns to realism are
themselves either controversial or logically flawed. Instead of seeing
states identifying negatively with the security of others, realists envision
states as egoists and therefore as indifferent to others' security, except as
it directly influences their own security.91 The security dilemma then
explains why egoists might see a positive interaction; that is, all else
being equal, increases in other states' security increase one's own secu-
rity. Wendt's claim that realists believe states should base their policies
on worst-case assumptions is also wrong (even though this position is
commonly attributed to realists and some offensive realists might try to
defend it). The core logic of the security dilemma makes clear that
90
Wendt (fn. 88,1992), 396; see also idem (fn. 88,1995), 72.
" This said, some realists have argued otherwise. See, for example, Joseph M. Grieco, Cooperation
among Nations (Ithaca, N.Y.: Cornell University Press, 1990). This is because they have conflated
means and ends. See Charles L. Glaser, "Correspondence: When Cumulative Relative Gains Matter,"
International Security 21 (Spring 1997). See also Robert Powell, "Guns, Butter, and Anarchy," Ameri-
can Political Science Review 87 (March 1993), 127.
198 WORLD POLITICS
worst-case analysis can be self-defeating, since forces that are larger or
more threatening than necessary can lead to reduced military capabili-
ties and/or negative spirals in beliefs about adversaries' motives. In fact,
Wendt s description of how states facing uncertainty about the motives
of others should interacts—his alternative to the realist prescription—is
essentially the signaling behavior envisioned by realists who emphasize
the role of the security dilemma.92
Second, in arguing that states can change "the intersubjective knowl-
edge that constitutes the system" by changing their practices, Wendt
implicitly rejects the existence of (standard/material) security dilem-
mas.93 The problem is that a security dilemma can sometimes make it
too risky for states to signal their benign motives. (As I have stressed,
however, this is not always true, since the risks depend on the severity
of the security dilemma, which varies.) Because Wendt believes the se-
curity dilemma is the creation of states, he overlooks the constraints it
can impose and consequently is too optimistic about the general ability
of states to change their practices and use signaling to avoid competi-
tive relations.94 As I have already explained, states may be able to over-
come these constraints by relying on unit-level knowledge of others'
motives. This, however, is not what Wendt is arguing, since his critique
is cast in terms of structural/third-image analysis.
CRITIQUE 3: OFFENSE-DEFENSE THEORY IS FLAWED
RESEARCH AGENDA
This article should leave little doubt that "Cooperation under the Se-
curity Dilemma" has both established the importance of key arguments
about the security dilemma and offense-defense theory and spawned a
large and impressive literature. What work remains to be done? I sug-
gest beginning at least with the following areas.
GREED VERSUS SECURITY
From a variety of angles, we have seen that the security dilemma and
offense-defense arguments are less important and insightful when
greedy states, especially secure greedy states, are at the heart of interna-
tional conflict. Research that helps determine the frequency and inten-
sity of greedy states would therefore be quite valuable. In addition,
concern about greedy states poses a basic challenge for theory building:
a reasonable goal would be to develop theories that deal adequately
with the mixture and varying intensity of states' motives—greed as well
as security.
EMPIRICAL TESTING OF OFFENSE-DEFENSE HYPOTHESES
Given their importance, the full range of offense-defense hypotheses
warrants further empirical testing. Valuable contributions can be made
by research (1) that examines whether states assess their security in
terms of power or military capability, (2) that explores how the balance
should be measured, and (3) that systematically applies these measures
to specific cases.
CAUTION VERSUS COMPETITION
As further empirical testing proceeds, the time is ripe to explore more
thoroughly the deductive strength of offense-defense hypotheses. Why
do countervailing considerations not moderate the impact of offense
SECURITY DILEMMA REVISITED 201
advantage? For example, if offense advantage means that attackers as
well as defenders could lose big in a major war, then why does offense
advantage not sometimes make states more cautious and war less
likely?100 If arms races become more dangerous, then why do states not
take greater risks to stop them? And so on.
RATIONAL VERSUS BIASED STATES
The explanatory and predictive value of the security dilemma depends
on the extent to which states suffer from psychological, bureaucratic,
and political biases. Maybe most important, the theory's optimistic pre-
diction that relations will be cooperative when the security dilemma is
mild depends on states accurately perceiving the conditions they face. If
states are inclined to exaggerate the advantages of offense, to ignore
that others face a security dilemma, or to overlook others' restraint,
then the opportunity created by objective conditions will be squan-
dered. Although a substantial amount of research has focused on mis-
perceptions, answers to critical questions remain wide open. How
important have misperceptions been in fueling competition and war?
Are states likely to be better at avoidingflawedpolicies in the future?
Given these topics, the already large body of work that builds on
Jervis's security-dilemma and offense-defense arguments is likely to
continue to grow in size and importance.
100
See fn. 60 for work that has already raised this possibility.