http://jcr.sagepub.

com/

Journal of Conflict Resolution

http://jcr.sagepub.com/content/54/2/354
The online version of this article can be found at:

DOI: 10.1177/0022002709355414
2009
2010 54: 354 originally published online 22 December Journal of Conflict Resolution
Daniel G. Arce and Todd Sandler
Terrorist Spectaculars: Backlash Attacks and the Focus of Intelligence

Published by:
http://www.sagepublications.com
On behalf of:

Peace Science Society (International)

can be found at: Journal of Conflict Resolution Additional services and information for

http://jcr.sagepub.com/cgi/alerts Email Alerts:

http://jcr.sagepub.com/subscriptions Subscriptions:

http://www.sagepub.com/journalsReprints.nav Reprints:

http://www.sagepub.com/journalsPermissions.nav Permissions:

http://jcr.sagepub.com/content/54/2/354.refs.html Citations:

at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
Terrorist Spectaculars:
Backlash Attacks and the
Focus of Intelligence
Daniel G. Arce
1
and Todd Sandler
1
Abstract
This article presents a signaling model of terrorist attacks, where the target govern-
ment faces a trade-off from its counterterrorism responses and the backlash (coun-
terreaction) that such responses incite. An endogenous characterization of terrorist
spectaculars is specified, given a governments counterterrorism stance and the
potential for backlash attacks. In particular, spectacular attacks are pooling, rather
than separating, phenomena, whereby the government cannot discern, based on
past attacks, the militancy of the terrorist group. The definition for spectacular ter-
rorist attacks is inversely related to the governments toughness and its belief that
it confronts a militant group. Policy recommendations are specified for non-event-
specific intelligence in relation to the avoidance of spectacular attacks or unnecessary
concessions. Intelligence must be focused on the propensity for counterterrorism to
give rise to a backlash attack.
Keywords
signaling games, terrorist spectaculars, value of intelligence, backlash attacks,
counterterrorism
The unprecedented suicide skyjackings of September 11, 2001 (henceforth, 9/11), dem-
onstrated that large-scale terrorist eventsknown as spectacularscan result in
huge human and monetary losses. Other spectacular terrorist incidents include the sui-
cide truck bombing of the U.S. Marine barracks in Beirut, Lebanon (October 23, 1983),
the downing of Air India flight 182 over the Atlantic Ocean (June 23, 1985), the down-
ing of Pan Am Flight 103 over Lockerbie, Scotland (December 21, 1988), the
1
School of Economic, Political and Policy Sciences, University of Texas at Dallas, Richardson
Corresponding Author:
Todd Sandler, School of Economic, Political and Policy Sciences, University of Texas at Dallas, 800 W.
Campbell Road, Richardson, TX 75080-3021
Email: tsandler@utdallas.edu
Journal of Conflict Resolution
54(2) 354373
The Author(s) 2010
Reprints and permission: http://www.
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0022002709355414
http://jcr.sagepub.com
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
simultaneous bombings of the U.S. embassies in Kenya and Tanzania (August 7, 1998),
and the bombing of a Bali nightclub (October 12, 2002). These spectacular attacks, and
others not mentioned, may be perpetrated by terrorists whose demands may or may not
be feasibly met by a target government. Hezbollah suspended its suicide bombings for
a time after the United States and other participants in the Multinational Force left Bei-
rut in 1983. In contrast, al-Qaida, responsible for the African embassy bombings and 9/
11, did not end its attacks after the United States partially conceded and closed its mil-
itary bases in Saudi Arabia.
Counterterrorism requires a judicious mix of defensive and proactive measures,
including the gathering of intelligence. Typically, the latter is understood to involve
information to prevent specific terrorist events through the use of signal interception,
group infiltration, or imagery. We, instead, view the value of intelligence in terms of
a signaling game of incomplete information, in which a target government is ill
informed about the nature of the terrorist group. Since 9/11, there has been an empha-
sis on understanding the enemy and the lengths to which it will go to achieve its
objectives (Scheuer 2006; Wright 2006). For example, al-Qaida has not folded to tre-
mendous proactive measures; instead, al-Qaida has decentralized with a regional
emphasis. Moreover, attacks on the Madrid commuter trains and stations on March
11, 2004, and on the London transport system on July 7, 2005, illustrate that offensive
measures against al-Qaida may incite backlash attacks by sympathetic terrorists.
The primary purpose of this article is to indicate how a target government should
tailor its counterterrorism response to an unknown terrorist threat. This tailoring
involves not only the amount, if any, of concessions but also the level of counterter-
rorism actions. Unlike earlier signaling models of terrorism (Arce and Sandler 2007;
Lapan and Sandler 1993; Overgaard 1994), government countermeasures may pro-
voke a backlash attack. Another innovation here is to permit partial concessions in
place of an all-or-nothing concessionary response. A secondary purpose is to offer
an endogenous characterization of terrorist spectaculars, based on a governments
counterterrorism stance and the likelihood for backlash attacks. Spectacular incidents
arise from a pooling equilibrium, whereby the government cannot distinguish the mil-
itant orientation of a terrorist group.
In the two earliest signaling models of terrorism (Lapan and Sandler 1993;
Overgaard 1994), the government has incomplete information about the resources
available to the terrorist group. Terrorists are, however, initially endowed with suf-
ficient resources to send a signal corresponding to a spectacular attack, with the
definition of spectacular given exogenously (in terms of the resource require-
ments). The main difference between these two models is that terrorists directly
benefit from violence in LapanSandler, whereas violence is payoff decreasing
as it removes resources from nonviolent political activity in Overgaard. Subsequent
to 9/11, a categorization of terrorist types has emerged in which incomplete infor-
mation stems from the differences in preferences rather than resources that differ-
entiate the LapanSandler and Overgaard models. For example, Hoffman and
McCormick (2004) distinguish absolute groups that would rather fight than com-
promise from political groups that would rather compromise than battle. Franck
Arce and Sandler 355
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
and Melese (2004) make a distinction between organizations that have chosen
terrorism as a means to political ends and fanatical groups that have a predilec-
tion for violence. In their typology, political terrorists primarily use violence to
attract attention or communicate demands. By contrast, fanatics are more intent
on inflicting damage than on sending political communication. In Abramss
(2006) typology, terrorists with limited objectives employ violence to win political
concessions, whereas maximalist terrorists have demands over beliefs, values, and
ideology that are more difficult for the target government to compromise over.
Finally, Arce and Sandler (2007) and Sandler and Arce (2007) explicitly combine
the terrorists types in LapanSandler and Overgaard to create a signaling model in
which the target government has incomplete information about whether terrorists
are militantly motivated (M-types) or politically motivated (P-types). M-types
accept only full concessions, whereas P-types accept partial concessions. M-types
expend resources on attacks if their demands are not fully met; P-types allocate their
remaining resources to political purposes if no accommodations are made. P-types
are also concerned about losing the high moral ground from a never-ending terro-
rist campaign. M-type terrorists, however, see violence as sanctified and not as
