Information about http://www.people.fas.harvard.edu/~esimpson/papers/diffusion.pdf

The Diffusion of Insurgency and Suicide …

Tags: american political science, american political science association, anecdotal evidence, explicit permission, fighting in iraq, flexible tools, harvard university, insurgency, kidnappings, late august, maoist, michael horowitz, phd candidates, political science association, roadside bombings, state actors, suicide bombings, suicide terrorism, terrorist groups, theorists,
Pages: 28
Language: english
Created: Thu Sep 15 14:32:37 2005

Display cached document

Page 1

Page 2

Page 3

Page 4

Page 5

Page 6

Page 7

Page 8

Page 9

Page 10

Page 11

Page 12

Page 13

Page 14

Page 15

Page 16

Page 17

Page 18

Page 19

Page 20

Page 21

Page 22

Page 23

Page 24

Page 25

Page 26

Page 27

Page 28

The Diffusion of Insurgency and Suicide

Terrorism Strategies

Michael Horowitz and Erin Simpson1

Prepared for the 2005 American Political Science

Association Annual Meeting: Please do not cite without

the explicit permission of the authors

1
The authors are both PhD Candidates in the Department of Government at Harvard University.
I. Introduction

In late August 2005, the BBC reported that jihadists were returning from Iraq to

Afghanistan in an effort to train locals in techniques they had learned while fighting in

Iraq--in particular, suicide bombings and kidnappings. Similarly, the New York Times

attributed a spike in roadside bombings in Afghanistan to Al Qaeda operatives with

experience in Iraq. These reports are not isolated events; there is no shortage of

anecdotal evidence reported in the popular media related to cooperation between rebel

and terrorist groups. But there is sadly little effort to understand this phenomenon in a

comparative or more scholarly context. What are the means by which non-state actors--

rebel and terrorist groups--adopt new tactics and strategies? Do these actors converge

on one superior strategy? Are their choices mediated by norms or culture, or are they

driven more by the dictates of rationalist efficiency?

We answer these questions in the context of a theory of diffusion. As described

below, the wide-ranging literature on diffusion provides both flexible tools and a well-

grounded framework for understanding the means by which ideas or technologies spread

throughout a social system. Borrowing from existing data, and creating our own new

measure of "insurgency" over time, we test the ideas of diffusion with regard to suicide

bombing and Maoist insurgency strategies. We find, broadly speaking, that the diffusion

of these tactics and strategies conforms to the expectations of diffusion theorists. While

preliminary, these findings give us reason to expect that there are elements of norms or

culture that affect the choices these groups make--they are not purely functionalist.

As the US continues to confront terrorist and other "irregular" threats, we believe

that a greater understanding is needed with regard to the evolution and development of

2
rebel or terrorist tactics and strategy. Obviously, this is only one part of what is an

exceptionally complicated equation. There are important economic and social variables

that also require more and better attention when trying to understand the micro-

foundations of non-state violence. That said, an understanding of these groups that

ignores how they actually fight the conflicts in which they are engaged is both incomplete

and misleading. This paper is a first step in understanding these dynamics.

II. Theory of Diffusion

The study of international processes through the lens of diffusion has received

intermittent attention over the last few decades. This section will describe some of the

underlying concepts behind diffusion research, explain some applications in the

international politics arena, and describe why it may be a useful avenue of approach to

tackle issues of insurgency and terrorism.

A. Theoretical Concept of Diffusion

The canonical text on diffusion is the Diffusion of Innovations, by Everett Rogers.

For over thirty years, Rogers has tracked research on diffusion as it rose and declined in a

variety of fields, for the most part ignored by political science researchers. An off-the-

cuff definition of diffusion is simply the way that a given innovation spreads or does not

spread throughout the population of interest. The more technical definition is that

"Diffusion is the process in which an innovation is communicated through certain

channels over time among the members of a social system" (Rogers, 2003: 5). Key to the

3
idea of diffusion is the notion of some form of communication, even if implicit, between

members of a social system. . That is to say, a diffusion approach begins with the

assumption that actors do not make 100% independent decisions without receiving any

information. For example, if two actors in different places simultaneously created hybrid

corn without the benefit of any cross-pollination of information, it would not represent a

diffusion process. However, if the two actors both adopt hybrid corn either through a

mutually constitutive process or through learning by one of the other, a diffusion

approach is applicable.

The simplest diffusion models are represented by S curves that show the slow

beginning, rapid dissemination, and leveling off of an innovation as it spreads throughout

the population. The X-axis generally represents time since the introduction of the

innovation and the Y-axis represents cumulative adoptions throughout the given social

system. Cumulative adoption can be represented in several ways, referring to the total

number of people that have adopted an innovation, the total number of new adopters in a

given period of time, or the total percentage of adopters out of the total possible set of

adopters. A model S curve is shown below in Figure 1.

