Beruflich Dokumente
Kultur Dokumente
Journal of Conflict Resolution, Vol. 51, No. 1 (Feb., 2007), pp. 33-57 Published by: Sage Publications, Inc. Stable URL: http://www.jstor.org/stable/27638537 . Accessed: 04/12/2012 11:03
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal of Conflict Resolution.
http://www.jstor.org
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
Rational
Journal of Conflict Resolution Volume 5 i Number 1 February 2007 33-57 ? 2007 Sage Publications 10.1177/0022002706296155 http://jcr.sagepub.com hosted at http://online.sagepub.com
more proactive in that the to directly attack terrorist net work structures. We argue that rational terrorists will attempt to thwart such policies and restructure themselves to be less penetrable. We model the trade-off between and communication faced by terrorists. The model is used to security intragroup counterterrorism far more became Patriot Act allowed the authorities freedom derive type, complexity, Keywords: the anticipated changes and success terrorist cells; in network structure and rate of potential structure; the consequent terrorist attacks. changes in the
the events
of 9/11, U.S.
network
counterterrorism;
terror networks
the events of 9/11, U.S. counterterrorism policy switched from being almost purely defensive toward being more proactive. As detailed in Enders After and Sandier (2006), the U.S. Patriot Act (HR 3162) allowed the authorities far greater latitude in surveillance and intelligence gathering than had ever been thought possible. For our purposes, it is noteworthy thatmany features of the Patriot Act were designed to launch a direct assault on terrorist networks by moni (1) expanded authority to interceptwire, oral, and electronic communications of suspected terrorists; (2) allowed for shared criminal investigative findings among foreign and domestic law enforcement agencies; (3) eased the restrictions on foreign intelligence gathering inside the United
seizures; (5) augmented the access toring their communications. Particularly controversial was the fact that the act
States;
message
of domestic
law-enforcement
to
Foreign Intelligence Surveillance Act records; and (6) reduced congressional over sight on intelligence gathering. The current firestorm surrounding the National
Security
Agency's
monitoring
of the communications
of suspected
terrorists
shows
against
laundering
enhanced
the government's
ability
to track
the
interna
33
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
34
Itwould be naive to think that terroristsand theirnetworks would remain invar iant tomeasures designed to track and infiltratethe innerworkings of theirorgani
zations. Hamas For was example, altered most observers to the believe Israeli that the organizational against structure the group. of in response crackdown After
lized, hierarchical leadership structure. In its place, the leadership developed a compartmented organizational structure such that the individual cells operate under the directorship of local commanders.1 Although it is hard to put toomuch
credence secretly from each other. The cells are connected to the extent that they operate
stan. This is in accord with the type of "swarming" behavior discussed inArquilla, Ronfeldt, and Zanini (1999). Although there is no precise definition of the term
swarming, the connotation is one
chises, it is clear that the network has become far less hierarchical and far less structuredas a result of thePatriot Act and the ouster of theTaliban fromAfghani
in the popular
view
affiliates
act
like MacDonald's
fran
radio network (see Wilkinson [2005]), Paul Wilkinson of St. Andrew's University argues that the group responsible for the London metro bombings is loosely orga nized. Consider the following:
Colvin: But does thatmean that Osama bin Laden would have given the order [regard or is it a new al-Qaeda almost like a which operates ing the London bombings], more sort of McDonalds with much of a franchise, people working independently large organisation? Wilkinson: I think that bin Laden of the Taliban tions. That has only rarely been in a position since the removal to actually coordinate the detail of any opera regime in Afghanistan is left to the sections of the worldwide network that are involved in that in accord with the wishes
terrorist network would be one with a nonhierarchical deployment of small terrorist subgroups communicating with each other. No one stingwould be critical, but the sum total of all attacks could be fatal. Rodriguez (2004) reports thatSpain's March 11 terrorist network is based on weak ties in the sense that the structure is not hier archical and thatmost of the links are not based on intense (such as familial) relationships.2 In a July 12, 2005, interviewwith Mark Colvin on Australia's ABC
of groups
of killer
bees
on
the attack.
swarming
.But area... know that this was particular they would and strategic objectives movement. of the al-Qaeda ... if bin Laden or somebody Colvin: So like him ... has come ... no chatter was other? picked up? Are through to each Wilkinson: nications measures Well, and
I think that they've always been very clever at hiding their commu as they've had to adapt to the severe they've got more sophisticated community. So what we've seen
is the morphing
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
35
Because
terrorists
act
in secret,
most
of
the evidence
tions of theirnetwork structures is anecdotal. Yet this need for secrecy is precisely
the reason works. that terrorist networks are organized setting, differently network from other design can be types of net on In a social, familial, or business focused
concerning
these
adapta
the efficiency and robustness of the information flow. Information flows are most likely to be unfetteredwhen all individuals are well connected. However, the ille gal nature of a terroristgroup dictates that its network design balance the trade-off
between information flow and group
The aim of this article is to model the process by which terrorists select between the competing ends of security versus the unbridled flow of information.
Obviously, counterterrorism
security.
work is bound to be inferior to one based on the recognition a dynamic group struc
ture. This is not
policy
based
on
of a static
group
net
structures.Articles such as Arquilla, Ronfeldt, and Zanini (1999), Carley et al. (2003), and Rodriguez (2004) all discuss the possibility that the structuresof terror
ist groups evolve over time. However, structure Specifically, a general climate we attempt
to say
that we
are
the first
changing
terror network
to place
the choice
within
the types
intimidation
counterterrorismpolicies
induce groups
bombings)
are
subject
to infiltration.