a (temporary) necessary evil. Such terrorists have an incentive to create backlash
attacks in reaction to their targets response and may even franchise new groups
for this purpose.
Based on these considerations, we would ex post classify Euskadi ta Askatasuna
(ETA), the Provisional Irish Republic Army (PIRA), the African National Congress
(ANC), the Stern Gang, Irgun, the Tamil Tigers, Hamas, Hezbollah, and most
ethno-nationalist terrorist groups as P-types. For example, ETA seeks regional auton-
omy and self-determination over social services, schooling, taxation, and law and
order. At times, ETA has agreed to cease-firesthe most recent ending in June
2007when it saw progress with respect to its demands. ETA limits collateral dam-
age and directs most of its violence against symbols of the government. When more
general targets are chosen (e.g., tourists hotels), ETA provides advanced warnings of
bombings to minimize casualties. A political partyBastasunarepresents the polit-
ical agenda of ETA. In the case of PIRA, the Irish Army Council ended its military
campaign on July 28, 2005, at which time it resorted to democratic means to pursue
political goals. The ANC, the Stern Gang, Irgun, and the Tamil Tigers all had or have
clear political goals. At one point, the Tamil Tigers temporarily suspended its terrorist
campaign when it perceived progress toward its autonomy goal. Although they are
formidable and violent, Hamas and Hezbollah provide social services so that attacks
have a real opportunity cost in terms of their political objectives.
In contrast, M-type terrorists present demands that cannot be partially satisfied
for example, al-Qaidas demands for fundamentalist governments and an end to Israel
or Jemaah Islamiyahs demand for a pan-Islamic state. The Egyptian Islamic Jihad
(al-Jihad) is also an M-type group, now allied with al-Qaida. In the past, the Abu
Nidal Organization (ANO) was bent on murder and presented ambiguous demands.
In Uruguay, the Tupamaros (196872) failed to win over a constituency owing to
its brutality (Enders and Sandler 2006, 17-18).
356 Journal of Conflict Resolution 54(2)
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
One must wonder why a government may not be informed about the nature of the
terrorist threat. Incomplete information may stem from myriad factors. First, the ter-
rorist group may be new with no past track recordfor example, al-Qaida in the early
1990s. Second, the group may have just splintered off from an established group.
Terrorist groups may have both militant and political factions vying for control.
Governmental partial concessions may result in the genesis of a harder-line
groupfor example, ANO and the Popular Front for the Liberation of Palestine,
which broke off from Fatah (Sandler and Arce 2003). Third, terrorist groups may
not claim responsibility for an attack so that the government is in the dark about
the groups orientationfor example, neither the Black Hawk down incident in
Somalia (October 1993) nor the bombing of the Khobar Towers housing U.S. Air
Force personnel in Dharan, Saudi Arabia (June 1996), were claimed. Fourth, multiple
competing claims of responsibility may be made by groups for the same attack or
attacks. Fifth, a terrorist group may acquire a new leader so that its orientation morphs.
Sixth, terrorists may purposely exhibit uncharacteristic behavior in the short term
to enhance their bona fides with their constituency or expand their base of political
support (Hoffman and McCormick 2004; Abrams 2006).
Given this lack of information concerning a terrorist groups intentions, a target
government may face two types of regret: P-regret from conceding to a political group
that would not have continued attacking and M-regret from not responding with the
proper counterterrorism measures that limit subsequent damage from militant attacks.
P-types have an incentive to mimic M-types with a spectacular incident that might
result in concessions. M-types have a rationale for holding back during their initial
attack to curb the governments counterterrorism response. Such issues and the proper
policy response are explored below, after the basic model is presented.
The Model
Our study is based on a two-period signaling model of terrorism without discounting.
We do not consider a repeated game approach because 90 percent of terrorist organiza-
tions have a life span of less than one year; and of those that make it to a year, more than
half disappear within a decade (Rapoport 1992; Blomberg, Engel, and Sawyer 2010). In
democratic target nations, the governments time horizon is also short because of the
turnover of policy makers. For a two-period model, both terrorist types would heavily
discount future benefits. The presence of discount factors implies that all groups are
biased in favor of first-period spectaculars. Given our focus on the generation of spec-
tacular attacks, we reduce this potential bias and investigate whether such spectaculars
occur even without discounting. Finally, in the majority of discrete signaling models of
terrorism, the discount factor serves as a surrogate for the effect of the governments
response to the initial level of terrorism. When the government does not concede, it
is assumed that the government raises its defenses, thereby reducing the costs associated
with second-period attacks. This effectively discounts the gain from second-period at-
tacks. In our model, this use of a discount factor is no longer required because the gov-
ernments gains and losses vary continuously with its response, r.
Arce and Sandler 357
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
We consider a situation where terrorists send a signal, s 0; 1 ; corresponding to
their first-period attacks, where terrorists have total resources at their disposal equal
to 1. Governments react to this signal with counterterrorism policy r 0; 1 ; where
nonintelligence action r decreases the effectiveness of a second-period attack on
a one-to-one (1:1) basis.
1
The upper limit of 1 implies that we are examining the rel-
ative trade-off between counterterrorism activities, r, and concessions, 1 r. In con-
trast to prior signaling models, this relative policy trade-off is now a continuous
variable rather than a discrete all-or-nothing choice. In keeping with prior signaling
models, we are interested in this policy trade-off rather than a simultaneous increase
in both concessions and counterterror activities. Similarly, we do not further decom-
pose r into proactive and defensive components (nor does any other signaling model)
because the different transnational externalities that these policies produce are not the
subject of inquiry.
Furthermore, (1 r)S translates into the concessions made to terrorists when the
government takes stance, S, against terrorism. The more hard-line the government,
the greater is S. As a governments declared stance of toughness increases, any con-
cessions become more costly since they indicate a greater compromise of the govern-
ments principles. By the same token, terrorists place a higher value on concessions
gleaned from a tough government. As the degree of concession is given by the term
1 r , partial concessions represent a cost of 1 r S to the government and
a gain of 1 r S to terrorists who accept these concessions.
We follow the unifying convention in which two types of terrorists are considered
P-type (political) and M-type (militant)as described earlier. This convention general-
izes prior models in which terrorists vary in terms of the resources at their disposal but
always have sufficient first-period resources to mount a spectacular attack.
In specifying the payoff function for M-types, we assume that they expend all
remaining resources in a second-period attack and receive a benefit from their first-
and second-period attacks. Prior signaling models employ a dichotomous government
strategy set defined as (concede, not concede). When the government conceded, it was
a full concession so that there was little doubt that M-types would accept it. By exten-