4
Figure 1: Model S Curve of Cumulative Adoptions (Source: Horowitz 2005)

Total
Adopters Takeoff Point
Slowdown Point

Time since First Mover Demonstration

The stages of the innovation diffusion process generally correspond to four or five

different sets of adopters: First Mover(s), Early Adopters, Early Majority Adopters, Late

Majority Adopters, and Laggards. Some typologies of diffusion merge the Early

Majority and Late Majority category into a simple "Majority Adopters" category.

The idea is that a function exists which relatively accurately describes the spread

of a given innovation throughout the social system under analysis. That function is used

to create what is called a "diffusion term," which in combination with the independent

variables of interest should accurately predict the actual rate and extent of diffusion of the

innovation of interest through the social system.

There have traditionally been two dominant types of diffusion models, an internal

influence model and an external influence model. The internal influence model assumes

a given level of interaction within the target population and further presumes that

interaction is a means through which knowledge of the innovation is transmitted. Its

cumulative adoption function (CAF) looks like Figure 1 above. The alternative is an

5
external influence model. It is still a diffusion model because it presumes that

communication about an innovation occurs and has an important impact on the spread of

the innovation, but it posits that there is little or no direct communication between

members of the social system. The transmission mechanism for diffusion in an external

influence model is usually mass media communications. A typical external influence

CAF is presented below as Figure 2.

Figure 2: External Influence Diffusion Model (Source: Mahajan and Peterson 1985)

Total
Adopters Slowdown Point

Time since First Mover Demonstration

Some things about the international relations context make the diffusion approach

complicated. First, the international environment does not naturally conform to either the

internal or external diffusion models. There is communication through both things like

mass media and through direct contacts. This means a "mixed" diffusion model is more

appropriate, which increases the degree of difficulty involved in fitting the diffusion

curve because there are more parameters to accurately predict (Mahajan and Peterson

1985).

6
Second, the idea of "potential adopters" is potentially tricky in the international

relations context if the object of analysis is the adoption of an innovation by a nation-

state. The set of potential adopters in a village is the people that live in the village. The

set of potential adopters for a consumer durable in the United States are those that can

afford the product. This means that in studies of nation-states, the set of potential

adopters writ large is the number of nation-states in the international system, or the

number of states that can afford to purchase a particular military technology or the

number of states in a given region of interest. It varies depending on the nature of the

research and topic of interest. One recent application of a diffusion approach in the

military realm is by Zarzecki (2002), who analyzed the spread of generations of advanced

weaponry throughout the international system using a diffusion model.

In addition, in many of the canonical diffusion examples--water boiling in Peru

or hybrid seed adoption amongst farmers--researchers assume a fixed population. No

one works under the assumptions that the wonders of boiled water will cause new people

to move to the village in question, or that more people will become farmers as a result of

new seed technologies. In our context, however, this may be exactly the case. Open

rebellion may only be seen as possible in one country once insurgent success has been

demonstrated elsewhere. Thus the diffusion of the technology may increase the number

of actual and potential adopters. The choices, then, in the international context are not

just adopt/not adopt. Actors have a prior decision regarding participation that may be

simultaneously affected by the diffusion of strategies or tactics.

A diffusion model is not appropriate in all cases: if total independence of

observations exists, a diffusion perspective does not apply. For better or worse, the

majority of international relations research on violence has assumed independence of

7
observations, approaching each war in the international system, for statistical purposes, as

if it is an independent observation totally unrelated to previous (or neighboring) events in

the international system. In recent years, some political science researchers have

attempted to develop models that deal with both spatial and temporal dependence in large

datasets. Much of this draws on the large time-series literature. More recently, scholars

have addressed problems of non-independence across units (countries or states) as well

(Pluemper and Troeger 2004; Beck and Katz 2001, 1995; Zorn 2001; Green, Kim and

Yoon 2001). While these papers tend to focus on technical issues surrounding bias and

efficiency in parameter estimation, they do begin to develop the tools one would need to

fully work within a diffusion context. This project stands in agreement with these

scholars and attempts to outline a diffusion-based approach to study insurgency and

terrorism in part as an attempt to move beyond the flawed independence assumption.2

B. Diffusion Research in Political Science3

Most of the most prominent diffusion-related research in political science has

involved the study of policy innovations by American states and the way that states learn

from each other (Walker, 1969). Berry and Berry's 1990 article studied the probability of

state lottery adoption, finding that a combination of internal and regional diffusion

models best predicted whether or not a given state would adopt a lottery (Berry and

Berry, 1990). Other American politics research on diffusion includes studies of tax

2
This creates a unique academic space for diffusion research on conflict. Even if such a research program
discovered very little in the way of positive results, it would still provide useful information for
international relations research on warfare in general by demonstrating the potential continued utility of the
independence of observations approach.
3
This section draws heavily on a longer summary of existing research on diffusion is available in Horowitz
(Horowitz 2005).