Transference
can
because
to substitute
approach to the study of terroristoptimizing decisions. The key insight of the approach is that rational terrorists optimize theirbehavior so as to bring about their
most works. tance, preferred The and outcomes. Also approach among discussed uses graph is the structural structural reachability to terrorist net approach to the linkages, theory analyze impor in a terrorist organization. participants such as al-Qaeda. benefits It is also
the various
We
cuss
sible to show how these adaptations will alter the types of incidents orchestrated by
the group. munication In the third links within section, a we focus the costs The and of forming terrorist network. complete choice-theoretic
is developed
will adapt
its network
in response
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
36
Models
toCombat Terrorism
There are two primary types of models of terrorism thathave been used to for
mulate anti-terrorism a terrorist group by assumes the model The rational-actor policy. are the result of an economic that a terrorist group uses model posits that the choices made process. Specifically, optimization resources to maximize its its scarce
expected utility. In contrast, the structural approach attempts to find and connect the links among the participants in the network. A graphical depiction of all such linkages is called a "sociogram." Individuals with many links to others can be
deemed have important their potential and targets weaknesses. for counterterrorist Typically, actions. Both models approaches do not strengths rational-actor
address the issue of how different individual preferences can be aggregated into a group decision, how information flows within the group, or how the individual
group issue members of how form and why a consensus. links The structuralist and maintained. approach Since never addresses are the static, that are formed sociograms
they are unable to indicate how the group's structure is likely to respond to changes
in its environment.
terrorists might protect the network from infiltrationand ways to respond technological developments such as the Internet. The Rational-Actor Approach
Combining
allows
us
to model
the ways
that theymight
The basic premise of the rational-actor approach is that the utility of a terrorist group is derived from a shared political goal. The shared goal might be the elimina tion of a grievance stemming from income inequality, ideological differences, his
torical
discussed inEnders and Sandier (1993), this shared goal can be obtained from the
consumption instability, fear and of a number popular intimidation. financial Given support The assets, of basic for commodities cause, group and and has such as media attention, of an set of political of resources, their the creation access atmosphere
inequities,
or
lack
of
political,
economic,
or
religious
freedom.
As
terrorist weapons
including abilities.
buildings, terrorist
entrepreneurial
its resources,
a rational
group
selects
that maximizes the expectation of its attaining the shared goal. The choices made by the group will be influenced by the relative prices of the basic commodities. The full price of any particular attack mode includes the value
of the resources of deaths Certain used and to plan casualties, and execute the attack, the loss of group members because ity breach. and the costs of defending to expose the group the group's from a secur membership
attack modes
are more
to capture and infiltrationthan others. For example, the price of a suicide bombing includes the direct costs of the bomb, the costs of grooming the perpetrator to
ensure that the attack takes place, and the cost of protecting the group's security
likely
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
37
against a failed attack. On the other hand, threatsand hoaxes typically require few will incorporate all of inputs and personnel so that costs are low. Rational terrorists
these costs into their calculations when
The key feature of any counterterrorismpolicy is that it can influence the prices, resource supplies, and the payoffs faced by terrorists.Enhanced airport security increases the logistical complexity of a skyjacking and raises its price. If, at the
same
selecting
their optimal
tactics.
contraband become relatively cheaper. Similarly, if immigration officials make it more difficult for terroriststo enter theUnited States, a terrorist group might attack U.S. interests located abroad (for example, tourists and firms). Enders and Sandier model as follows: (1993, 2004) summarize two of thekey propositions of the
1: An increase Proposition the terrorist group to substitute are now relatively less expensive. in the relative price of one type of terrorist activity will cause out of the relatively expensive activity and into activities that
time, governments
do not
increase
security
at ports
of entry,
attacks
relying
on
The direct implication is that an increase in the costs of maintaining intragroup connections will induce the group to substitute away from a tightlyorganized net
work structure toward 2: An normal a more loosely organized form. set will it to produce and
increase goods
induce
If logistically complex events are normal goods while logistically simple events are inferiorgoods, improvements in the technology available to the group should
increase threats ple the occurrence and letter bombs. connectivity of complex Since will tend events and reduce require as the number more of events such than as complex events coordination available sim
events,
to increase
the technology
to terror
These propositions have been used to explain many of the substitutions that
seem to make in response to changing to various some circumstances. Notice, the group. however, Unlike have the no of
that the approach never directly addresses the issue of how the group's decisions are
made model direct and how of a way they are passed along consumer or firm, single with various between may members of For necessarily of many a of members terrorist group As
security
reasons,
guarded.
some
know
the identities
The Structuralist
The alternative
Approach:
way of viewing
Sociograms
terrorists is to map the structure of their organiza
As summarized by Farley (2003), each indivi tions using network or graph theory. or terroristcell within an organization can be represented by a point dual terrorist
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
38
called sent
a node. a direct
Lines,
sometimes
connecting of
repre net
communications
work. Even if two nodes are not directly connected, it is still possible for them to communicate through the use ofmultiple links.The density of a network is defined as the number of links in the network divided by the maximum number of all possi which information travels througha network ble links.Typically, the speed through
is an links function increasing so that information of the density can quickly of the network. from one Dense networks have many travel node to any or all of the others.
subcomponents
Figure 1 contains Krebs' (2001) sociogram of the nineteen 9/11 hijackers. The key insight of the structuralistapproach is that the effectiveness of the network can Nawaf al Hazmi is be eroded by breaking the links between the nodes. Notice that a well-connected to such links. with six the connected network Eliminating directly individual is likely to undermine the effectiveness of the entire organization. Since Abdulaziz al Omari is a bridge to three of the network's members (Satam al Waleed andWail al Shehri), eliminating him would leave Suqami and thebrothers these three hijackers disconnected. In contrast, Ahmed al Ghamdi and Majed
Moqed are not well connected; might as want such, if the network structure is known, the coun terterrorism authorities to target other network members.
Figure 2 shows the basic schematics for the two fundamental network types? the star and the chain. For simplicity,we describe thepure forms of each type using only five nodes. Note that the nodes can represent individuals or groups of indivi
duals that we refer to as cells. For
expositional
purposes,
we
suppose
that there
is a
select the structure that best suits the types of operations it intends to carry out
its available resources.
disadvantages
for a terrorist
terrorist
group
given
The chain structure is clearly sequential in that communications flow from the
to nodes 2 through 5 in a predetermined order. In contrast, each node has a
leader
direct tie to the leadership in the star structure.As such, all communications flow from node 1 directly to the individual nodes. Notice that this structure allows the leadership to simultaneously coordinate the behavior of the individual cells, but the individual nodes do not communicate directly with each other. Since communica tion links to the leadership are all direct, this structure is not especially secure, since every node has the possibility of providing useful information about the loca tion of the leadership. The strengthof the star structure is that it is less vulnerable
to logistical bers of failure than nodes the chain can structure. If, say, out node 3 is compromised, mission. mem To take the other continue to carry their assigned
a specific example, it is possible to suppose that nodes 2 through 5 represent the four hijacked planes of 9/11. Even thoughUnited Airlines flight 93 crashed near
Shanksville, Pennsylvania, the remaining planes were tragically successful.