sion, we assume that M-types do not seek or accept partial concessions. This is con-
sistent with the description of absolute (Hoffman and McCormick 2004), fanatical
(Franck and Melese 2004), and maximalist terrorists (Abrams 2006), a collection of
terms falling under our M-typology. For example, the radical Islamist movement
has never had a clear idea of participation in governance, or even much interest in
it. Purification is the goal (Wright 2006, 247), which corresponds to full concessions
here. Our assumption is consistent with M-types willingness to accept (de facto) full
concessions in prior models, with the added implication that partial concessions imply
that less funds are expended on counterterrorism. This then increases the potential
damage of a second-period attack by M-types. Effectively, the benefits that M-types
perceive from second-period attacks under reduced defenses are greater than those
from partial concessions.
In addition, M-types receive an increase of resources stemming from the govern-
ment response equal to br; where 0 b < 1: This backlash term is investigated in
358 Journal of Conflict Resolution 54(2)
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
prior nonsignaling models by Rosendorff and Sandler (2004), Faria and Arce (2005),
Jacobsen and Kaplan (2007), and Siqueira and Sandler (2007). The upper bound on
b is justified because a response is likely to include defensive and proactive compo-
nents, and only the latter is likely to elicit a backlash. Indeed, Kydd and Walter (2006)
classify one possible motivation for terrorism as an attempt to provoke targets into
overreactions that turn public opinion into support for terrorists. Similarly, Wright
(2006) finds that one of Osama bin Ladens goals in implementing the attacks of 9/
11 was to draw al-Qaidas opponents into responses that turn out to be repressive blun-
ders, thereby encouraging attacks by other Muslims.
2
The M-types payoff is given as,
Ms; r
1 br r when r 0; 1; and
S s when r 0 full concessions:
_
The 1 term in M s; r 1 br r is the benefits that militant types receive from
expending all of their resources over both periods by violent means. It is the sum of
the damages done in the first-period attack, s, and the second-period attack, 1 s: These
benefits are offset by the governments counterterrorism policy, r; and are augmented
by the backlash effect, br, which is treated as an additional resource used in M-types
second-period attack. Furthermore, the governments response, r; depends on the initial
level of terrorism, s: Partial concessions, 1 r; enter into M s; r 1 br r through
two channels. First, as partial concessions increase, the government reduces its second-
period countermeasures, thereby increasing M-types benefits from a second-period
attack. Second, increased concessions translate into a reduced r; which diminishes
the backlash effect, br: The target population for concessions and backlash are often
the same (Faria and Arce 2005; Frey and Luechinger 2003).
Under full concessions, M-types benefit from the first-period attack, s; plus the
concession, S; and do not attack in the second period. With full concessions, the
top term in M s; r is 1. M-types are better off accepting full concessions when
1 < S s; that is, the benefit from expending all their remaining resources in a
second-period attack, 1 s; is less than what is received from full concessions,
S: Furthermore, the following inequality holds: S > 1 s: The upper bound on
a first-period attack in terms of terrorists resources explains why 1 s: The funda-
mental notion of terrorism as asymmetric conflict justifies S > 1: Specifically, the
consequences of fully conceding are greater for the government than is the amount
of resources terrorists have on hand to mount attacks. An alternative interpretation
of S > 1 stems from r 0; 1 : Given that a second-period attack cannot exceed 1
and that the governments response reduces the effectiveness of an attack on a 1:1
basis, the response need not exceed 1. Hence, S > 1 reflects reality because the gov-
ernment has more at stake when it fully concedes than what it budgets for
counterterrorism.
In the absence of full concessions, M-types send signal s with anticipated response,
r > 0; instead of signal s
0
with response r
0
> 0; whenever 1 br r 1 br
0
r
0
:
That is,
Arce and Sandler 359
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
M signals s when r
0
r 0: Recall that b <1: 1
P-types view their first-period signal as a pure cost and receive benefits for their re-
maining resources, 1 s; in the second period as they are applied to a political goal.
With partial concessions, P-types receive payoff,
P s; r 1 s 1 r S:
The first term in P s; r is the degree to which violence constitutes an opportunity
cost relative to using resources for nonviolent political purposes. The second term is
the concessions that the government grants in response to s. In comparing payoffs for
signal s and response r to those of s
0
and r
0
; we find that,
P signals s when r
0
r
1
S
s s
0
: 2
The terrorists (senders) payoffs are nonincreasing in the governments (receivers)
response, which is atypical for signaling models. When the government increases
its response, this decreases the payoffs for either type of terrorist. By contrast, in
the educational signaling model, an increased wage (employer response) raises the
payoff for both types of potential employees, regardless of their intrinsic level of pro-
ductivity. In addition, terrorists do not all wish to be perceived as one type, as is gen-
erally the case in signaling models (e.g., to be perceived as a high-productivity
worker). Instead, there are advantages for P-types to be perceived as M-types if this
convinces the government to grant greater concessions to avoid a feared backlash.
As shown below, the governments optimal response, r, is tempered to reduce the
potential for a backlash, and what remains, 1 r; can be interpreted as partial conces-
sions meant to limit the backlash effect. P-types benefit from these partial conces-
sions. Furthermore, M-types can benefit from being perceived as P-types if the
governments restrained response augments the prospects of second-period attacks.
The governments (Gs) payoffs are zero sum in attacks and concessions. If m is the
governments belief that it faces an M-type when signal s is sent, then the govern-
ments expected payoff is,
Gs; r
1 s r br 1 1 rS s; for r > 0; and
S s; for r 0 full concessions:
_
In the top expression, the first term in brackets is the expected loss from a second-
period attack by M-types, 1 s, which is offset by counterterror policy, r, and
augmented by the backlash that the response provokes, br: The second term is the
expected loss for making concessions to P-types, while the third term is the damages
from the first-period attack, s. In the bottom expression, the government makes full
concessions and loses S along with the first-period attack, s: Against demands
that correspond to purification or the loss of the targets status as a political
entity, we assume that full concessions are too costly for the government; that is,
360 Journal of Conflict Resolution 54(2)
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
G s; r > G s; 0 for all levels of terrorist signals s 0:
3
In an analysis of twenty-
eight recent terrorist campaigns, Abrams (2006) finds the instance of total
successin terms of terrorists expressed objectivesto be exceedingly rare, con-
sistent with our assumption of prohibitively costly full concessions. In contrast to
terrorists types, the government stance is known by the terrorists. There is little
doubt as to the terrorism stance of Israel and the United States in light of the
demands outlined above.
The governments payoff function is essentially an expected loss function that is
appended by the cost of the initial attack, s. Hence, the governments objective is
to select a policy mix, r, that limits its expected losses given the cost of first-period
attack, s; that is, G s; r