8
policy, tort reform, education reform, workplace regulations, and child case (Mintrom

and Vergari, 1998; Zylan and Soule, 2000; Canon and Baum, 1981; Berry and Berry,

1992; Moore 2005).

Three sets of research within international relations have most prominently dealt

with diffusion issues. First, scholars writing in the war contagion literature, which treats

war as an epidemiological condition and attempts to explain it as such, have sometimes

adopted a diffusion approach, though the models utilized have generally not drawn very

heavily on the more econometric diffusion tradition in other fields. Siverson and Starr

find that as the number of border states involved in a war increases and alliance

relationships with those states increase, states are more likely to be drawn into wars

(Siverson and Starr, 1991). To some extent this finding is intuitive, and Simowitz has

argued that the research program is a failure in part because it has not justified the value

of the contagion approach as opposed to other political science approaches (Simowitz

1998a, 1998b; also see response by Starr and Siverson, 1998). Other studies of war and

diffusion have included region-specific studies of Africa and Latin America (Starr and

Most, 1983; Hammarstrom, 1994) as well as more general studies of the spread of

coercive practices in international disputes (Bremer, 1982).

A second area of diffusion research involves the spread of different types of

political regimes, mostly democracies. Gleditsch and Ward, most prominently, have

studied the spread of democracy and utilized diffusion models as a tool to determine the

probability that a given state becomes democratic (Gleditsch and Ward, 1997, 1999a,

1999b, 2000; Ward and Gleditsch, 1998; Cederman and Gleditsch, 2004). Their research

produced an S-shaped curve showing the relationship between the probability of a

democratic or autocratic transition and the regime type of neighboring states. The results

9
show a definite tipping point may exist on the regional level where, once the number of

democracies in a given region approaches N, the probability of a state in that region

becoming a democracy dramatically increases (Cederman and Gleditsch, 2004).

Third, there has been growing interest in the last few years at viewing financial

and trade policy instrument adoption as a diffusion process (Simmons and Elkins, 2004).

The degree of information sharing between countries or even the spread of information in

general is very highly correlated with the adoption of similar financial policy instruments.

A somewhat surprising finding in their research is the importance of cultural variables

like religious similarity in predicting policy adoption. States are much more likely to

adopt policies from countries they view as "like" them than countries that are viewed as

different. Though not surprising to any student of psychology, the empirical finding of

this relationship on the international level is striking. However, regardless of the specific

results, their research and associated papers show the utility of diffusion theorizing in

international relations research.

C. Existing Research on Insurgencies and Terrorism

The main debate that occurs throughout discussions of suicide terrorism and

insurgencies is that of motivation. It is possible to contrast functionalist, norms, and

cultural or religious approaches to thinking about suicide bombing and insurgencies. The

functionalist perspective is represented by Arreguin-Toft (2000). Toft argues that, over

time, insurgent groups increasingly fight with "indirect" insurgent strategies, as opposed

to trying to raise an army and fight "directly," because they decide that it is the best

10
strategy for victory. His argument that groups see the success of others and mimic argues

for a limited, or external influence, diffusion curve.

A second perspective on strategic choice is that these groups chose to behave the

way they do due to norms of appropriateness. This argument is best made by Farrell

(1997, 1999) who argues that military organizations choose strategies on the basis of

what other military organizations do, but as a form of status-seeking or legitimacy-

creating, not as a heuristic to adopt best practices. For example, Farrell writes that when

they achieved independence, the Irish resistance chose to establish a "regular"

conventional army modeled after the British, despite the fact that in the case of a British

attack, the independence of Ireland was likely to be best preserved through the guerilla

strategy utilized to achieve their independence in the first place.

A third perspective views the diffusion process through the lens of cultural

learning. Groups learn from each other over time and often due to perceptions of

"sameness". That is, as Simmons and Elkins find in their study of financial policy

adoption, groups are more likely to mimic other groups that they perceive to be in a

similar situation. While this may occur due to material or geographic similarities,

frequently sameness is a metric decided on the basis of ideology or religion.

As we will describe below, the results show that a combination of these

perspectives best explains insurgencies and suicide terrorism, though the cultural and

normative findings are somewhat more surprising. For the most part, academic research

in political science has tended to view insurgencies and terrorist groups as relatively

independent actors. Policy analysts have scrambled to try and understand the "networks"

that may exist between terrorist groups. Work focusing on how terror groups adapt

(Jackson, et al. 2005; Cragin et al., forthcoming), for example, has highlighted the several

11
of the ways in which terror or insurgent groups learn directly or indirectly from each

other. Some partnerships are not surprising: Hamas has received extensive training from

Hezbullah, for example. Other linkages are less direct: suicide bombing innovations

developed by the Tamil Tigers in Sri Lanka were mimicked in the Middle East without

any formal contact (Chalk nd)4.

