Content uploaded by Erica Chenoweth
Author content
All content in this area was uploaded by Erica Chenoweth on Dec 09, 2019
Content may be subject to copyright.
This article was downloaded by:[Harvard University]
On: 20 June 2008
Access Details: [subscription number 769788083]
Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954
Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Defense & Security Analysis
Publication details, including instructions for authors and subscription information:
http://www.informaworld.com/smpp/title~content=t713412200
On Classifying Terrorism: A Potential Contribution of
Cluster Analysis for Academics and Policy-makers
Erica Chenoweth a; Elizabeth Lowham b
aBCSIA, Harvard University, Cambridge, MA, USA
bDepartment of Political Science, California Polytechnic State University, San Luis
Obispo, CA, USA
Online Publication Date: 01 December 2007
To cite this Article: Chenoweth, Erica and Lowham, Elizabeth (2007) 'On
Classifying Terrorism: A Potential Contribution of Cluster Analysis for Academics
and Policy-makers', Defense & Security Analysis, 23:4, 345 — 357
To link to this article: DOI: 10.1080/14751790701752402
URL: http://dx.doi.org/10.1080/14751790701752402
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction,
re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly
forbidden.
The publisher does not give any warranty express or implied or make any representation that the contents will be
complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be
independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings,
demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or
arising out of the use of this material.
Downloaded By: [Harvard University] At: 04:20 20 June 2008
INTRODUCTION
One need only consider for a moment the vast array of terrorist groups and their
ideologies to grasp the diversity of motivations, tactics, and damage wrought by terror-
ists around the world. In spite of these differences, scholars have begun to compare and
contrast terrorist groups in different polities, arguing that we can learn lessons from
such comparisons – especially in transferring counter-terrorism policies from one
context to the next. Recent studies on terrorism have provided new ways to compare
terrorist groups across cases. Analyses of terrorist groups and incidents have begun to
recognize the evolutionary nature of terrorism, as well as its ability to adapt to counter-
terrorism tactics. Moreover, scholars have noticed that terrorist tactics tend to be
“contagious” in that some terrorist groups either emulate other groups’ successes or
learn from their mistakes.2
Despite such observations, most comparisons of terrorist groups include either
small-ncase studies or large-nquantitative analysis. Small-nanalyses often provide
thick description while neglecting the “replication” or transferability of observations to
a more general explanation of a phenomenon. On the other hand, large-nstudies are
unable to capture contextual qualities, peculiarities among cases, and omitted
variables. As such, researchers often sacrifice causal inference when applying either
methodology.
While there are advantages and weaknesses to each approach, we suggest that large-
scale comparisons of terrorist incidents can benefit from the application of cluster
analysis, a context-sensitive statistical method. In the search for a more comprehensive
classification system, we suggest using cluster analysis to re-classify terrorist groups
Defense & Security Analysis Vol. 23, No. 4, pp. 345–357, December 2007
ISSN 1475-1798 print; 1475-1801 online/07/040345-13 © 2007 Taylor & Francis 345
DOI: 10.1080/14751790701752402
On Classifying Terrorism: A Potential
Contribution of Cluster Analysis for
Academics and Policy-makers
Erica Chenoweth1
BCSIA, Harvard University, 79 John F. Kennedy Street, Cambridge, MA 02138, USA
Elizabeth Lowham
California Polytechnic State University, Department of Political Science, One Grand Avenue,
San Luis Obispo, CA 93402, USA
23-4 master finished 20/11/07 9:03 AM Page 345
Downloaded By: [Harvard University] At: 04:20 20 June 2008
based upon both their motives and their means. Such a classification method takes into
account not only the terrorists’ motivations, but also the weapons, targets, location, and
level of destructiveness invoked by the terrorist attack. While the clusters provide a way
to analyze trends in terrorist attacks that transcend the motive (usually described as
religious, ethnic, or politically-based), they also provide interesting insight into similar-
ities between the terrorist attacks of different groups in different time periods, as well as
a way to view the dynamic nature of terrorism over the past few decades. Such method-
ologies provide more comprehensive ways of classifying terrorist incidents across cases
while maintaining sensitivity to the context of the cases.