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
39
Sale
?+\Naw?fai Hazmi
Ahmed Ghamdi ai ., ais
;*
:Saeed
x? Khaiidai Mihdhar
ai Ghamdi
"x:
Ahmed
a! Haznawi
.-* Ziad
Jarrah
Shehhi
;
/ Mohamed Atta
Satam
ai Suqami
Source: Krebs
Arquilla, Ronfeldt, and Zanini (1999) suggest that the chain structure is secure so that it is especially well suited for smuggling types of operations. Viewers of the classic film The Battle of Algiers will recall Colonel Mathieu's famous characteri zation of the purported chain formation of theFront de Lib?ration Nationale:3
The military head of the executive bureau finds a competent person and names him . no 1.No. 1 finds others: nos. 2 and 3... Now 2 and 3 each select two men: nos. 4, 5, 6, and 7. The reason only three other members: In point of fact, they don't is that each for these geometries the one who chose him and know each other. knows member organization . the two he himself chose...
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
40
Panel
1 :Star Pattern
that node 3 is compromised as a result of antiterrorism efforts. Since 1 and 2 no longer have a direct or indirect communication linkwith 4 and 5, these two nodes
are cut off from
at any
of
the communications
junctures.
Suppose
and/or 5 will lead to a logistical failure. On the other hand, the leadership is secure, since 3 cannot inform the authorities about the location of the leadership. More
over, the chain
the remaining
members
of the network.
Any
operations
requiring
of accession should the leadership be compromised. For example, after 9/11, the Bush administration targeted the top leadership of the al-Qaeda network by placing special emphasis on Osama bin Laden and Ayman al Zawahiri. If al-Qaeda were
arranged no natural chain as a star network, successor. the successful the chain execution pattern In contrast, leave of the plan would a clear successor suggests it with in the
pattern
has
greater
long-run
viability
in that there
is a natural
order
of command.
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
41
Panels 3 and 4 indicate the effects of allowing an additional linkage between two of the nodes. Begin with a basic chain structureand allow nodes 2 and 5 to commu mission to nicate directly with each other. The additional linkage allows the terror be more logistically sophisticated in that2 and 5 are able to directly coordinate their actions. The possibility of logistical failure is diminished in that 5 may not be iso lated if the network is infiltrated.If node 3 is compromised, 2 can relay any impor tant information directly to 5. The disadvantage of the additional link is that the network becomes less secure in that2 and 5 now have informationabout each other.
Similar remarks can be made about the star structure. The additional linkage
allows the terror mission to be more logistically sophisticated in that 2 and 5 are able to directly coordinate their actions. However, allowing 2 and 5 to directly interact compromises the security of the network. After all, ifnode 5 is infiltrated, it could compromise the security of 1 and/or 2.
Before grams for proceeding, formulating we a want to emphasize the policy. limitations The of the use of socio is to counterterrorism standard procedure
collect information about the key individuals in the network and the links connect ing them.The aim is to identifyand remove those individuals who are important in the social network. Importance is typically identifiedby the number of links to the
nodes. Thus,
other
rather than on the role that the individual plays in the system.Consider the follow than point j if ithas more direct links to other points" and that ".. .more important
persons influence less important persons." Nevertheless, the number of as links the authors does not themselves necessarily acknowledge, in many circumstances, ing statement of Brams, Mutlu, and Ramirez (2006): ".. .point / is more important
importance
is measured
by
how
many
individuals
one
knows
correspond to importance. For example, in panel 4 of Figure 2, the leader is the least connected node even though it influences all nodes and no node has a direct influence over the leader. The four pilots of 9/11 (Mohamed Atta, Marwan al Shehhi, Hani Hanjour, and Ziad Jarrah) are neither importantnor weakly connected in Krebs's sociogram. The issue is that a sociogram does not distinguish indivi duals in terms of their function. In addition, articles such as Farley (2003) and Rodriguez (2004) argue that it is the weakly-linked individuals that provide the most opportunity for successful infiltration. After all, eliminating an individual
such over, as Abdulaziz there al Omari, since who serves as a bridge, accurately may segment the group. actually can be More the is always structure, the issue of how the authorities operations know
network's
Clearly, incomplete information about the network can result in missing nodes and/or links. and the Security-Communication Trade-Off
government
intelligence
faulty.
Density
The trade-offbetween network density and information flows is illustrated in Figure 3. Notice that each panel of the figure depicts a nonhierarchical network
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
42
is directly connected to each other node. The density of the group (p) is unity, since
each node is connected
maximized,
number sions. Note
to every
other
node.
Not
only
are
the communication
flows
of subgroups
threemembers
ways, and two
in six different
ways.4 Panel 2
shows the effects of removing a link. Since the group is not hierarchical, itdoes not
which of the six links is removed; all five-link patterns are necessarily trans
matter
formations of each other.With only five links (p = 5/6), group flexibility and communication become more difficult, since there is not a direct path from 1 to 4. Resiliency and flexibility decline, since it is impossible to have a team of four
members only member because nodes to form a 1 and 4 are not directly team, only connected. are only Similarly, five ways two groups to connect there are a two ways teams.5 three-member and 3a and there 3b are show there to form of two
team with
two ways
to form ways
= = using four links (p 4/6 2/3). If you examine these two panels, it should be clear that communication and flexibility are furtherdiminished by the elimination of the second link. The point ismade even more forcefully in panels 4a and 4b, wherein there are only three links connecting the nodes (p = 3/6 = 1/2). Note the chain in panel 4a and the star of panel 4b are the only two possible structureswith
three nication links are among four nodes. linear Even in panel though 4a, there and is no leader, the lines must of commu go through necessarily all communications
four nodes
less flexible,
most links. If the antiterrorismauthorities are able to compromise because ithas the
any
work is placed in jeopardy. In panel 2, only nodes 2 and 3 place the entire network at risk, since 1 and 4 are not directly linked. Ifwe use a modification of the argu a ment made by Farley (2003), suppose that the antiterrorismauthority can infiltrate single node with a probability of success equal to p?. Further suppose that the agency is unsure of the network structure so that it is equally likely to attempt to
infiltrate
in panel
1 by means
such
as
infiltration,
torture,
or
bribery,
the entire
net
any one particular node is 0.25). Thus, in panel 1, the probabil attempts to infiltrate ity that the entire network is brought down is p?, regardless of which node is infil trated. If eliminating any one node has a payoff of 1, the expected cost to the terrorists of infiltrationin panel 1 is p)A. In panel 2, there is a 0.5 chance that the
any
one
particular
node
(i.e., we
assume
that authority
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
43
Panel 1: p = 1
Panel 2: p = 5/6 1
node 2 or 3 and a 0.5 chance that it attempts to infil authority attempts to infiltrate tratenode 1 or 4. Thus, the probability that the entire network is brought down is p*0.5, and the probability that threeof the nodes are eliminated is also p*0.5. In this of infiltrationis p* (0.5*4 + 0.5*3) = /?*3.5, case, the expected cost to the terrorists
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
44