However, academic research generally assumes the independence of terrorist

campaigns and groups. To the extent the issue is addressed at all it is usually in case-

specific research, not attempts to systematically look at the incidence of terrorism

throughout the international system. Those large-scale attempts to look at terrorism in

the international system, like the ITERATE and MIPT data gathering projects, are

generally interested in variables of clear interest to the general population like casualties,

meaning these efforts do not focus on potential links, either implicit or explicit, between

groups. This project, by focusing on potential inter-group diffusion of innovative

practices, specifically suicide bombing, attempts to move beyond this existing research

and evaluate terrorism through a different lens.

In terms of our work on insurgency, there is a rich, though incomplete, literature.

While there are many good volumes on particular cases (Malaya, Algeria, and Vietnam

being favorites), there is relatively little comparative work. Most comparisons focus on

either the common elements of insurgent success (see Ellis 1999; Lacquer 1998) or fine-

tuning of counter-insurgency practices (as the masters theses written every year at

Carlisle Barracks and Ft. Leavenworth demonstrate). What one almost certainly does not

find is research that problematizes the question of rebel strategy; this is nearly always

4
As noted below, there were contacts between the Tamils and Hezbullah prior to the Israeli invasion of
Lebanon in 1982. There is no evidence of direct communications after this date, however.

12
deus ex machina. Jeremy Weinstein's (2003) work on the various incentives and

constraints that affect rebel behavior vis-�-vis their supporting civilian population is a

very notable exception, as is the work by Arreguin-Toft (2000) mentioned above (see

also Kalyvas forthcoming). In analyzing military strategies in asymmetric conflicts

(which include several of the civil and colonial wars under review here), Toft arrives at a

functionalist interpretation of rebel strategies: weak actors adopt insurgent strategies

because they work. He posits that other actors see this success and mimic it over time.

He does not, however, test this claim. This study represents a plausibility probe

representing a possible test of that claim. Our results follow a brief discussion of our data

sources.

III. Data

We examine two different sources of data to probe for the utility of using a

diffusion perspective: civil war strategies and suicide bombing campaigns. As such we

use the population of cases defined by Fearon and Laitin (2003) for civil wars, and

Robert Pape's newly published data on suicide bombings. While these data have some

limitations and/or controversial aspects, they provide a useful starting point for our first

cut at this question. We have no reason to believe that the coding rules developed by

Pape and Fearon and Laitin would skew the results described below in any particular

direction. If anything, Fearon and Laitin's more permissive rules (vis a vis COW) are

more likely to allow lower-level, more sporadic forms of violence to be counted as civil

wars--biasing the data against the set of cases we are most interested in. On the other

hand, Fearon and Laitin correct a fundamental problem with the COW data by combining

13
the colonial wars (Algeria, Malaya, Indonesia, etc.) that meet all other components of

their coding framework. These wars are perhaps disproportionately insurgent in nature.

Pape's dataset is billed as the most complete dataset on suicide bombing ever

assembled. It covers the 1980-2003 period and includes both infamous cases of suicide

bombing by Middle Eastern terrorist groups and suicide campaigns by lesser-known

groups (in America) like the Tamil Tigers. Pape gathered his data by scouring publicly

available news articles through research services like Lexis-Nexis and interviewing

knowledgeable people in an attempt to uncover every instance of successful suicide

bombing during the study period.5 If there is any likely "bias" in Pape's dataset, it is his

coding of some campaigns that arguably have involved multiple groups acting somewhat

independently in the Israeli-Palestinian context, as single-actor "Palestinian" suicide

terrorism campaigns. This ignores the diversity of perspectives and motivations that

groups like Fatah, Hamas, and Islamic Jihad have brought to the table over the last

decade. Condensing down multiple campaigns into a single campaign is likely to

eliminate evidence of diffusion where cross-group learning, especially given that the

groups in question operate in the same geographic locations and against the same

adversary, likely occurs. Therefore, the dataset is biased against finding any evidence of

diffusion processes.

While recent work has drawn attention to the idea of insurgent strategies affecting

the likelihood and duration of civil wars (Fearon and Latin 2003; Fearon 2004; see also,

Arreguin-Toft 2000), there are no comprehensive strategy codings for the population of

civil wars. This contrasts sharply with the interstate war literature, which generally

5
The use of Pape's data should not be interpreted as endorsement of Pape's methodology or conclusions.
There are serious selection issues with Pape's data, most notably the failure to include non-suicide
terrorism campaigns and the way some campaigns, like the current Israeli-Palestinian campaign, are coded.
Pape's data is also not cross-correlated with datasets like ITERATE or the MIPT dataset.

14
speaking has a richer theoretical tradition, but also a comprehensive set of strategy

codings (see, Stam 1996). For our purposes, we constructed a fairly crude dataset

cataloguing the use of insurgent strategies for each civil war in the Fearon/Laitin dataset.