To illustrate our argument, we utilize a biased dataset – the 2003 US State Depart-
ment’s Chronology of Significant Terrorist Incidents – in order to demonstrate how cluster
analysis can provide a broader basis for comparison. The State Department dataset is
not ideal from a purely scientific standpoint because of reporting bias (a focus on
incidents involving American citizens, etc). However, the clusters resulting from even a
biased dataset indicate that there is much to learn from clustering terrorist events based
upon means, motives, and methods. While current scholars usually classify groups
according to their religious sect or ethnic group, including the means of attacks shows
linkages across cultures and time periods and allows for a more comprehensive, cross-
case comparison.
The clusters resulting from our initial analysis demonstrate that terrorism is a con-
stantly-evolving and contagious phenomenon. While this is not a particularly original
observation, the clusters provide comprehensive mechanisms through which US
officials can identify potentially informative experiences from clusters of terrorist
activity abroad. From a methodological standpoint, the results reveal that cluster
analysis provides a technique for large-scale comparisons while still maintaining the
contextuality and comprehensiveness of individual incidents.
WHY “RE-CLASSIFY” TERRORISM?
The aim of this article is to explore alternative ways to conceive of terrorist typologies,
or the classification of terrorist groups, for analysis and policy response. As such, we
define terrorism in accordance with the US State Department’s definition, which
follows that of scholars such as Bruce Hoffman and others. This definition is “Premed-
itated, politically motivated violence perpetrated against noncombatant targets by sub
national groups or clandestine agents, usually intended to influence an audience” in
Title 22 of the United States Code, Section 2656f(d).3
While defining terrorism is a daunting task, so is classifying terrorist groups, as
evidenced by the hundreds of terrorist typologies that exist in the literature. Schmid
and Jongman identified 31 in 1988, before terrorism studies was a well-developed field
of inquiry.4Terrorist typologies, though numerous, are surprisingly limited in their
scope. Most classification systems focus on the origins or motivations of the group of
interest. Often this manifests in a description of the group as “nationalist”, “Marxist”,
or “Islamic fundamentalist”. For instance, the US State Department’s database of sig-
nificant terrorist incidents classifies terrorist groups with such descriptions as “Muslim
extremists” and “Palestinian suicide bombers” to the exclusion of other attack charac-
346 •ERICA CHENOWETH AND ELIZABETH LOWHAM
23-4 master finished 20/11/07 9:03 AM Page 346
Downloaded By: [Harvard University] At: 04:20 20 June 2008
ONCLASSIFYING TERRORISM •347
teristics such as type of target, number of casualties, or method of attack. The State
Department, therefore, usually classifies the attacks in terms of who is responsible,
rather than the outcome or method of the attack. Such practices are common in other
terrorist events databases, such as the “International Terrorism: Attributes of Terrorist
Events” (ITERATE) and “Terrorism in Western Europe Database” (TWEED)
databases.5
In academia, many scholars use encyclopedic entries or databases such as the State
Department’s list of Foreign Terrorist Groups to classify terrorist organizations.6
Although these databases include a multitude of information concerning the motives
and means of attacks, most scholars who derive terrorist typologies of terrorism ignore
the means and classify groups based upon motives alone. This type of classification
system makes sense in light of its simplicity, as well as the growing attention paid to
groups whose interests are incompatible with US national interests.
Despite these trends, Ariel Merari has devised a useful table distinguishing conven-
tional warfare, guerrilla warfare, and terrorism that accounts for more than just the
motive. His typology identifies unit size in battle, weapons used, tactics, targets,
intended impact, legality, etc. However, such a typology has not yet been quantified and
clustered for use in examining the similarities and differences between groups.7
The tendency to view terrorist groups solely on the basis of their ideologies has been
mitigated by recent literature that focuses more on the common tactics employed by
terrorist groups of all ideologies. For instance, both Pape and Bloom have investigated
suicide bombing as a distinct phenomenon that occurs among both religious and
secular groups in a number of ethnic contexts.8Such studies, however, are uncommon
and do not generate mechanisms for international and inter-temporal comparisons.
There are several reasons to reconsider the most common classifications of terrorist
groups. A more comprehensive system would contribute to the literature in many ways.