of any one which is less than that in panel 1. In panel 3a, the successful infiltration cost of infiltra node leads to the elimination of three terrorists. the Thus, expected tion is p*3. In panel 3b, the successful infiltrationof nodes 1 and 3 leads to the elimination of threenodes, the successful infiltration of node 2 leads to the elimina tion of the entire network, and the elimination of node 4 leads to the elimination of two nodes. As such, the expected cost of infiltration is p*(0.5*3 + 0.25*4+ = p*3. Finally, the expected costs of infiltrationcan be shown to be pj 0.25*2) = = + p*2.5 in panel 4b. p*2.5 in panel 4a and p) (0.75*2 + 0.25*4) (0.5*2 0.5*3) A concept related to a network's density is the reachability or detectability of a node is likely to be specific individuals within the network. Given an infiltration, detectable if there are many direct and indirect links to that node. Suppose that if of ability of 0.5. Hence, returning to panel 1 of Figure 2, the successful infiltration node 1 reveals the identityof node 1with probability 1, and the successful infiltra tion of nodes 2, 3, 4, or 5 reveals the identity of node 1 with probability 0.5. As such, the reachability of node 1 is given by 1 + 4(0.5) = 3. As such,with five identi will reach cal individuals in the network, the probability a successful infiltration node 1 is p*3/5. Node 2 can be reached by direct infiltration and indirectlyby infil of node 1 reveals 2 with probability 0.5, tratingnode 1, 3, 4, or 5. The infiltration and the infiltrationof any of nodes 3, 4, or 5 reveals the identityof node 2 with = 0.25. Hence, the reachability of node 2 is given by 1 + probability of (0.5)(0.5) = 2.25. It should be clear that the + 3(0.25) (0.5) reachability of nodes 3, 4, and 5 is
precisely detectable the same in the as that of node 2. Hence, network. Nevertheless, there in the star pattern, node is not a one-to-one 1 is the most relationship successfully infiltrated, each node reveals a directly connected node with a prob
between density and reachability. To illustrate the point, note that the density of the network in panel 2 is identical to that of panel 1 but the reachability of the var ious individuals differs across the panels. In panel 2, the reachability of node 1 is 1+ 0.5 + (0.5)2 + (0.5)3 + (0.5)4 = 1.9375 in that it can be reached directly and through routes passing along (2, 1), (3, 2, 1), (4, 3, 2, 1) and (5, 4, 3, 2, 1). By using similar calculations, the reachability of nodes 2, 3, 4, and 5 can be shown to be
2.375, 2.5, 2.375, and 1.9375,
in the star pattern have different degrees of reachability than those in the chain. However, for our purposes, the distinction between density and reachability is not especially important because, for a given structure, adding an additional link
respectively.
Thus,
density,
individuals
increases both the network's density and the reachability of the nodes. Even though reachability can be useful for cases inwhich the counterterrorism
has detailed is not knowledge of the network's the point structure, in most cases, such that, information available. Nevertheless, of the exercise is to argue
authority
all else being equal, terrorists would want tominimize the number of links if their sole goal was tominimize the expected damage to the network. If the network is hierarchical, the network would want to protect important individuals by reducing theirreachability.
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
45
Combining
The advantages. rorist group A
rational-actor strength
will
cussed above, there is little recognition of the communication flows within the organization itself. In contrast, the structuralistapproach highlights the linkswithin
an
optimize
dimensions
the counterterrorist authorities target the lines of communication within the network, terrorists will find that the relative price of links has increased. As links become less
secure mize on (or are expected to become less secure), a rational group will want As to econo such, coun the number of communication links within the organization.
organization,
but
it usually
takes
as given.
However,
once
terterrorism policy will alter the form of the organization itself.Not only will the group's density decline, but the group will find that the relative price of tactics that
are intensive attack, in communication a suicide bombing links has increased. secure In contrast since to a coordinated it involves few armed is a very attack mode
means to trace the successful bomber back individuals, and the authorities have little to thenetwork. In the same vein,Merrari (1999) argues that terrorist groups adopting
loosely connected
ing to limit theirability to conduct logistically complex activities such as the acquisi tion and use of chemical, biological, radioactive, or nuclear (CBRN) weapons. A
degree of centralization allows a group to acquire and coordinate the resources
organizational
structures
to minimize
infiltration
risk are
also
act
and personnel necessary to conduct this type of logistically sophisticated attack. Similarly, Enders and Sandier (2006) argue that al-Qaeda's decentralized structure protected itduring the post-9/11 attacks but at the price of not being able to develop CBRN weapons. In contrast,Aum Shinrikyo was a highly centralized group. Until the recent spate of anthrax attacks immediately following 9/11, it was the only terrorist group using a CRBN weapon. However, the group stayed centralized and sustained an organizationwide setback once theirheadquarters were raided in 1995.
high
nation of the two. The precise pattern will depend on the various types of attacks the group plans to undertake and on othermicrofactors such as kinship, childhood Before proceeding, some preliminary relationships need to be developed. Sup pose there are N > 2 members in the group. It is easy to show that themaximum
friendship, who goes to the same training camp as whom, and so forth.6
the group
chooses
a star pattern,
a chain,
or a
complex
combi
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
46
? number of links connecting the members is N(N l)/2.7 Also suppose that the actual number of links is L. Since the density (p) is the number of links (L) divided by the maximum possible number of links, it follows that p = ? a mea l))/2) G [0,1]. In a sense, the density parameter p is summary L/((N(N surement of the complexity of the network structure in that it shows the degree of minimum connectivity among the cells. Notice that if the network is connected, the ? 1. Consequently, when p g [0,2/Af), the group necessarily number of links isN
falls
p g [2/Af,1] is itpossible that the entire group forms a connected network. and Infiltration Risk
apart
into
subgroups
that
are
disconnected
from
one
another;
only
when
Density
The essential feature of our model is to capture the relationship between a as the group's optimal density and the intensity of governmental efforts such Patriot Act to infiltratethe group and to curtail its activities. To model this process, we assume that the successful infiltrationof node j compromises j and all nodes directly connected to j. It should be clear that the greater the number of links, For example, in panel 1 of the greater the expected cost of any one infiltration. can cause the downfall of the a member the infiltration of 3, group single Figure of nodes 1 or 4 will cause only entire group. However, in panel 4a, the infiltration one othermember to be compromised, and the infiltration of 2 or 3 will cause only two others to be compromised. More generally, if node j is connected to k other of j is nodes, the number of individuals compromised from a successful infiltration of node j to 1+ k. In principle, it is also possible to allow a successful infiltration
compromise j, all k nodes connected to j, and with some probability less than unity, the individuals connected to this latter set of individuals. However, nothing of substance would be changed as long as the infiltrationof one node does not
cause the entire network to be compromised. network, In the circumstance there As is a corner such, we that a successful such infiltration agents are compromises connected the entire or all act solution on that all
individually.