Each war, then, was coded as a being characterized by an insurgent strategy (1) or not

(0).6 For this preliminary exercise, we relied on Clodfelter (2000) and Philips and

Axelrod (2005) as general references for each of the civil wars. Wars were coded as

having an insurgent strategy if they met the following criteria:

1) Well developed political organization with discernible political goal
2) Guerrilla tactics (avoidance (at least initially) of direct confrontation with
government forces)
3) Rural base

This conceptualization is, of course, the Maoist vision of insurgency. While restrictive, it

does allow us to test the idea of diffusion more carefully as it prescribes several

observable elements (guerrilla tactics and rural base, for example) that distinguish it as a

specific technology (or software) of warfighting. But to be clear, the presence of

guerrilla tactics alone is not sufficient evidence of an insurgent strategy as we have

defined it. Weaker military actors have adopted guerrilla tactics for centuries (if not

millennia). Sniping from trees and ambushing static defenses or barracks are time-

honored weapons of the weak. We are more interested the diffusion of the specific

Maoist vision of revolution and resistance.

There are, however, difficulties with this approach. One persistent frustration is

the degree to which all rebel groups were described as "guerrillas" and "insurgents" (with

shocking disregard for the sensitive restrictions of our coding rubric!). Secondly, in the

general reference sources we consulted, there is often rather little information about the

6
This coding was applied to the entire duration of the war; variations over time were not catalogued.

15
rebels themselves. Information concerning their organizational structure or political

platform was not readily available for many of the smaller wars that made it into the

Fearon/Laitin dataset. Obviously, future work will need to make use of more descriptive

literature on specific case histories. There is no reason to believe that the omissions by

Clodfelter or Philips and Axelrod would skew the data in our favor. If anything they

privilege large (possibly insurgent) civil wars, which would lead to a fairly steady rate

over time. As discussed below, that is not what the data reveals.

IV. Results

Both the suicide bombing and insurgency data show preliminary evidence of a

diffusion effect in their loosely S-shaped curves (figures 3 and 4).7 Both are also

characterized by volatility at the annual level, with zeros resulting in several aesthetic

faux pas. We therefore rely on the 5-year moving average lines as a way of summarizing

the trends presented in each graph. This is especially necessary in the suicide bombing

case. Given the small longitudinal period in which suicide bombers have operated, single

years where groups did not conduct attacks can have a large impact on the results,

meaning moving averages are the most appropriate way to show trends.

The suicide bombing results are remarkably consistent. Whether looking at the

number of active campaigns or the number of groups who have engaged in suicide

bombing, we see evidence of a latency period followed by rapid adoption after a period

of 8 or 10 years. We do not, however, see the leveling off that is commonplace after the

social system becomes "saturated," leaving few additional adopters. We discuss this in

7
It is importance to recognize that the existence of any clear mathematical function that could describe the
spread of a given innovation is amenable to a diffusion approach. The S curve is merely the most "classic"
and well-known example, given its prevalence in the natural and social world.

16
more detail below, but one obvious reason may be the continuous creation of additional

adopters in the form of new terrorist groups. As mentioned above, the diffusion literature

considers both new and total adopters to be valid measures of diffusion activity. If we

restrict ourselves to only the number of new adopters no clear pattern emerges. There is

some evidence of a small increase after an initially stagnant period, in the five-year

average. But with few observations per year and a compressed timeline, the data do not

support any strong conclusions with regards to the rate of adoption.

Figure 3: Total Number of Active Suicide Campaigns by Year: 1980-2003

7
Active Suicide Terrorism Campaigns
6

5 5 per. Mov. Avg. (Active Suicide Terrorism
Campaigns)
4

0
r

01
a
Ye

17
Figure 4: Total Number of Active Insurgencies Per Year: 1945-2003

40
Total # Insurgencies
35

30 Total CW ongoing

25
5 per. Mov. Avg. (Total #
20 Insurgencies)

0
1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000

The insurgency data, displayed in Figure 4, tells a largely similar story. In terms

of the total number of insurgencies at any given point in time, we again see the familiar

S-curve. However, instead of leveling off, we observe a marked decline. We discuss the

many potential reasons for this trend below, but, sufficed to say, we believe the end of the

Cold War plays a prominent role. This post 1990 trend is supported in the graphs

documenting the percentage of all civil wars that are fought with insurgent strategies

(Figure 5) and the percentage of new wars fought with insurgent strategies. In the latter

two, we also see an uptick in the number of insurgent wars after 1996 or so. So if you

split apart Figure 5 into two component parts, there are almost two curves. A first curve

that begins after World War II and tails off at the end of the Cold War and a second curve

that begins in the mid-1990s.