A rigorous construction of terrorist group clusters would provide a framework by which
they could be compared over time and space. Such a system would contribute to
terrorist events databases because it would actually combine factors of terrorist attacks
to produce “most similar” and “most different” classification systems based upon com-
prehensive criteria. For instance, classifying terrorists on both their motivations and
their methods could shed new light on the phenomenon, illuminating attack patterns
previously unnoticed. Moreover, such a method would allow large-ncomparisons of
groups previously examined only through small-ncase studies without losing sensitiv-
ity to context.
Second, a revised method such as the one we propose may show some interesting
relationships between seemingly unrelated groups. Mia Bloom’s study demonstrates
the counter-intuitive findings that both secular and religious terrorist groups engage in
suicide terrorism in order to outbid one another for public support.9Cluster analysis of
terrorist groups based upon comprehensive criteria reveal such dynamics, as well as
other potential commonalities among groups whose common behaviors may otherwise
go unnoticed. Clustering these groups may reveal some productive information con-
cerning interactions between them.
Furthermore, by viewing terrorist groups by their motives and attempting to an-
ticipate attack patterns, officials may inadvertently neglect important elements of
23-4 master finished 20/11/07 9:03 AM Page 347
Downloaded By: [Harvard University] At: 04:20 20 June 2008
counter-terrorism and response. For instance, while the “whodunit” version of
terrorist classification may be important in tracking down these individuals, a different
classification system may be more useful to local or state officials, who are most likely
to respond to the immediate effects of terrorism. The prescriptions associated with
counter-terrorism would benefit from insights derived from previous encounters with
similar “types” of terrorist attacks experienced in other states. By looking at clusters of
terrorist attacks, officials could identify the localities that have dealt with a similar
“type” of terrorism in the past, allowing them to share experiences with one another.
Terrorist groups should thus be classified not only on the basis of their motives, nation-
alities, and religions, but also on the basis of their tactics, destructiveness, and targets.
CLUSTER ANALYSIS FOR COMPREHENSIVE
CLASSIFICATION
The data set on terrorist incidents used in this study identifies “Significant Terrorist
Incidents” as categorized by the US State Department between 1961 and 2003 for a
total of 259 incidents. Before describing the method, we should discuss some
important limitations of the data. Most obviously, they are biased toward terrorism
against Americans. For instance, a kidnapping of an American citizen in Colombia for
ransom is considered a significant terrorist attack, although a kidnapping of a
Colombian citizen in Colombia would not be considered a significant terrorist attack
by the US. Second, the data are biased toward terrorism in recent years. If looking
solely at this data set, it would appear as though there has been a dramatic increase in
terrorism since 2000.
Due to these shortcomings, the State Department data represents a “hard case” for
the argument that cluster analysis can provide novel information. As biased as it is, if
the State Department data can reveal important patterns based upon attack character-
istic criteria, then this method should be applied to other, more objective and complete,
databases. Since we are simply proposing and demonstrating a new method of catego-
rizing terrorist events, rather than a strict set of terrorism categories, the limitations of
the State Department data are less damaging to our overall argument.
Using the “Significant Terrorist Incident” summaries, we coded each terrorist
incident using a series of dichotomous variables representing the type of group involved
in each attack, general characteristics of the attack, the types of weapons, the attack
types, the target types, the target populations, and the destructiveness of the attack in
terms of casualties and deaths. Each terrorist incident is therefore described on 45
different dichotomous variables, where a 1 indicates that the feature is present, and 0
indicates that the feature is absent (see Table 1 in the Appendix to this article for a list
of variables and their descriptions).
There are two critical choices in setting up a cluster analysis. The first is the choice
of the measure of similarity within the data; the second is the choice of the algorithm to
determine groupings. Using all 45 variables, we ran a cluster analysis on the 259
incidents using a Jaccard co-efficient as measure of similarity and an average between
groups linkage as the computational algorithm. The Jaccard co-efficient is simply the
sum of all the positive matches between two cases as a proportion of the total possible
348 •ERICA CHENOWETH AND ELIZABETH LOWHAM
23-4 master finished 20/11/07 9:03 AM Page 348
Downloaded By: [Harvard University] At: 04:20 20 June 2008
ONCLASSIFYING TERRORISM •349
positive matches (45). For example, if two incidents were both suicide attacks (meaning
that they both have a value of 1 in the suicide variable), SPSS counts this as a positive
match contributing to the numerator of the co-efficient. On the other hand, if no-one
claimed responsibility for the attacks in the same two incidents (meaning that they both
have a value of 0 in the responsibility variable), Statistics Package for Social Sciences
(SPSS) does not count this variable toward the measure of similarity. Thus, the Jaccard
co-efficient can be thought of as the proportion of characteristics that match across any
two cases.