s: Another way to interpret this equality is that the

game is zero sum in attacks and concessions; hence, if we remove the s term
from both sides of the equality, the governments objective is to minimize the
expected losses associated with a second-period attack and/or concessions given the
damages done in the first-period attack. In this interpretation, the governments trun-
cated payoff function is a loss function whose minimum expected value is zero; that is,
the government selects r such that m 1 s r br 1 m 1 rS 0:
Solving for r in either equality yields the loss-minimizing government policy:
r

1 s 1 S
1 b 1 S
: 3
This response, r

, sets the governments expected losses equal to zero exclusive of the

initial attack, and the losses are equal to s inclusive of the initial attack, s. As the
expected losses cannot be further reduced via a different response, r

is the govern-
ments best response given the terrorists strategy, s, and government beliefs, : Con-
sequently, counterterrorism policy is a function of the incomplete information that the
government has about the type of terrorists that it faces, as reflected by the term in
equation (3). There is therefore a role for intelligence to better inform policy, a subject
we investigate in detail below. Briefly, under complete information, let r
M
be the gov-
ernments policy when it faces an M-type group and let r
P
be its policy when it faces
a P-type group. When the government knows that it confronts a P-type, 0; imply-
ing that r
P
1 from equation (3). The government makes no concessions when know-
ingly facing a P-type. Similarly, when the government knowingly faces an M-type, so
that 1; equation (3) implies that r
M
1 s = 1 b : The government responds
to the anticipated second-period attack by accounting for the backlash, b, that favors
the terrorists. If b 0; then the government expends counterterrorism resources to
completely offset the effect of the second-period attack: r
M
1 s: By definition,
it must also be the case that r
M
1; that is, b s; so that the second-period multi-
plier effect of a response is less than the damage done by the first-period attack.
Pooling Equilibria
Intelligence is relevant in signaling games of terrorism when the targeted government
cannot distinguish between terrorist types and appropriately tailor its counterterrorism
Arce and Sandler 361
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
response. We have already discussed that M- and P-types may have incentives to
mimic one anothers first-period attacks. When both types mount the same first-period
attack, the government cannot use this as a signal to update its priors about the terrorist
threat and adjust its response accordingly. Under the appropriate conditions (identified
below), this game has a continuum of pooling equilibria in which both types select
initial signal/attack level s, relative to out-of-equilibrium signal s
0
: The groups are
pooling on s.
Pooling equilibria are often sensitive to the beliefs and strategies that support them
off the equilibrium path. In perfect Bayesian equilibria, beliefs are unrestricted off
the equilibrium path. Although defined for general signaling games, most signaling
refinements are motivated by examples where the senders payoff is monotonic in
type; for example, low versus high productivity in educational signaling or weak ver-
sus strong in the beer-quiche game. In such cases, the motivation for refining the set of
off-the-equilibrium-path beliefs is based on interpreting an out-of-equilibrium mes-
sage as a higher order signal. As we have discussed above, the game under study is
non-monotonic in type, with each type having some incentive to be taken for the other
type. When the senders preferences are no longer monotonic, beliefs and responses
to off-the-equilibrium-path deviations can no longer be nested by type and signaling
refinements no longer apply (Banks 1991).
In particular, when s
0
is the out-of-equilibrium signal,
0
is the governments belief
that an M-type sent this signal and r
0
is the governments off-the-equilibrium-path
response. When s > s
0
; condition (2) on P-types rationality is the binding constraint
on terrorists signals, and when s
0
> s; equation (1) is the binding constraint. We there-
fore make the following assumption based on the non-monotonic nature of our game.
Assumption: For any (perfect Bayesian) pooling equilibrium where r; is the
governments strategy/belief assessment on the equilibrium path, s, the
assessment for off-the-equilibrium-path signal s
0
< s is r
0
;
0
1; 0 ;
and r
0
;
0
r; is the governments assessment for off-the-equilibrium-
path signal s
0
> s:
This set of beliefs recognizes the non-monotonic nature of types in our game by spec-
ifying no concessions and maximum defenses for off-the-equilibrium-path signals (at-
tacks) that are less than the pooling signal.
4
It recognizes the incentive for each type to
look like the otherthrough an out-of-equilibrium signaland, through equation (3),
sets r accordingly. Furthermore, it is consistent with a no negotiation/concession stance
that often characterizes counterterrorism policies. In a pooling equilibrium, the govern-
ments belief that it is facing an M-type on the equilibrium path is equal to the prior
probability that terrorists are M-types. Hence, throughout the remainder of this section,
we interpret 0; 1 as this prior probability. The condition r
0
;
0
r; for
s
0
> s implies that the government does not further update from its pooling strategy
if the off-the-equilibrium-path attack is greater than the pooling level. Again, non-
monotonicity comes into play because M-types have an incentive to seek a greater
response to provoke a backlash, and, given that no concessions occur for lower than
pooling attacks, s
0
< s; this is the only path where concessions are a possibility (from
362 Journal of Conflict Resolution 54(2)
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
the perspective of P-types). Hence, there is no rationale for the government to update
any further. Finally, given that a pooling equilibrium must simultaneously satisfy con-
ditions (1) and (2), our assumption gives the following intuitive characterization.
Result 1: Pooling requires the government to believe that it is more likely
(greater than .5) to be facing an M-type.
Proof: Applying the assumption about out-of-equilibrium beliefs to equation (2)
yields,
1
1 s 1 S
1 b 1 S
..
r

s
S
; which reduces to;
s b
1 b 1 S

s
S
: 4
Cross-multiplying, we have,
sS bS s bs sS sS: 5
Equation (5) becomes s 2S b 1 S bS: As the right-hand side of this
inequality is positive, it is necessary for the left-hand term in brackets to be positive
for the inequality to hold (recall s > 0). The bracketed term is positive when,
>
S
2S b 1
:5
:51 b
2S b 1
=