18
Figure 5: Percent of all Civil Wars fought with Insurgent Strategies

100.00%

90.00%

80.00%

70.00%

60.00%

50.00%

40.00% % of Insurgencies out of Total CW
30.00%
5 per. Mov. Avg. (% of Insurgencies out of
20.00% Total CW)

10.00%

0.00%
1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000

While this could be an interesting finding if the trend persists, for now it seems to

be an artifact of a shrinking denominator: there are fewer new and ongoing wars overall

each year. As in previous decades, however, insurgent wars persist longer than their less

organized counterparts, thus increasing their percentage over time.

This issue of persistence causes some problems, however. As noted by other

authors when debunking the post-Cold War "lift the lid" theory of civil war explosion,

many of these wars had accumulated over time. In fact, many of these accumulated wars

are insurgencies. Thus, part of the increase revealed in the total wars graph (figure 4) is a

result of the cumulative nature of these wars: many persist for an exceptionally long time

(see, Burma). We are reassured, however, by the information offered in the graphs,

which take into account the total number of new and ongoing wars. While these graphs

do not fully share the S-shape in figure 1, they also document the secular decline in

19
insurgencies after 1991. This question continues to provide interesting grist for the mill,

as many authors (Toft 2000 and Hammes 2004 among others) suggest that insurgency is

the strategy of choice for nearly all rebels because it is best able to defeat strong

adversaries, like state militaries. These results suggest that this is only part of the story.

But why?

V. Discussion

Both the suicide bombing campaign trends identified in Figure 3 and the available

evidence on suicide bombing supports the utility of a diffusion approach to suicide

bombing. The figure highlighting active suicide campaigns per year shows a relatively

clear S curve when measured in terms of the moving average of suicide campaigns over a

five-year period. As noted above, the only exception is that suicide bombing as a

strategy does not appear to have leveled off the way diffusion theorists would generally

predict. However, there are two possible explanations for the lack of leveling, both of

which still support a diffusion approach. First, suicide terrorism could still be on the

upswing as a terrorist strategy, meaning the lack of leveling is to be expected. Second,

the success of suicide terrorism as a strategy, or at least the perception that it is a

successful strategy, may mean it is unlikely to level off at any point in the near or

medium term as it continues to inspire new groups to adopt such tactics.

Pape's explanation for suicide terrorism, like Toft's explanation of insurgencies,

is highly functionalist � terrorist groups adopt suicide bombing because they think it

works. However, while Pape's failure to study terrorist groups that do not use suicide

bombing clearly places limits on the explanatory value of the evidence, the data he

20
presents strongly supports a mixed diffusion approach. For example, Pape argues that the

initial suicide bombings by Hezbollah played an important role in influencing the Tamil

Tigers (LTTE) in two important ways. First, as evidence uncovered by Israel during its

original invasion of southern Lebanon in the early 1980s demonstrated, Tamil separatists

had established some training camps in Lebanon, highlighting the direct transmission of

ideas between Hezbollah (and potentially the PLO) and the Tamil Tigers. This would be

a classic case of internal diffusion (Pape 2005: 142; Council on Foreign Relations 2004).

Second, there seems to be direct evidence that the leader of the LTTE, Prabhakaran,

specifically viewed the 1983 Hezbollah suicide attack on the US Marine barracks in

Lebanon as a classic terrorist success case and sought to emulate that success by adopting

similar tactics. This would be a case of external diffusion, similar to functionalist

explanations, in that the perception that the strategy "worked" led another group to adopt

similar tactics (Pape 2005: 142; Hoffman and McCormick 2004: 259). An external

diffusion model also appears to be at work in the spread of several "innovations" in

suicide bombing, like the jacket lined with explosives, originally utilized by the LTTE

but quickly adopted by Palestinian groups like Hamas in the mid-1990s.

Other linkages between suicide terrorism groups, both strong and weak, highlights

the validity of a diffusion approach. First, one difference between the first Chechen

rebellion against Russia and the second is the use of suicide bombing tactics by the

Chechen's in the second conflict. Same group, same conflict, different tactics. Why?

The answer appears to be a part-ideational/part-functional explanation. Al Qaeda

operates have reportedly trained many of the second-generation Chechen rebels, with the

relationship cemented in part by renewed religious devotion on the part of the Chechen

rebels, demonstrating clear internal diffusion between Al Qaeda and the Chechen's. On

21
the other hand, the need for financing and the desire to produce the publicity that

generally comes with suicide bombing campaigns also likely influenced the Chechens.

But regardless of the motivations, the diffusion of tacit knowledge of suicide bombing

strategy from Al Qaeda to the Chechen's seems to have occurred.

Second, we know that Hezbullah has also explicitly trained several Palestinian

groups over the years, including Hamas and Palestinian Islamic Jihad (Catignani 2005;

Shatz 2004; Moghadam 2003). These ties were originally established when officials

from Hamas and PIJ were expelled from Israel and took refuge in Lebanon in the late

1980s and early 1990s, which heavily influenced the post-Oslo suicide campaigns in

1993 and 1994. One effect of Israel's near-term success in thwarting the Second Intifada

has been to push Hamas in particular back into the arms of Hezbullah. Recent

descriptions of rocket attacks and roadside bombs are almost certainly the result of

renewed coordination between the two groups (Fishman, 2005c; BBC 2004; Barel, 2004).