SPSS begins by calculating a similarity matrix of all incidents based on the Jaccard
co-efficient, then grouping the two (or more) incidents that have the highest similarity
together. The program then averages the Jaccard values across the remaining
individuals. Essentially, the average between groups algorithm repeats these steps,
grouping the most similar cases and then averaging the similarities of the remaining
individuals. Finally, in the last step, the entire data set of 259 cases becomes one cluster.
Cluster analysis itself is simply a way of detecting natural groupings within a dataset
based on the information available. By using cluster analysis, relationships and
dynamics which may not be immediately obvious begin to emerge. Depending on the
purposes of the scholar or policy-maker, such classifications can provide useful infor-
mation on possible response or prevention measures, risk assessments, or simply
re-categorizations of attacks based on variables besides nationality or ethnicity for
analysis. Moreover, innovative, context-sensitive methods of inquiry such as cluster
analysis could inform mainstream terrorism studies in ways that could bridge the gaps
between policy-makers and scholars.
ANALYSIS
Using the cluster analysis described above, ten core clusters of terrorist incidents
emerge from the State Department data. These core groups represent 69 per cent of the
total number of cases in the data set. We consider the remaining 31 per cent of cases as
periphery groups, meaning that they are watered down versions of the core incident
types. It is important to understand that these core types do not represent any particu-
lar incident, but rather that they are the average score on each variable for all incidents
within a particular cluster.
To interpret the ten core ideal types, we considered all variables with a score of 0.6 or
higher as important characteristics of the group. That is, if more than 60 per cent of the
cases in any given cluster have a particular characteristic, we considered it important for
defining the cluster. Interestingly, the type of group committing the act is only crucial
in two of the ten clusters. Although we did not include year or country of attack as
variables in the cluster analysis, we did investigate whether or not there were patterns in
time or in location in the clusters that emerged.
We report best specimens in the descriptions below. The best specimens are found
by correlating each incident in a cluster with its ideal type (defined by the means of each
variable). The incident with the highest correlation with the core ideal type is consid-
ered the best specimen of the type. In some cases, more than one case had the same
Pearson’s R with the ideal type, in which case, only one of the incidents is described.
23-4 master finished 20/11/07 9:03 AM Page 349
Downloaded By: [Harvard University] At: 04:20 20 June 2008
The bombing clusters
Of the ten core clusters, five use bombs at least 79 per cent of the time (see Table 2 in
the Appendix of this article for a list of clusters). These five clusters cannot, therefore,
be distinguished according to the type of weapon used. Differences in target types,
target populations, whether or not a group takes responsibility and the results of the
attack largely differentiate these clusters. In only one of the bombing clusters is the type
of group committing the attack an important characteristic of the core type.
Bombings of a public population where a liberation group takes responsibility (A)
In this cluster, all attacks are bombings of a public population by liberation groups who
take responsibility for the attacks. In other words, all 22 cases have these four charac-
teristics. In 95 per cent of the cases, a group also claims responsibility for the attack.
Finally, in 81 per cent of the attacks, casualties range from 36 to 97 individuals. All but
two of these incidents occurred in Israel during or after 2001. One of the best
specimens of this ideal type is a suicide bombing on a bus in Haifa, Israel in December
2001. Hamas claimed responsibility for the attack, which killed 15 people and
wounded 40.
Bombings of a public population at a commercial target where groups take responsibility (B)
Nearly all the cases in this cluster are bombings (89%) of a public population (89%) at
a commercial target (84%). In 70 per cent of the cases, a group takes responsibility for
the attack, though there is no clear group domination of this type of incident. Finally,
in almost 60 per cent of the cases, the casualties are greater than 100 individuals. Unlike
the previous cluster, there is no dominant country or time pattern. The best specimen
for this cluster is the bombing of the Central Bank in Colombo, Sri Lanka by the Tamil
Tigers in January 1996. The attack killed 90 individuals and wounded over 1,400.