: 6
Several implications follow from this result. First, the response, r

, that satisfies
equation (4)characterized in terms of ; s; b; and Sinvolves partial concessions
r

< 1 : In prior signaling models, pooling spectaculars occur under full concessions;
yet full concessions are rarely granted in practice. Moreover, for studies that call the
rationality of terrorism into question, given the lack of evidence that terrorist groups
successfully achieve their demands (e.g., Abrams 2006), we find an equilibrium ratio-
nale for terrorism that is based on pooling the goals of partial concessions for P-types
and backlash effects for M-types. Furthermore, even though M-types do not accept
partial concessions, target governments should understand the policy implication
that partial concessions reduce the backlash effect that M-types seek to produce by
placating the M-types target constituency rather than the M-types themselves. Sec-
ond, in the vernacular of the intelligence community (Safire 1995), all that a pooling
equilibrium requires is that it is a probability that terrorists are militant, where the term
probability refers to a likelihood that is greater than .5.
Second, we can endogenously derive the first-period level of attack that allows for
pooling. Under the above conditions, equilibrium condition (5) reduces to pooling on
an attack, s, which meets or exceeds spectacular level, s

, where,
s
bS
2S b 1 S
s

: 7
In prior signaling models of terrorism, s

is exogenously defined as a large incident,

whereas s

is now derived in terms of the equilibrium requirements for pooling. This is

Arce and Sandler 363
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
novel for two reasons. First, we have endogenously generated the lower bound on the
pooling attack. Second, we can demonstrate that the spectacular interpretation
is, indeed, correct. In a separating equilibrium, P-types attack at level s
P
0 and
M-types attack at level s
M
s

M
; where s

M
is the upper bound on an M-types separat-
ing attack. We later show in result 5 that s

> s

M
; so that the pooling level of attack is
greater than any level associated with a separating equilibrium, thereby making s

the
lower bound for spectacular attacks.
Comparative statics on s

yield,
s

bS
2
2S b 1 S
2
< 0:
The greater the (prior) probability that the government believes that it is facing an M-
type, the smaller the pooling spectacular attack. Effectively, P-types can free ride on the
likelihood of M-types in terms of the resources necessary to mount a spectacular. P-
types do engage in spectacular attacks to gain concession by appearing bent on even
greater destruction if not appeased. Irgun Zvai Leumi bombed the local British military
headquarters at the King David Hotel in Jerusalem on July 22, 1946, killing ninety-one
(Enders and Sandler 2006, 250). This incident greatly contributed to the British decision
to end its occupation. Hezbollah employed two large-scale suicide truck bombings
against the U.S. Marines and the French Paratroopers in Beirut on October 23, 1983.
The Multinational Force departed Lebanon, as demanded, shortly after these incidents.
Other P-typesfor example, Tamil Tigersemploy large-scale suicide attacks.
Finally, the al-Qaida offshoot responsible for the Madrid train station bombings threat-
ened to turn Spain into an inferno if Spain did not withdraw its troops from Iraq.
Another comparative static result regarding spectacular attacks relates to the
toughness of the governments no-concession stance, S:
s

2
b b 1
2S b 1 S
2
< 0:
The lower bound on a spectacular attack is inversely related to S. When a government
is hard-line (high S), the threshold falls for classifying spectaculars. Finally, we have,
s

S 2S 1 S
2S b 1 S
2
>
<
0:
In a pooling equilibrium, :5; 1 so that the effect of backlash attacks on what con-
stitutes a spectacular is ambiguous. When the government is uncertain that it faces an
M-type :5 ; greater backlash lowers s

. When, conversely, the government is

certain that it faces an M-type 1 ; greater backlash raises s

. For the latter,

the presence of backlash raises the stakes in terms of the required size of a pooling
spectacular. The government realizes that first-period attacks are intended to provoke
a large response resulting in more backlash, so government restraint is required.
364 Journal of Conflict Resolution 54(2)
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
Result 2: Spectacular attacks that meet or exceed the level of s

given in equa-
tion (7) are consistent with a continuum of pooling equilibria. The defining
level for spectaculars is inversely related to the governments stance against
terrorism, S, and its (prior) belief that it is facing an M-type. In contrast, the
influence of a backlash on the defining level of spectaculars is ambiguous.
Another important value is the lower bound on the governments prior that it is fac-
ing an M-type, consistent with the pooling equilibrium,