In this instance, then, we see the diffusion of both suicide and quasi-insurgent tactics

between two similarly situated rebel groups.

It is important to note the way Pape's selection on the dependent variable,

meaning he only studies suicide terrorism, not non-suicide terrorism, makes it difficult to

make analytically correct judgments about the diffusion of terrorism. It is not possible to

create figures for suicide terrorism comparable to the insurgency figures because we do

not currently have adequate data on non-suicide terrorism groups.8 The lack of data on

non-suicide terrorism groups skews the results because we cannot tell the percentage of

groups that adopt suicide terrorism or compare the spread of information about suicide

8
While it would probably be a relatively simple analytical task to put it together from other existing
datasets, it would be a bit labor intensive and is beyond the scope of this paper.

22
bombing tactics to groups whose existence overlaps with the suicide terrorism era but

have chosen not to utilize the method, like the IRA and ETA. This means the results are

stacked against finding evidence of diffusion, so the evidence this paper has described

means a diffusion perspective is especially likely to be effective. Given the available

data, Figure 6 below demonstrates one theory as to possible linkages between the suicide

terrorist groups identified by Pape.

Figure 6: Network Chart of Suicide Terrorist Groups Over Time, 1980-2003

Hezbullah:
Hezbullah 1982-1986

LTTE: 1990-2001
LTTE
Hama Hamas II:
Hamas s I: 2000-Present
1994-
1997

BKI BKI

PKK PK PK

IJ I Islamic Jihad
Islamic Jihad II: 2000-

Al Qaeda Al Qaeda: 1996-
Chechens:
Chechen 2000 -
This is more recent � not Kashmir:
Kashmir posited as AQ origin 2000-
Iraq:
Iraq 2003-

1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002

Solid Line: Direct Contact, Dotted Line: Indirect Contact (Mass Media)

The most important finding with regard to insurgencies is the degree to which

they fall off as civil war strategies in the 1990s. It is possible to argue that the decline of

23
insurgency is not an actual decline in groups interested in insurgency strategies, but the

result of several long-lasting insurgencies finally coming to and end. Indeed, this is true.

For example, insurgencies in Cambodia, El Salvador, Peru, and the Philippines ended in

this period. Of the 11 civil wars that terminated between 1990 and 1993, 6 were

insurgencies; of the 23 that ended between 1990 and 1995, 8 were insurgencies. Clearly

this die-off contributes to the overall dip. But it's not the whole story. The figures

detailing the percent of all wars fought with insurgent strategies and the number of new

insurgencies also reveals a marked decline in the 1990s. This is all the more telling given

that it occurs in the context of a sharp (but short-lived) spike in the total number of new

civil wars (Figure 5). So despite an increase in the overall number of wars, we still see a

decrease in both total and new insurgencies.

In contrast to previous decades, new rebel groups in the 1990s did not turn to

classical insurgent strategies with nearly the same enthusiasm as did those in the 1960s

and 1970s. The post-Cold War insurgencies took place in: Somalia, Afghanistan, Nepal,

Aceh (Indonesia), Israel/Palestine, and Iraq. Several of these countries experienced

previous civil wars dominated by insurgent strategies, suggesting that the organizational

legacy of those previous wars played a role in the strategic choice of the current ones.

Really, only the Maoists in Nepal newly adopted the now old-school insurgent tactics.

In many ways, these trends are not surprising. Seen within its original context,

insurgencies were part and parcel of a particular political ideology--communism--and

we should therefore not be surprised that similarly situated groups adopted Mao's

strategy. Indeed, that is wholly consistent with the diffusion framework. In Roger's

words, these groups are "homophilic", making them more receptive to adopting ideas

from each other. The end of material superpower support for these groups combined

24
with the bankruptcy of communism's ideational content to seriously diminish Maoist

insurgency as an attractive strategic alternative. Here, we should keep in mind the

stringent demands of Mao's version of revolution vis-�-vis its alternatives: the amount of

time and resources devoted to a well-developed political organization, the exceedingly

long time horizons for eventual victory, etc. If one removes the financial backers, tactical

trainers, and ideological allure of Maoist revolution, there is rather little to keep the train

on the tracks.

This version of events is not, however, the one told within mainstream political

science circles. Here, insurgency has been divorced from its ideational roots. Toft's

vision of groups adopting these techniques in an effort to mimic the success of previous

insurgencies (in effect, being the only real option if one wishes to topple an existing

government) is not able to fully explain the decline we see in the 1990s. To square his

explanation, he would need to document the realization that such insurgent strategies

were, in fact, no longer successful. We are not aware of any strong evidence that this is

the case. It seems, then, his functionalist explanation is, at best, incomplete.