Bombings of a public population at a commercial target by unknown groups (C)
This cluster is composed entirely of bombings of public populations at commercial
targets. The groups committing the attacks in this cluster are largely unknown (85%).
The casualties from these attacks typically range from 36 to 97 individuals (69%) and
there are generally between five and 15 deaths from such attacks (62%). There is no
clear domination of this cluster by country or year. The best specimen is the car
bombing of a shopping center near the American Embassy in Peru in 2002. The
attack, which no group claimed responsibility for, killed nine individuals and
wounded 32.
Bombings of official populations at official targets by unknown groups (D)
Most of the cases in this cluster (79%) are bombings of official populations (93%) at
official targets (83%) by unknown groups (62%). There is no clear pattern of casualties
or deaths within this cluster, perhaps because of the nature of the target. There is also
no clear pattern of country or year in this cluster, although a fair number (11) of the
incidents occurred in Iraq in 2003. The best specimen in this cluster is the suicide car
bombing in Kirkuk in late November 2003. The suspected target was the headquarters
350 •ERICA CHENOWETH AND ELIZABETH LOWHAM
23-4 master finished 20/11/07 9:03 AM Page 350
Downloaded By: [Harvard University] At: 04:20 20 June 2008
ONCLASSIFYING TERRORISM •351
of the Patriotic Union of Kurdistan. The bombing, for which no group claimed respon-
sibility, killed five individuals.
Bombings of foreign populations at military targets where a group takes responsibility (E)
Nearly all the incidents in this cluster are bombings (85%) of foreign populations
(85%) at a military target (92%). In roughly 70 per cent of the incidents a group takes
responsibility for the attack. There are generally more than 15 deaths resulting from
these incidents (62%). There is no clear country or time pattern in this cluster. The
best specimen for this cluster is the bombing of a military compound near Riyadh,
Saudi Arabia, in November 2003. The Islamic Movement of Change claimed respon-
sibility for the attack which killed over 40 individuals, including several foreign
nationals.
The non-bombing clusters
The remaining five clusters include an assortment of different attack types differenti-
ated by weapons used, the type of attack, and the target attack. As opposed to the
bombings, the results of these incident clusters tend to be on the small side. In only one
of these clusters does group type play a distinguishing role among the core types.
Gun attacks where a righteous vengeance group takes responsibility (F)
In nearly all the incidents in this cluster, a righteous vengeance group (90%) takes
responsibility for the attack (90%). The attack tends to be carried out with guns (70%)
and tends to be focused on a religious population (60%) at a religious target (70%).
There is no clear death or casualty result from these attacks. Of the ten incidents in this
core cluster, five occurred in Pakistan after 2001. However, the best specimen for this
cluster occurred when Sikh terrorists took control of the Golden Temple in Amritsar,
India in 1984. 100 people were killed as Indian security forces regained control of the
temple.
Assassination of foreign population with guns by unknown groups (G)
This cluster is composed entirely of incidents where the casualties and deaths are small.
Two-thirds of the incidents are assassinations by gun. The attack is largely carried out
on a foreign population (73%) in an open-air target (60%) by unknown groups (87%).
Six of the 15 attacks in this cluster occurred in Iraq during 2003. The best specimen for
this cluster occurred in November 2003 when two gunmen assassinated the deputy
mayor in Baghdad, Iraq.
Attacks on foreign, official populations in open air targets where groups take responsibility (H)
All of the incidents in this cluster are attacks in the open where the casualties and deaths
are both small and groups take responsibility for the attack. The attacks tend to be
foreign in nature (83%) and tend to be focused on official populations (83%). There is
no clear weapon pattern, indeed, none of the attacks in this cluster indicate weapon use
and the cluster is split roughly 60 per cent kidnappings where the victim is killed and 40
per cent assassinations. There is no clear time or country pattern in this cluster. One of
23-4 master finished 20/11/07 9:03 AM Page 351
Downloaded By: [Harvard University] At: 04:20 20 June 2008
the best specimens for this cluster occurred when the Islamic Jihad kidnapped and later
killed a US Embassy officer in Beirut, Lebanon in 1984.