focus
the network's
that the probability of a successful infiltration is given by p?. If the group is not
hierarchical and
single
node
and
able to assume that each node has an equal probability of being infiltrated. Hence, if there are TV group members, the probability that any particular node is infiltrated in l/N. Given the group's size and the number of links L, it is straightforward to show that the expected number of compromised nodes (/) from an attempted infil tration is8 I = p,(\+2L/N)
the authority
does
not know
the group
structure,
it seems
reason
(1)
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
47
Intuitively, if there are L links among TV people, the average person is connected to L/(N/2) ? 2L/N others. Hence, an infiltrationcompromises 1+ 2L/N others with probability p?. Since we want to express / as a function of the density p, it is
useful to rewrite the equation as
I = Pl[l + (N-l)p}.
(2)
For example, in panel 1 of Figure 3, p =1 and N ? I = 3, so thatany attempted infiltration would be expected to compromise 4 people with probability p?. Simi would be expected to compro larly, in panels 4a and 4b, an attempted infiltration = + mise 2.5 (2.5 1 3*0.5) people with probability p?. The point of the exercise is to show that the expected number of compromised individuals (/) is an increasing function of p?, N, and p. Although it is relatively straightforwardto extend the ana
where the counterterrorism lysis to situations authority we to conserve do not pursue that exercise. For space, tries our to infiltrate purposes, two nodes, point the key
is that the expected number of compromised individuals is increasing in the group's density; hence, the group can decrease its infiltrationrisk by reducing p. We can also use equation (2) to illustrateKrebs's (2001) sociogram of the 9/11 network shown in Figure 1. There are nineteen nodes (N = 19) and twenty-seven links = maximum possible number of links is 17*16/2 = 136, the value (L 27). Since the we can calcu of p = 27/137 = 0.19708. Although we do not know the value of /?/, late I/pj = 1+ (N - l)p as 1+ 18*0.19708 = 4.5474. Had there been a 10 per
cent chance of a successful infiltration, the
expected
number
of
compromised
of a Logistical
literature,
Failure
we divide the government's counterterrorism
policies into those that are primarily offensive and those that are primarily defen sive. As described in Enders and Sandier (2006), defensive policies Go represent
measures such as enhanced
the existing
airport
security,
embassy
fortification,
and
inspection
focus
the aforementioned
features of the Patriot Act, designed to infiltrate terroristnetwork structures.The distinction is important, since an increase in the government's infiltrationefforts should result in an increase in p?. In contrast, defensive policies have no direct
impact on the value of p?. that the terrorist group uses to plan, organize, and con Similarly, the technology
duct an attack can be divided into two categories. The terroristoffensive technol ogy To represents innovations inweaponry such as improvised explosive devices or CBRN weapons. The terrorist defensive technology Tp represents innovations in
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
48
such as the
the abil
ity of terrorists to form secure links. In our framework, such technological advances act to reduce the value of p?, while changes in the level of To have no
direct effect on p?.
improve
The point is that the probability of a successful infiltration p? depends on the of the and the terrorist defen offensive efforts G o magnitude government's policy sive technology TD. As such, we can model the probability of a successful infiltra tion as pi = pi(Go->TD), where the partial derivative with respect to the first argument is positive, while thatwith respect to the second argument is negative.
However, the value of p? to keep the notation so that simple, we let A denote any autonomous increase in
At this point, we can model the probability of a failed attack p. Besides the extent of infiltration/, a logistical failuremight be because of enhanced defensive counterterrorism efforts Go- As such, we let the probability of a failed attack be increasing inGo and /: = p = p(GD, /) p{GD, p/(A)[l + (Nl)p]) (4) ? l)p].
where we have used (2) and (3) so thatwe can express / as pi(A)[l+(N Consider the function9 p = p(A,GD,p,N) From the discussion above, it follows that:
Assumption 1: px > 0, p2 > 0, p3 > 0; p\3 > 0, P23 = 0, p-& = 0.
(5)
that defensive policies combat terrorismby increasing the value of Gd, the types of offensive policies we consider act to combat terrorism by increasing A. Both types of policies work by increasing the probability of a failed attack so thatp\ > 0 and pi > 0. An increase in density increases the expected loss of an attempted infiltrationso thatp3 > 0. Articles by Bueno deMesquita (2005) and Faria and Arce (2005) provide formal models of the recruitmentprocess (i.e., the selection of the optimal value ofN). In while
contrast, we treat N as a fixed factor
Note
determinants of the group's optimal density. This is not to say the density and size are independent of each other; instead, we want to refrain frommaking specific extension of the model
assumptions concerning the partial effects of N on p. Of course, a
of production
so
that we
can
focus
on
other
straightforward
and p.