: From equation (6), it is

clear that

=S < 0 and

=b < 0: Thus, the greater is the governments stance

against concessions, the less certain that government needs to be that terrorists are an
M-type for a pooling equilibrium, in which the government does not fully concede.
Moreover, the greater is the backlash, the less certain that the government needs to
be that terrorists are M-types for a pooling equilibrium.
Result 3: The (prior) belief that the government is facing an M-type, consistent
with a pooling equilibrium, is inversely related to the governments stance
against terrorists and the potential backlash.
The Value of Intelligence
In models with incomplete information about terrorists resources or preferences, the
value of intelligence stems from better informing the government on its choice of
strategy. For example, Lapan and Sandler (1993) measure this value regarding
whether it is optimal for a government to concede or not. Arce and Sandler (2007)
instead identify a potential trade-off between conceding to an M-type versus decreas-
ing the effectiveness of a second-period attack. In these models, the value of intelli-
gence involves the avoidance of ex post regret. Given a pooling equilibrium, ex
post P-regret occurs when the pooling payoff is compared to the no-concede pay-
off, which is the optimal strategy when the government faces a P-type. Ex post
M-regret is measured as the difference between the pooling payoff and the govern-
ments optimal policy when it knows that it confronts an M-type. We also determine
the value of intelligence by comparing the governments ex post payoff for a given
type under its pooling response with its optimal complete information strategy.
Because the policy response is a continuous variable in our model, intelligence tells
the government not only whether to concede or not but also how to tailor its response
in anticipation of a second-period attack. Hoffman and McCormick (2004, 247) sim-
ilarly observe that intelligence is needed not only to find, fix, and destroy terrorist
groups but also to develop an accurate picture of the threat that the group is likely
to pose down the road. Specifically, in our model, intelligence guides the optimal
counterterrorism response in recognition of the responses potential to create a back-
lash that fuels the terrorist campaign, as given in equation (3). The potential for intel-
ligence to inform counterterrorism policy distinguishes this study from prior signaling
models and further illuminates intelligences role as a complement to event-specific
information.
Arce and Sandler 365
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
We use equation (3) to ascertain the governments optimal payoff with complete
information. For P-type terrorists, G s; r
P
equals s; as 0 ) r
P
1; for
M-type terrorists, G s; r
M
also equals s because 1 ) r
M
1 s = 1 b :
In the latter case, the governments countermeasures completely offset the second-
period attacks by accounting for any backlash. Given that the governments complete
information payoff is s; we now calculate the value of intelligence.
Case 1: Ex post the government faces a P-type. The governments ex post pay-
off is, 1 r S s; corresponding to the concession that it makes and its
losses from the first-period attack. In the pooling equilibrium, the govern-
ments response is given by equation (3). Substituting this value for r into
the ex post payoff yields,

s b
1 b 1 S
_ _
S s:
Since the value of intelligence is the avoidance of ex post regret, it is the difference
between the complete information payoff, s; and the pooling payoff above. For
a P-type, this value, v
P
; is given by,
v
P

s b
1 b 1 S
_ _
S: 8
Case 2: Ex post the government faces an M-type. The governments ex post
payoff is, 1 s r br s 1 r br 1 s r br s
1 r br : Substituting the pooling response, r, based on equation
(3), we find that the governments ex post payoff for an M-type is,

s b
1 b 1 S
_ _
b
1 s 1 S
1 b 1 S
_ _
s:
The difference between the complete information payoff, s; and this pooling payoff
is the value of intelligence when facing an M-types, in a pooling equilibrium, v
M
:
v
M

s b
1 b 1 S
_ _
b
1 s 1 S
1 b 1 S
_ _
: 9
We now ascertain whether a ranking exists between the value of intelligence to
avoid P-regret versus M-regret. We have assumed that S > 1 r; so that the cost
of fully conceding to terrorists (S) exceeds the budgeted response to terrorist attacks
1 r .
5
When comparing the two values of intelligence, we have,
v
P
> v
M
, s b S > s 1 b :
Since b < 1; the right-hand side of this inequality is positive. Given that S is also pos-
itive, a necessary condition for this inequality is > b=s: As we earlier derived that
366 Journal of Conflict Resolution 54(2)
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
r
M
1 implies s b; the necessary condition holds when the prior probability of an M-
type takes an extreme value; specifically, b=s; 1 : If s b, this condition cannot
be met and v
P
< v
M
: Furthermore, when falls within this range, the government is
relatively certain that it is facing an M-type. There is very little need for intelligence in
such circumstances. By contrast, when takes an intermediate value,

; b=s ;
this is consistent with the notion of a pooling equilibrium in that the government is
relatively uncertain about the type of terrorist it is facing.
6
In other words, the value
of intelligence is an issue when m takes an intermediate value rather than an extreme
one. This implies that M-regret is the overriding issue; that is, v
P
< v
M
. We can thus
conclude that, in the presence of backlash attacks (b s0), the value of intelligence is
defined by the avoidance of M-regret.
This observation calls into question the focus on P-regret in prior signaling models
where the government either fully conceded or held firm. In the current analysis, the
value of intelligence is almost always defined by the avoidance of M-regret. The iden-
tification of the nature of intelligence comes from our generalized framework that
allows for a continuous response strategy that weighs defensive benefits and counter-
productive backlash reactions.
Because
M
is the likely measure of the value of intelligence, we are interested in
the comparative statics of
M
relative to the governments stance and backlash. To
begin, we have v
M
=S 0 (see the appendix). In Arce and Sandler (2007), the rela-
tion between the value of intelligence and the governments stance is ambiguous
because of the discrete nature of the policy response that allows for both forms of
regret. Our unambiguous comparative static finding means that a tough stance to
weather attacks is best served by investing resources in strong counterterrorism meas-
ures, which includes event-preventing intelligence. With a tougher stance, there is less
to gain from knowing the nature of the threat since concession is less of an option.
Moreover, the ways in which s

and v
M
are related to the backlash term have impor-
tant policy implications. First, it is unclear whether the potential for backlash increases
or decreases the magnitude of spectaculars (s

) required for pooling equilibria (result

2); yet in a pooling equilibrium, intelligence is required to avoid M-regret. Second, the
value of information to avoid M-regret is inversely related to the size of the backlash
to the governments response: v
M
=b < 0 (see the appendix). This follows because
knowledge of b is effectively a substitute for intelligence regarding terrorist types,
given that only M-types seek to create a backlash to the government response. If, con-
sequently, the government determines the potential for a backlash, it then knows
whether M-types have a greater incentive to mount a spectacular attack in a pooling
equilibrium. In other words, intelligence should be focused on the propensity for coun-
terterrorism to result in a backlash.
Separating Equilibrium
We now consider the case where s
M
6 s
P
( 0). In such a separating equilibrium, the
government has perfect information about which type of terrorist it is facing at the
information sets following signals s
M
and s
P
; respectively. Under these circumstances,
Arce and Sandler 367
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
the government recognizes political types for what they are; hence, political types
attain their goals via political channels rather than terrorist actions that lead to (partial)
concessions. For example, a potential terrorist organization whose own strategic cal-
culations lead it to conclude that the cost of violence is too high may instead resort to
civic action to highlight its agenda. Indeed, such a calculation often leads to splinter-
ing, a common phenomenon under which the more violently inclined members fight
for other terrorist groups or form their own organization. By contrast, terrorist groups
that attack at level s
M
are known to be militant. We characterize the type of attack
made by militant terrorists in the ensuing section.
Under complete information the governments responses are r
M
and r
P
; respec-
tively. The best-response characterization in equation (1) for an M-types signal is
r
P
r
M
0; while the best-response characterization in equation (2) for a P-type sig-
nal is r
M
r
P

s
M
S
; or r
P
r
M

s
M
S
: Combined, these inequalities indicate that,
s
M
S
r
P
r
M
0:
Furthermore, 1 at the information set corresponding to s
M
, and 0 at the
information set corresponding to s
P
: From equation (3), this results in
r
M
1 s
M
= 1 b and r
P
1: The government adjusts its response, r
M
; accord-
ing to the anticipated second-period attack and potential backlash, and it does not con-
cede to P-types. Combining these values with the above inequality produces,
s
M
S
1 1 s
M
= 1 b 0 or
s
M
S
s
M
b = 1 b 0:
Cross-multiplying and solving for s
M
gives an upper bound for the separating
equilibrium, s