Attacks on official populations at official targets with no deaths where a group takes responsi-
bility (I)
All of the incidents in this cluster are attacks on official populations at official targets
where there are no deaths resulting from the attack. In three-quarters of the incidents,
a group takes responsibility for the attack and there are generally less than six casualties.
There is no clear country or time domination in this cluster. The best specimen for this
cluster is the abduction of four UN military observers in Georgia in February 1998.
Kidnappings at open-air targets with small casualties, no deaths (J)
All of the incidents in this cluster are kidnappings at open-air targets which, while
resulting in small casualties, do not result in any deaths. In nearly all of the cases, a
group takes responsibility for the attack (94%). The kidnapping tends to be foreign in
nature (81%). Ten of the 16 incidents in this core cluster occurred in Colombia. One of
the best specimens for this cluster is the Colombian People’s Liberation Army kidnap-
ping of a US citizen in 1999 in an unsuccessful attempt to extract a ransom.
DISCUSSION
Re-classifying terrorist incidents based upon our analysis does not provide any answers,
but rather raises a number of interesting questions for future inquiry. Using this
method, for instance, we can derive clusters that yield significant research puzzles from
the clusters that emerged from this analysis. For instance, why would groups take
responsibility for attacks against military populations (Cluster E) and not for bomb
attacks against public officials (Cluster D)? Why would groups take responsibility for
gun attacks and assassinations against official populations (Clusters H and I) and not
for bombings against official populations? Moreover, what explains regional trends in
methods, such as the righteous vengeance gun attacks located in South Asia in Cluster
F, or the Colombian kidnappings in Cluster J? And why do some clusters concentrate
around one time period (Clusters A and F), whereas others sustain over long periods of
time (Clusters H and I)? These are just a few of the possible research questions that
emerge from this particular analysis.
Given the limitations of the State Department data set, we are not proposing that our
classification scheme for terrorist incidents is ideal or lasting. The aim here is simply to
illustrate that methods exist through which terrorism scholars can identify patterns in
terrorist group behavior that include a large-n while maintaining context and cohesion
of cases. Including various measures of means, motives and opportunities leads to a
different scheme of classification than those traditionally used to discuss terrorist
incidents, and cluster analysis is a useful tool for understanding these categories.
Using cluster analysis to re-classify terrorist groups based upon motives, means, and
effect of attack provides a more systematic and comprehensive way to analyze terrorism
without biasing the implications of the trends to favor motive-based analyses.
Moreover, the clusters identify emergent patterns between and among the terrorist
352 •ERICA CHENOWETH AND ELIZABETH LOWHAM
23-4 master finished 20/11/07 9:03 AM Page 352
Downloaded By: [Harvard University] At: 04:20 20 June 2008
ONCLASSIFYING TERRORISM •353
attacks by different groups in different time periods and demonstrate the evolutionary
nature of terrorism over the past half century. The clusters show that terrorism is not an
isolated event emanating from a single religion or ideology, but is rather a tactic of
unconventional warfare used almost universally among different people and in different
times to achieve political ends. We propose that scholars and policy-makers should
attempt to form cluster-based analyses of terrorist incidents using other, more compre-
hensive, datasets in order to identify important trends, patterns, and puzzles for further
inquiry.
APPENDIX
Table 1: Variable List
For all variables, 1 = Yes; 0 = No
Case Case id
Ccode Country code (cowcode)
Year Year of attack
Group Group name
Foreign Attack on foreign citizens
Suicide Suicide attack
Response Has a group claimed responsibility?
Assass Whether or not assassination
Biologic Whether or not biological weapons used
Bomb Whether or not bombs used
Bombsieg Whether or not bombs used in a siege
Gas Whether or not gas used
Guns Whether or not guns used
Carjack Car jacking or not
Kidnap Kidnapping or not
Kidnapki Kidnap and kill or not
Planecra Plane crash or not
Planehij Plane hijack or not
Siege Siege or not
Airport Airport target
Commerci Commercial target
Education Educational target
Military Military target
Official Official target (government, police, fire station, etc.)