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
49
we aim
could to reduce
also
generalize
to allow personnel.
for proactive
policies
that
the terrorists'
Now
second-order
the pro
a techno group cross
derivative between Go and p is 0, because the government's defensive efforts do nothing to directly facilitate or impede the information flow within the terrorist
network structure. And last, the second-order derivative of p is 0 (i.e, 7733 = 0)
security
breach
severe.
because, from equation (2), the infiltration probability is linear in p. The Terrorists'
Terrorists However,
Production
basic
Technology
commodities tractable using and a number to focus on of factors connectivity, of production. we assume
produce to keep
that the output of the basic commodity F?say represented by: F = F(T0,N,p)
Assumption 2: F\ > 0, F3 > 0, F13 > 0, F33 < 0.
the problem
be
(6)
Note thatF\ > 0 since the expression To represents technological change in the production of thebasic commodity. For example, the development of plastic explo stant the other factors of production.With fixed technology,more links within the network imply better informationflow and better coordination among themembers, so that marginal product of density on the output of fear is positive (i.e., F3 > 0). The positive cross-derivative between To and p (i.e., F13 > 0) implies that better additional links.And it is a conventional assumption that themarginal benefit of an additional link diminishes as there aremore and more existing links; hence, F33 < 0. The Cost Function The cost function faced by the terroristorganization depends only on the num ber of people in the organization and the density of the organization. Hence, we posit a cost function of the form: C = C{N,p) (7)
Assumption 3: C2 > 0, C22 > 0. technology to plan and organize attacks enhances the marginal value of forming sives might mean that terrorists can create more intimidation and fear holding con
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
50
Given that it takes additional effort to establish and maintain an additional link, we make the assumption that themarginal cost of increasing density is positive > 0). Also as the number of existing links increases, it ismore and more costly (C2 to build an additional link so thatC22 > 0.10
Determinants of theDensity
The aim of the group is tomaximize the expected output of the basic commod a terrorist group chooses the ity, (1 ?p)F, minus costs, C. With fixed group size TV, to solve the maximization problem: optimal group density p following max[l p where the expression ?
is fixed or predetermined.11
?, p)
C(?, p) (8)
mal density p. Since the only choice variable is p, the arguments of a function are omitted when there is no risk of confusion. The first-ordercondition on p (assum ing an interiorsolution) is
(l-/?)F3-p3F-C2 = 0. (9)
The first term, (1 ?p)F^, is positive and represents the expected marginal pro duct of an additional link. The second and third terms are negative and represent themarginal cost of an additional link,p^ F represents themarginal cost of a link
because
marginal cost at the optimum, maintaining the link.The marginal benefit equals the
so that the terrorist ber of links. group has no incentive either to increase condition or to decrease for maximization the num is also In this problem, the second-order
of a security
breach,
and C2
represents
the marginal
cost
of setting
up
and
satisfied:
(l-p)F33-2p3F3-C22<0.
(10)
Given assumptions 1, 2, and 3, the optimal density (p*) is the solution to (9) and is unique. Since this optimal value is implicitly a function of A, Go, and To, it is straightforward to derive the determinants of p* and the consequent changes on p and F using the implicit function theorem (IFT). Counterterrorism: Infiltration Effort
will substitute out of One important result of themodel is that rational terrorists as infiltration risk increases. activities Similarly, if Internet connec high-density tions become more secure (so thatTD increases), the group should seek to add links to take advantage of the lower risk. In fact, it is straightforward to derive these
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
51
predictions. If we apply the IFT to (9) and take the total derivative with respect to
A, we have
((1
p)F33
2p3F3
C22){dp*/dA)
Since the left-hand-side (LHS) coefficient is negative while the right-hand-side (RHS) is positive, it immediately follows thatdp*/dA < 0. Since A is increasing in Go and decreasing in To, the group's density will fall as a result of increased infil
tration efforts and rise as a result of
extreme, counterterrorism efforts could split the network. If the optimal density falls such that p* < 2/N, the network would find itmore efficient to work in Martyrs' completely separate subgroups than in a single large group. Like the Jenin Brigade formed as a breakaway group fromHamas after an Israeli raid on a Palesti nian refugee camp in 2003, the breakaway group could act autonomously from the larger group.12 Alternatively, the original network could simply be an umbrella
group for the resultant
improved
communications
technology.
In the
subgroups.
Go
technologies
have
occurred
simultaneously it would
(i.e., be
both
if theMacDonald's
the case
franchise
that the effortsof the counterterrorism authorities have kept ahead of the terrorists' adoption of Internet technologies. The substitution toward a lower density implies a substitution from logistically complex attacks to logistically simple ones. All else being equal, logistically com plex incidents should further the terrorist's cause more than simple incidents. Yet
because many factors are
matic consequences. For example, there is little doubt that theMarch 11, 2004, Madrid commuter train bombings had a larger total impact than themore coordi
nated Salaam, simultaneous Tanzania, attacks on August on the U.S. 7, 1998. embassies Nevertheless, in Nairobi, Kenya, and Dar es to the extent that the density
in play,
some
relatively
simple
incidents
can
have
dra
of al-Qaeda has decreased, themodel predicts a substitution toward events like the Madrid bombings and away fromhighly sophisticated and coordinated attacks. One interesting result is that enhanced infiltrationactivities may actually result
in the terrorists
terroristssubstitute out of complex attacks with a lower probability of success and into simpler attacks that are more likely to succeed. To obtain this result, take the differentialof p = p(A1Go, p)' dp/dA = px + p3{dp*/dA). (12)
adopting
attack
modes
with
higher
success
rates.
In a
sense,
the
is negative, Since the first termp\ is positive and the second termpi(dp*?dA) it follows that at the optimal network structure,dp/dA may be positive, zero, or negative. Simply put, more intense proactive policies to fight terrorismneed not
necessarily make it harder to carry out an attack successfully. On one hand, the
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
52
each link.On the other hand, the government becomes more efficient at infiltrating terrorism group becomes more discreet, establishes fewer links and switches to The net effect depends on the logistically simple attacks thatare harder to infiltrate. comparison of the two effects.Of course, it also follows thatan increase in To will
have
Although the effect of A on the probability of a logistical failure p is ambigu ous, the overall impact of an increase inA is to decrease the expected output of the basic commodity. To obtain this result, take the total derivative of expected output (1 ?p)F with respect toA: d(\ p)F/dA = -PxF + {-p3F + (1 ? ? p)F3)(dp*/dA). = (13) ?
an ambiguous
effect
on p.