M
:
s
M

bS
S b 1
s

M
: 10
Result 4: A separating equilibrium (s
M
, s
P
) exists in which P-types do not attack
and receive no concessions s
P
0 and 1 r
P
0 and M-types initially
attack at or below s

M
. The governments optimal response to the attack is
r
M
1 s
M
= 1 b :
The comparative statics on upper bound s

M
are as follows:
s

M
S

b b 1
S b 1
2
< 0; and
s

M
b

S S 1
S b 1
2
> 0:
In a separating equilibrium, M-types must attack belowthe level, s

M
, which is an inverse
function of the governments hard-line stance, S. When facing a tough government, mil-
itant terrorists are better off by not provoking an immediate response and, instead, in-
tertemporally substituting into second-period attacks. If, however, the resulting
backlash to the governments response is expected to be high, then militants should
368 Journal of Conflict Resolution 54(2)
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
instead move their plans forward to promote recruitment. The intertemporal nature of
attacks is therefore characterized by a fundamental trade-off between the effectiveness
of government counterterrorism policy and the backlash that the policy provokes.
Spectacular Attacks Are Pooling Phenomena
The trade-off between the additional benefit that M-types seek by restraining car-
nage in the first period versus the potential for provoking a backlash raises the issue of
whether the upper bound on M-types separating attack, s

M
, crosses the lower bound
on the pooling equilibrium spectacular, s

. Formally, we have,
Result 5: (Spectacular) pooling attacks cannot be considered militant attacks,
consistent with a separating equilibrium; that is, s

M
< s

:
Proof: See the appendix.
The fact that the pooling attack, s

, exceeds the level of an M-type attack associated

with a separating equilibrium, s

M
, is what allows us to label s

as a spectacular.
Results 2 and 5 allow us to endogenously derive and characterize the value of a spec-
tacular, in contrast to its exogeneity in other signaling models of terrorism. Instead of
the conventional wisdom that M-types seek to distinguish themselves through spectac-
ular attacks, we have the novel result that spectaculars are pooling phenomena. In
a separating equilibrium, the government does not concede to P-types, and it knows
to tailor its response to M-types based on the expected magnitude of the second-period
attack and the potential backlash that the government response may elicit. Partial con-
cessions are made in the pooling equilibrium to reduce the backlash effect of a gov-
ernments response. This implies a benefit for M-types because the presence of
incomplete information means that the governments response does not fully antici-
pate M-types second-period attack. Furthermore, P-types directly benefit from the
partial concessions. Consequently, both M- and P-types have an incentive to create
a pooling equilibrium through a spectacular attack, thereby preserving a situation of
incomplete information. By contrast, in a separating equilibrium the government
knows when it is facing an M-type. Consequently, a spectacular is suboptimal given
the governments response, r
M
:
Concluding Remarks
This article presents a signaling model of terrorist attacks where the government is
uninformed about whether the terrorists are politically or militantly motivated. The
former view attacks as a pure cost and will accept partial concessions, while the latter
derives fulfillment from the attack itself and will accept only full capitulation. We
derive a characterization of spectacular terrorist incidents as a function of the countrys
declared stance of toughness and the terrorists goal of inciting counterterrorism-
induced backlash recruitment. Although M-types do not accept partial concessions, a tar-
get government offers such concessions to limit backlash, thereby reducing militants
payoffs. We associate spectacular incidents with a pooling equilibrium, where the
Arce and Sandler 369
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
government cannot distinguish the nature of the terrorist threat. That is, M-types engage
in large-scale incidents to prod the government into a response that provokes a backlash.
P-types also resort to large-scale incidents in the hopes of receiving partial concession,
given by governments to limit a backlash. In a pooling equilibrium, the value of intel-
ligence derives fromthe proper policy mix of countermeasures and concessions to com-
bat just M-types. Thus, intelligence on the potential for a backlash serves as an effective
substitute for identifying the nature of the terrorist type. This insight is not available
from prior signaling models.
Two additional insights arise from this analysis. First, it is in the governments
interest to minimize the backlash effect of counterterrorism policy because this limits
future militant terrorists resources and the governments need for intelligence. For
example, policing is a particularly effective form of counterterrorism activity with
minimal backlash effects as compared to direct military action (Jones and Lubicki
2008; Sandler, Arce, and Enders 2009). This suggests that backlash should be a pri-
mary consideration when designing counterterrorism policy. It also suggests a future
extension of the model in which the governments ability to gain intelligence is endo-
genized. Finally, although terrorism has been shown to be an inefficient means for
fully gaining a groups demands (Abrams 2006), we have established that partial con-
cessions are often the result of strategic interaction between terrorist groups and target
governments, particularly when terrorists seek to produce a backlash reaction to a gov-
ernments counterterrorism response.
Appendix
Proof that v
M
=S 0 : From equation (9), v
M

sb
1b1S
_ _
b
1s1S
1b1S
_ _
:

M
S

1 s b
1 b 1 S
2
_ _
b
1 1 b 1 S 1 1 s 1 S
1 b 1 S
2
_ _
:
The denominators are common and positive, so that they can be eliminated. By divid-
ing through by 1 > 0 and aggregating terms, we have,
sign

M
S
_ _
sign s b bs b f g sign b 1s b f g:
Since b < 1 and s b; v
M
=S 0:
Proof that v
M
=b 0: Using the value of v
M
given in equation (9), we have,

M
b

1 b 1 S
2
s b
1 b 1 S
2
_ _
b
1 s 1 S
1 b 1 S
2
_ _
:
370 Journal of Conflict Resolution 54(2)
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
We again eliminate the common positive denominator. Next, we divide the numerator
by > 0; which leaves,
sign