Openair In open-air public or not (i.e. an open-air market or on a street)
Plane On a plane
Resident At a residence
Public At a public area (in a public building)
Religious At a religious site
Commpop Target commercial individuals
Milpop Target military individuals
Offpop Target official individuals
Pubpop Target public populations
Relpop Target religious populations
23-4 master finished 20/11/07 9:03 AM Page 353
Downloaded By: [Harvard University] At: 04:20 20 June 2008
Catsmall Casualties < = 6
Catmed Casualties between 6 and 35 (casualties = target + terrorists injured or
dead)
Catlarge Casualties between 36 and 97
Catsuper Casualties > = 100
Nodeath No deaths from attack (only target dead included)
Smdeath Fewer than 4 deaths
Mddeath Between 5 and 15 deaths
Lgdeath Greater than 15 deaths
Other Group type does not apply to following categories
Unknown Unknown group type
Marxist Marxist group
Liberation Liberation group
Army Military group
State State terror
Self-defense Group claiming defense (i.e. defense league)
Righteous vengeance Religious/righteous group
Source: US State Department. 2004. “Significant Terrorist Incidents: A Brief Chronology.”
http://www.state.gov/r/pa/ho/pubs/fs/5902.htm. Last accessed: 5/17/04. Also, MIPT-RAND list of
terrorist groups (for help identifying group type only). Available online at
http://www.tkb.org/Home.jsp. Last accessed 9/1/05.
354 •ERICA CHENOWETH AND ELIZABETH LOWHAM
23-4 master finished 20/11/07 9:03 AM Page 354
Downloaded By: [Harvard University] At: 04:20 20 June 2008
ONCLASSIFYING TERRORISM •355
Table 2: Core Cluster Results with Highlighted Variables Correlated > .6
A CORE B CORE C CORE D CORE E CORE F CORE G CORE H CORE I CORE J CORE
Group
Liberation 1 0.351351 0 0.103448 0.153846 0.1 0 0.25 0.333333 0.3125
Marxist 0 0.027027 0 0 0.307692 0 0 0.25 0.25 0.5
Unknown 0 0.162162 0.846154 0.62069 0.153846 0 0.866667 0.166667 0.083333 0
Other 0 0 0 0.034483 0 0 0 0.166667 0.333333 0.0625
Army 0 0.108108 0 0.034483 0.076923 0 0.066667 0 0 0.0625
Righteous
vengeance 0 0.27027 0.076923 0.172414 0.230769 0.9 0.066667 0.166667 0 0.0625
State 0 0.081081 0.076923 0.034483 0.076923 0 0 0 0 0
General
Foreign 0 0.297297 0.538462 0.37931 0.846154 0.1 0.733333 0.833333 0.583333 0.8125
Suicide attack 0.954546 0.243243 0.230769 0.37931 0.153846 0 0 0 0.083333 0
Responsibility
claimed 1 0.702703 0.307692 0.310345 0.692308 0.9 0 1 0.75 0.9375
Weapon
Biologic 0 0 0 0 0 0 0 0 0 0
Bombing 1 0.891892 1 0.793103 0.846154 0.2 0.2 0 0.5 0.0625
Bomb-siege 0 0 0 0.034483 0 0 0 0 0 0
Gas 0 0.027027 0 0 0 0 0 0 0 0
Guns 0 0 0 0.206897 0.076923 0.7 0.666667 0 0.083333 0
Attack Type
Assassination 0 0 0 0.103448 0.153846 0 0.666667 0.416667 0 0
Carjacking 0 0 0 0 0 0 0 0 0 0
Kidnapping 0 0 0 0 0 0 0 0 0.333333 1
Kidnap-killing 0 0 0 0.034483 0 0 0 0.583333 0 0
Plane crash 0 0.027027 0 0 0 0 0 0 0 0
Plane hijack 0 0.054054 0 0 0 0 0 0 0 0
Siege 0 0.027027 0 0 0 0.2 0 0 0.083333 0
23-4 master finished 20/11/07 9:03 AM Page 355
Downloaded By: [Harvard University] At: 04:20 20 June 2008
Target Type
Airport 0 0 0 0 0 0 0 0 0 0