From the first-ordercondition (1 p)F3 0, it follows that C2 p3F = -PlF + C2(dp*/dA)<0. (14) d(\ p)F/dA Thus, even though dp/dA is ambiguous, enhanced infiltrationdoes effectively reduce the expected output of the group. Counterterrorism: Defensive Policies
The signs of the comparative statics of an increase in defensive policies are identical to those of enhanced infiltration policies. In particular, dp* /dGo < 0, and at the optimal p*, dp/dGo is ambiguous while d(l ? p)F/dGo < 0. Ifwe apply the IFT to (9) and take the total derivative with respect to Go, we obtain ((1 Since the LHS _ p)p33 _ 2p3F3 C22)(dp*/dGD) = p2F3. (15)
dp*/dGD<0. Even though the government's defensive technology has no direct impact on the terrorist network structure, it has an indirect effect. Since p\ > 0, the direct effect of a defensive policy is to increase the probability of a logistical failure. The group will find it advantageous to offset this by reducing its density; the reduced density acts tomake the group more secure and to reduce the likelihood of a failed attack. It is easy to check thatdp/dGo = Pi 4- p?>{dp* ?dGjf) may be positive, zero, or efforts (i.e., increased A), observing negative. As in the case of increased infiltration
more successful attacks
following
stepped-up
counterterrorism
policy
does
not
p)F/dGD < 0.
features in the chain of operation; yet both have similar effects in that theydecrease density and expected output of the basic commodity. Moreover, both have ambigu ous effects on the probability of a failed attack.
target
different
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
53
Terrorist Offensive
Technology
It is straightforward to show that the sign of dp*?dTo is ambiguous. Applying the IFT to (9) and taking the total derivative with respect to To, we obtain ((1 Although
are of opposite
p)F33
2p3F3
C22)(dp*/dT0)
P3Fi
(1
p)F]3.
(16)
dp*?dT0 has an ambiguous sign. This may seem counterintuitive at first glance; yet the ambiguity is perfectly consistent with the rational-actormodel (see proposi
tion
they can have more of the basic commodity without expending any of their resources. Unless it can determined whether density is normal, the sign dp* dTo is j nology and the density (Fi3 > 0) suggests that an increase in To will be accompa nied by an increase in p. However, To has a positive income effect in thatoutput of the basic commodity can be increased without an increase in p. As such, in the face of a positive technological change, it might be worthwhile for the terroriststo give up some of the basic commodity by reducing p in exchange for a higher probability
of a successful attack. The net effect ambiguous. Intuitively, the complementarity between the terrorists' offensive tech
2).
The
increase
in To
acts
as
a pure
increase
in the terrorists'
income,
since
we make the rather innocuous assumption However, the ambiguity disappears if that the technological change has theHicks-neutral form. Specifically, we assume that the output of the basic commodity is proportional to the magnitude of the tech nological change: F(TD,N,p) = TDf(NlP) (17)
depends
on
the comparison
of the two.
where Fi = f{N, p) > 0, F3 = TDf2 > 0, F13 = /2 > 0. ? ? ? Now the term p^F\ ? (1 ? p)F\-$ can be written as psf(N,p) (1 p)fi ? ? we know that p-$F? (\/TD)[p3F (1 p)F^\. From the first-order condition,
(1 ? p)F$ ? ? C2 < 0. Hence, the ambiguity disappears since we can write
P)F33
2p3F3
C22)(dp*/dT0)
-C2/TD (18)
technological
improvements
in weaponry
can
be
expected
to increase
net
has importantpolicy implications. Sparrow (1991), Farley (2003), and Carley et al. (2003) discuss methods to destabilize a terrorist network that rely on the
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
54
structuralist approach. The problem is that the sociogram for period t may not be the same as thatfor period t+ 1 ; this is especially true in the presence of increased
counterterrorism it can successfully efforts. As hope the structure also of the group change. By changes, combining the types of actions to undertake the structuralist
approach with an optimizing model of terrorism,it is possible to anticipate how the network will change in response to counterterrorism policies and to technological
innovations counter such as the Internet. at We argue and that rational restructure security and terrorists themselves intragroup will to be attempt less pene to increased model efforts the infiltration between
trable. We
trade-off
communication
changes
Although themodel is highly stylized, it does lead to a number of potentially importantpolicy implications. First, we do not expect to see the type of extensive swarming behavior envisioned by Arquilla, Ronfeldt, and Zanini (1999). As long as terroristsneed tomaintain secrecy, theywill act to limit theirdensity so as to reduce infiltration risk. While it is true thatdevelopments in communications tech
have enabled terrorist groups to maintain contact with members across great
nology can
Web distances, the need for clandestine behavior remains. Although Internet cit details about forthcoming attacks. Moreover,
has entered Second, an Internet in response groups such are chat room, and e-mail to counterterrorism predicted to reduce provide some useful information to group members, they cannot provide
sites
expli
munications, cated
activities they are able to conduct will necessarily be altered. Logistically compli
events, as the 9/11 attacks are reduce and the coordinated to decline grievances recent on the U.S. that there embassies infiltration in Kenya technologies and Tanzania, do not predicted Given
types
the terrorists'
or resources,
is a predicted substitution into less coordinated incident types. Clearly, the March
11 attacks ings were on the Madrid simple attacks commuter to plan trains and and more London subway bomb or relatively execute compared to the embassy attacks
the simultaneous
of 9/11.
Third, both defensive and infiltrationcounterterrorism policies may have the effect of actually increasing the probability of a successful attack (i.e., decreasing the value of 1? p). The direct effect of either type of counterterrorismpolicy is to
reduce
the probability
of a successful
attack.
However,
rational
terrorists
can
seek
extend themodel such that there are two explicit attack modes. The logistically complex mode (e.g., hostage taking) could be intensive in connectivity, while the other attack mode could be relatively simple (e.g., letterbombs) in that it requires
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
55
few links to be successful. In thisway, itwould be possible to trace out the direct
substitution of one attack mode for another. Second, since our aim was to focus on
connectivity and logistical complexity, we assumed that all factors of production in the terrorist group were fixed. The terroristssimply chose the value of p thatpro
vided
nectivity should be complementary with some factors and substitutes for others. Allowing for variable labor inputs as in Bueno de Mesquita (2005) and Faria and Arce (2005) would make it possible to trace out the effects of changing resource
levels ously on connectivity. and have Third, opposite in the model, effect on the values of Go and To as are exogen and given the parameter A. However, in Arce
the greatest
level
of fear
and
intimidation
in excess
of costs.