M
b
_ _
sign 1 b 1 S s b f
b 1 s 1 S g:
Aggregating terms, we get,
sign

M
b
_ _
sign b 1 1 S 1 s f g:
Since ; b < 1; s 1; and S > 1; this implies v
M
=b < 0:
Proof of result 5: The result holds when s

M

bS
Sb1
<
bS
2Sb1 S
s

; because
then any pooling spectacular that exceeds s

will not be chosen by M-types in a sepa-

rating equilibrium. As shown when interpreting equation (5), the denominator on the
right-hand side of this inequality is positive; otherwise, no pooling equilibrium exists.
The denominator of the left-hand side is positive because S > 1: Cross-multiplying,
we have, 2bS
2
b
2
S bS bS
2
< bS
2
b
2
S bS: By aggregating
terms, we get, bS
2
< bS
2
or < 1; which holds by definition.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the authorship and/or pub-
lication of this article.
Financial Disclosure/Funding
The authors disclosed receipt of the following financial support for the research and/or author-
ship of this article: This study was funded, in part, by the U.S. Department of Homeland Secu-
rity through the Center for Risk and Economic Analysis of Terrorist Events at the University of
Southern California, Grant 2007-ST-061-000001. Any opinions, findings, and conclusions or
recommendations are solely those of the authors and do not necessarily reflect the view of
the Department of Homeland Security.
Notes
1. If, instead, the response reduces the effect of an attack by a factor of , then r reduces the
effect of an attack on a 1:1 basis and the government would, for all practical purposes, be
choosing r as its strategy.
2. As a point of clarification, it is the governments response that produces a backlash. We are
assuming that M-types seek to provoke this backlash while P-types do not. Alternatively, one
could think that M-types receive a greater benefit from the backlash because they are not
interested in political activities or partial concessions that benefit P-types. In this interpreta-
tion, we are normalizing the backlash benefits that P-types receive to zero while normalizing
the political benefits that M-types receive from partial concession to zero.
Arce and Sandler 371
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
3. This reduces to S > 1 s r br 1 1 r S , which is the prior probability
of an M-type multiplied by the M-types payoff from a second-period attack plus the prior
probability of a P-type multiplied by the partial concessions received in the second period.
In other words, the government has more at stake from fully conceding than the aggregate
damage because of partial concessions.
4. This is equivalent to the off-the-equilibrium path belief structure in Arce and Sandler (2007),
where only s
0
< s is possible.
5. If, however, S 1 , then clearly v
M
> v
P
.
6. The interval

; b=s is well defined when b=s >

or b > s

. In other words, the back-

lash effect must have sufficient potential; otherwise, there is no incentive for M-types to pro-
voke a response with a spectacular.
References
Abrams, Max. 2006. Why terrorism does not work. International Security 31 (2): 42-78.
Arce, Daniel G., and Todd Sandler. 2007. Terrorist signalling and the value of information.
British Journal of Political Science 37 (4): 573-86.
Banks, Jeffrey S. 1991. Signaling games in political science. Langhorne, PA: Harwood.
Blomberg, S. Brock, Rozlyn C. Engel, and Reid Sawyer. 2010. On the duration and sustainabil-
ity of transnational terrorist organizations. Journal of Conflict Resolution 54(2): 303-30.
Enders Walter, and Todd Sandler. 2006. The political economy of terrorism. Cambridge, UK:
Cambridge University Press.
Faria, Joao, and Daniel G. Arce. 2005. Terrorism support and recruitment. Defence and Peace
Economics 16 (4): 263-73.
Franck, Raymond E., and Francois Melese. 2004. Exploring the structure of terrorists WMD
decisions: A game theory approach. Defense & Security Analysis 20 (4): 355-72.
Frey, Bruno, and Simon Luechinger. 2003. How to fight terrorism: Alternatives to deterrence.
Defence and Peace Economics 14 (4): 237-49.
Hoffman, Bruce, and Gordon H. McCormick. 2004. Terrorism, signaling, and suicide attack.
Studies in Conflict & Terrorism 27 (4): 243-81.
Jacobsen, Daniel, and Edward H. Kaplan. 2007. Suicide bombings and targeted killings in
(counter-)terror games. Journal of Conflict Resolution 51(5):772-92.
Jones, Seth G., and Martin C. Lubicki. 2008. How terrorist groups end. Santa Monica, CA:
RAND.
Kydd, Andrew, and Barbara Walter. 2006. Strategies of terrorism. International Security 31 (1):
49-80.
Lapan, Harvey E., and Todd Sandler. 1993. Terrorism and signalling. European Journal of
Political Economy 9 (3): 383-97.
Overgaard, Per B. 1994. The scale of terrorist attacks as a signal of resources. Journal of Con-
flict Resolution 38(3):452-78.
Rapoport, David C. 1992. Terrorism. In Routledge encyclopedia of government and politics,
vol. 2, ed. Mary Hawkesworth and Maurice Kogan, 1067-70. London: Routledge.
Rosendorff, B. Peter, and Todd Sandler. 2004. Too much of a good thing? The proactive
response dilemma. Journal of Conflict Resolution 48(5):657-71.
Safire, William. 1995. Spookspeak in Deutchland. New York Times, November 19, SM540.
Sandler, Todd, and Daniel G. Arce. 2003. Terrorism and game theory. Simulation & Gaming 34
(3): 319-37.
372 Journal of Conflict Resolution 54(2)
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from
Sandler, Todd, and Daniel G. Arce. 2007. Terrorism: A game-theoretic approach. In Handbook
of defense economics, vol. 2, ed. Todd Sandler and Keith Hartley, 775-813. Amsterdam:
North Holland.
Sandler, Todd, Daniel G. Arce, and Walter Enders. 2009. An evaluation of INTERPOLs
cooperative-based counterterrorism linkages. Center for Global Collective Action working
paper, University of Texas at Dallas.
Scheuer, Michael. 2006. Through our enemies eyes. Rev. ed. Washington, DC: Potomac
Books.
Siqueira, Kevin, and Todd Sandler. 2007. Terrorist backlash, terrorism mitigation, and policy
delegation. Journal of Public Economics 91 (9): 1800-1815.
Wright, Lawrence. 2006. The looming tower. New York: Knopf.
Arce and Sandler 373
at University of New South Wales on August 15, 2010 jcr.sagepub.com Downloaded from