Commercial 0.272727 0.837838 1 0.034483 0.076923 0 0 0 0 0
Education 0 0 0 0 0 0.1 0 0 0 0
Military 0 0 0 0 0.923077 0.1 0 0 0 0
Official 0 0.081081 0.153846 0.827586 0 0 0.133333 0 1 0
Open air 0.136364 0.027027 0 0.068966 0.076923 0.1 0.6 1 0 1
Plane 0 0.108108 0 0 0 0 0 0 0 0
Residential 0 0 0 0.068966 0 0 0.333333 0 0 0
Public 0.545455 0.027027 0 0.034483 0 0.2 0 0 0 0
Religious 0.045455 0.054054 0 0 0 0.7 0 0 0 0
Population
Type
Commercial 0 0.081081 0 0 0 0 0.2 0 0 0.125
Military 0 0 0 0.034483 1 0.1 0.066667 0.083333 0 0
Official 0 0.054054 0.153846 0.931035 0 0 0.533333 0.833333 1 0.1875
Public 1 0.891892 1 0.068966 0 0.5 0.066667 0.083333 0 0.5625
Religious 0 0.027027 0 0 0 0.6 0.133333 0 0 0.0625
Results
Small
casualties 0.090909 0.243243 0 0.068966 0.307692 0 1 1 0.75 1
Medium
casualties 0.045455 0.081081 0.153846 0.448276 0.076923 0.4 0 0 0.25 0
Large casualties 0.818182 0.081081 0.692308 0.206897 0.384615 0.3 0 0 0 0
Catastrophic
casualties 0.045455 0.594595 0.153846 0.275862 0.230769 0.2 0 0 0 0
No death 0.045455 0.324324 0 0.034483 0.230769 0 0 0 1 1
Small deaths 0.272727 0.054054 0.307692 0.172414 0.153846 0 1 1 0 0
Medium deaths 0.454546 0.189189 0.615385 0.448276 0 0.5 0 0 0 0
Large deaths 0.227273 0.405405 0.076923 0.344828 0.615385 0.5 0 0 0 0
356 •ERICA CHENOWETH AND ELIZABETH LOWHAM
23-4 master finished 20/11/07 9:03 AM Page 356
Downloaded By: [Harvard University] At: 04:20 20 June 2008
ONCLASSIFYING TERRORISM •357
NOTES
1. The authors would like to thank Ron Brunner for his useful comments.
2. Manus Midlarsky, Martha Crenshaw and Fumihiko Yoshida, “Why Violence Spreads: The
Contagion of International Terrorism”, International Studies Quarterly, Vol. 24 No. 2, 1980,
pp. 262–298; Walter Enders and Todd Sandler, The Political Economy of Terrorism, New York:
Cambridge University Press, 2006; Robert Pape, Dying to Win, New York: Random House,
2005. Also Martha Crenshaw, Ter rorism in Context, University Park: Pennsylvania State Uni-
versity Press, 1995.
3. Quoted in Bruce Hoffman, Inside Terrorism (revised edn), New York: Columbia University,
2006, p. 31.
4. Alex Schmid and Albert Jongman, Political Terrorism, New Brunswick, NJ: Transaction
Press, 1988.
5. TWEED databases: see Edward F. Mickolus et al., “ITERATE: International Terrorism:
Attributes of Terrorist Events”, Vinyard Software, 2006; also, Jan Oskar Engene, Terrorism
in Western Europe, Cheltenham: Edward Elgar, 2004.
6. See, for example, Schmid and Jongman, op. cit., 1988; also, US State Department, Signifi-
cant Terrorist Incidents: A Brief Chronology, Washington DC, 2004, http://www.state.gov/r/pa
/ho/pubs/fs/5902.htm, last accessed 17 May 2004.
7. Ariel Merari, “Characteristics of Terrorism, Guerilla, and Conventional War as Modes of
Violent Struggle”, in V. S. Ramachandran (ed.), Encyclopedia of Human Behaviour, Vol. 4,
San Diego: Academic Press, 1994, p. 401.
8. Robert Pape, Dying to Win, New York: Random House, 2005; also, Mia Bloom, Dying to
Kill: The Allure of Suicide Terror, New York: Columbia University Press, 2005.
9. Mia Bloom, op. cit.
23-4 master finished 20/11/07 9:03 AM Page 357