One way of attempting to rescue this explanation is to briefly indulge the greed-

grievance debate that has dominated much civil war research in recent years. In its

simplest terms, these authors (Collier and Hoeffler 2000; Fearon and Laitin 2003) argue

that civil wars occur where they can--where environments are most permissive. They do

not occur, as a rule, due to overwhelming grievances against the government. Fearon and

Laitin take this a step further and argue that civil wars occur under conditions that favor

insurgency--weak central government, oil revenues, and difficult terrain (mountains or

forests). These authors are working under broader, more colloquial definition of

25
insurgency, but many of the points still hold. How do these arguments dovetail with

Toft's explanation of strategic choice?

These authors would point out that under the conditions prevalent in many states

experiencing civil wars in the 1990s, the meticulous political organizing required by the

Maoist strategy were unnecessary. Indeed, it may have been counter-productive. They

would point to changing war aims as a way of explaining why these groups would

eschew the insurgent model. Rebel groups no longer seek to control the full apparatus of

the state; they are less interested in governing than in pillaging. Under these conditions

rebels don't need popular legitimacy, and they thrive in this lawless environment. Thus

while there no system shock that convinced rebel groups that insurgent strategies failed,

political conditions changed such that these groups were no longer interested in the end-

state that these strategies were designed to bring about.

This interpretation is largely driven by the experiences of countries like Liberia,

Sierra Leone, and the DRC. It is far from clear that all modern rebel groups resemble

marauding armies or diamond smugglers. In fact, neither Fearon and Laitin nor any other

quantitative researchers have been able to replicate Collier and Hoeffler's natural

resource finding. Greed plays a role, but does not tell the whole story.

And, of course, the effectiveness of any given alternative is part of the diffusion

story. Obviously new adopters evaluate the potential benefits of a new technology (or

strategy) before adopting it. Essentially, insurgency adoptions appear to have occurred

when both material and ideological factors point in the same direction. This argues for a

mixed diffusion model that incorporates elements of both the internal and external

models in its functional form. But regardless of the specific balance of material and

ideational factors, the process going on is definitely one of diffusion: groups both learn of

26
the strategies of others through the media and through direction communication. This

means the "independence" assumption of the limited existent insurgency research is

flawed and a diffusion perspective is more appropriate.

VI. Conclusion

Our initial results give good credence to using a diffusion approach to understand

the spread of rebel or terrorist tactics and strategy. While it is clear that many groups

evaluate the effectiveness of such tactics in a rational manner, the functionalist approach

is incomplete. The ideational content underlying group linkages in both the suicide

bombing and Maoist insurgency cases is a powerful driver in determining the global

pattern of adoption.

The preliminary nature of our results (especially those using the insurgency data)

should give us caution. We do not wish to prematurely close the door on alternative

explanations. In particular, we remain concerned that the non-independence assumptions

we discussed above may be even stronger (and more pernicious) than we originally

anticipated. The ability of a successful demonstration of one or both of these strategies

has the ability to create (or at least energize) new terrorist or rebel groups. This is not a

common dynamic in the mainstream diffusion literature. Further work along the lines

laid out here will have to deal with the interaction between strategic choice and war onset

before it can fully validate one theory or another. Despite the large-N nature of the work

presented here, the micro-level detail needed to answer these questions still calls for

rather detailed case analysis (especially if one takes the time-series component seriously).

27
Thus while the diffusion framework provides what appears to be a useful tool for

systematic analysis of these phenomena, it cannot take the place of careful case work.

Despite these caveats, we are strongly encouraged by the results presented here.

They are a first step toward unpacking the question of how groups choose the strategies

and tactics that they do and encourage scholars to think much more clearly about the

micro-level behavior of rebel and terrorist groups. As our data document, there is a

significant amount of variation within this population of cases. But it is important to

think systematically about the elements that link these groups as well. The diffusion

framework points us in several key directions as we explore the degree to which new

ideas spread amongst groups that are more or less similar.

This first cut, though tentative, provides some strong support for the power of

ideas in influencing the diffusion of insurgency and suicide bombing strategies. While

insurgent strategies spread throughout the system during the 1960s and 1970s, they

quickly tailed off in the early 1990s. Contrary to expectations, the ideological content of

this military strategy was quite relevant. Similarly, while suicide bombing has shown its

effectiveness in the Middle East and South Asia, it is nearly absent from Latin American

and Europe. These were and are not ideas spreading like wildfire across the breadth and

width of the globe. Rather there appear to be somewhat significant limitations on the

population of potential adopters. We encourage further exploration of both the ideational

and material constraints that affect strategic decision-making among these groups. The

combination of a robust framework like the one presented here and detailed casework

may yet reveal uncover the why and how of rebel violence.