However,
con
Sandier (2005), the terroristsand the authorities could be allowed to play a game
against game each other such that each would selects allow theoretical framework policies. In a static its optimal strategy. us to the government's analyze a total resource constraint, facing setup, the optimal the gov the of
counterterrorism
ernment can select the optimal counterterrorism policy by changing the relative
emphasis oretical on proactive versus defensive us framework would allow In a dynamic policies. to study the cat-and-mouse setup, the game game innovation
In particular,
the type discussed in Enders and Sandier (2006) and Faria (2003). The nature of that game is such thata successful terroristinnovation (such as the use of the Inter net) puts pressure on the authorities to combat the terrorists (possibly by inter
cepting terrorists (such as e-mail adopt using a successful innovation After messages). a new strategy to offset the government's secure messaging). by the government, the counterterrorism strategy
Notes
1. Gambill Hamas 2. This bombings (2002) provides a detailed discussion of the ways changed in response to the Israeli counterterror initiative. so-called 3/11 network is the al-Qaeda in Madrid. that the organizational structure of
affiliate responsible
3. Translation from the Criterion Collection (minute 58) of the 1966 film The Battle of Algiers, and Saadi Yacef. The producer, Saadi directed by Gillo Pontecorov and produced by Antonio Musa in the Casbah and was arrested in 1957. Yacef, was the head of the Front de Lib?ration Nationale Granted amnesty, he collaborated with director by Gillo Pontecorvo and played himself in the film. 4. The four possible teams with threemembers each are (1, 2, 3), (1, 2, 4), (1, 3, 4), and (2, 3, 4). The six possible teams of twomembers each are (1, 2), (1, 3), (1, 4), (2, 3), (2, 3), and (3, 4). 5. The two possible teams with three members each are (1, 2, 3) and (2, 3, 4). The five possible teams of twomembers each are (1, 2), (1, 3), (2, 3), (2, 3), and (3, 4). 6. Since we are concerned about density but not the group's actual structure, it is possible to con
ceptualize a situation inwhich the group has a two-step optimization process. In the first stage, the group selects its density, and in the second stage, it selects the precise pattern of links among the individuals. 7. The hypergeometric distribution indicates that N items taken two at a time (without regard to = order) can be done inN\/[(N N{N l)/2 ways. 2)!2!] 8. Suppose there are N group members and L links joining them. Let n? denote the number of group members linked to exactly i other members. Hence, n\ denotes the number of members linked to only
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
56
and ni denotes the number of members N-l that N = ^2 n?- The total number of /=i ... + (N ?
+ + ^ [n\ 2n2
of nodes connected
to exactly ?others and each node has a probability of 1/iV of being infiltrated, the expected number of nodes whose activities become known to the authorities is
I=
^(2wi
(N-\
+ 3n2+
N-l
... +
\
= NnN-i), or I
N-l
^^
i?\
+ \)m (/
=
9. Throughout to its ithargument.
^??i?i
+ ??.j=i'/(l+2L/W)
the article, the notation x? denotes the partial derivative of the function x with respect of scale in creating links such thatC22 is sufficiently negative, some of the statics results are reversed.
11. The joint maximization of both TVand p depends critically on the cross-derivatives of TVand p in the production and cost functions. Since we want to focus on density and there is little empirical evi dence to justify any assumption concerning the sign of the cross-derivative, we abstract from the joint optimization. 12. The Memorial jsp?groupID=3499 Institute for the Prevention contains an extended discussion site http://www.tkb.org./Group. of Terrorism Web of the JeninMartyrs' Brigade.
References Arce, Daniel G., and Todd Sandler. 2005. Counterterrorism: A game-theoretic approach, Journal of Conflict Resolution 49 (2): 183-200. Zanini. 1999. Networks, netwar, and unformation-age Arquilla, John, David Ronfeldt, and Mich?le terrorism. In Countering theNew Terrorism, edited by Ian Lesser, Bruce Hoffman, John Arquilla, David Ronfeldt, and Mich?le Zanini, chap. 3. Santa Monica, CA: RAND. Brams, Steven, Hande Mutlu, and Shawn Ramirez. 2006. Influence in terrorist networks: From undir ected to directed graphs. Studies inConflict and Terrorism 29 (7), 679-94. Bueno de Mesquita, Ethan. 2005. The quality of terror.American Journal of Political Science 49 (3): 515-30. Carley, Kathleen, Matthew Dombroski, Max Tsvetovat, JeffreyReminga, and Natasha Kamneva. 2003. covert networks. In Proceedings of the 8th International Command and Con dynamic Destabilizing trol Research and Technology Symposium. Conference held at theNational Defense War College, papers/a2c2_carley_ Washington D.C. www.sandia.gov/ACG/focusareas/previousfocusareas/seldon/
-. -.
(accessed May 2, 2006). 2003_destabilizing.pdf and Todd Sandier. 1993. The effectiveness of anti-terrorism policies: A vector Enders, Walter, autoregression-intervention analysis. American Political Science Review 87 (4): 829-44. 2004. What do we know about the substitution effect in transnational terrorism? In Research on terrorism: Trends, achievements and failures, edited by Andrew Silke, 119-37. London: Frank Cass. 2006. The political economy of terrorism. Cambridge, UK: Cambridge University Press. Nonlinear Dynamics and Econometrics 1 (1): 1-11. Faria, Jo?o R. 2003. Terror cycles. Studies in
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions
57
Faria, Jo?o R., and Daniel G. Arce. 2005. Terror support and recruitment.Defence 16 (4): 263-73. Farley,
and Peace
Economics
Jonathan David. 2003. Breaking al Qaeda cells: A mathematical analysis of counterterrorism operations (A guide for risk assessment and decision making). Studies in Conflict and Terrorism 26:399-411. Gary. 2002. Sponsoring terrorism: Syria and Hamas. Middle East Intelligence Bulletin 4 (10).
Gambill,
(accessed May 2, 2006). http://www.meib.org/articles/0210{\_}si.htm Krebs, V. E. 2001. Mapping networks of terrorist cells. Connections 24 (3): 43-52. National Commission on Terrorist Attacks Upon theUnited States (2004). The 9/11 Commission New York: Norton. 1999. Terrorism Merrari, Ariel. Violence 11 (4): 52-65. Rodriguez, Jose. 2004. The March Sociology Analysis as a strategy of struggle: Past and future. Terrorism 11th terroristnetwork: In itsweakness University of Barcelona,
Report.
and Political of
of Organizations:
www.ub.es/epp/redes.htm
(accessed
May 2006). 1991. The application of network analysis to criminal intelligence: An assessment of Sparrow, Malcolm. the prospects. Social Networks 13:251-74 Wilkinson, Paul. 2005. Concerted effort needed to beat al-Qaeda. Transcript of radio interview available at www.abc.net.au/pm/content/2005/sl412793.htm (accessed April 2006).
This content downloaded by the authorized user from 192.168.82.206 on Tue, 4 Dec 2012 11:03:42 AM All use subject to JSTOR Terms and Conditions