Is Terrorism Really a Weapon of the Weak? Debunking the Conventional Wisdom

Abstract

That terrorism is a “weapon of the weak” is such deeply held conventional wisdom it has become almost a cliché. “Weak” means many different things in the literature, however, and little rigorous empirical research has tested the contention that weaker groups, however conceived, are more likely to employ terrorism. This article explores prominent weapon of the weak arguments to develop testable hypotheses about group strength and the prevalence of terrorism. Using measures of deliberately indiscriminate attacks on civilians by rebel groups in civil conflicts, as well as multiple measures of rebel strength, it examines systematically whether weaker groups are more likely to employ terrorism. I find surprisingly little empirical support for the conventional wisdom. There is no clear or consistent evidence that deliberately indiscriminate terrorism is a weapon of the weak rather than the strong.

Keywords

terrorism civil wars rebellion capabilities civilian casualties conflict internal armed conflict power

Terrorism is frequently said to be a “weapon of the weak.”¹ Indeed, this conception is so common in the terrorism literature that it is almost a cliché.² However, little empirical work has tested whether weaker groups are more likely to employ terrorism than stronger ones; it is simply taken as a given. Is the truism that terrorism is a weapon of the weak true?

The conventional wisdom is also undertheorized. What people mean by “weak” varies considerably, and however it is conceived, the rationale for why weaker groups should be more likely than strong ones to employ terrorism is rarely articulated. This article does not propose new theory so much as explore prominent existing arguments to elicit clear empirical hypotheses about strength and the prevalence of terrorism. I test these in the context of civil conflicts. I focus on deliberately indiscriminate terrorism as the form of terrorism for which weapon-of-the-weak arguments should be most applicable.

While some variations of the argument fare better than others, I find remarkably thin support for the deeply held conventional wisdom that terrorism is a weapon of the weak, despite numerous research decisions designed to make for an easy test of the argument. Given the ubiquity of the claim in the academic literature, policy discussions, and journalism, this absence of evidence is substantively important.

Defining “Terrorism”

Terrorism is a loaded term, making definitions notoriously contentious. As another cliché goes, ‘one person’s terrorist is another’s freedom fighter,’ perhaps particularly so in the context of civil wars. The term is also often used inconsistently, in politically biased ways. I define terrorism as political violence against public civilian targets to influence a wider audience. In this project, I focus on deliberately indiscriminate attacks on civilians. This definition captures aspects considered fundamental in the terrorism literature.³ Like many, but not all definitions of terrorism, mine focuses on deliberate attacks on civilians, excluding attacks on military and state targets that all rebels conduct by definition.⁴

I narrow the focus further to exclude forms of violence against civilians that almost all rebel groups (and almost all governments involved in civil wars) engage in, namely, violence to induce civilian cooperation or deter collaboration with the enemy.⁵ Much of the civilian targeting and “one-sided violence” literatures, including prominent work by Weinstein (2007), Kalyvas (2006), Wood (2010), Hultman (2007), and others, focus on such violence to control the population. This type of violence is ubiquitous in civil wars (Stanton 2016, 30), but is not what we normally think of as “terrorism.” Focusing instead on deliberately indiscriminate violence, I seek to capture the inherent randomness that makes terrorism so terrifying.⁶

This narrower lens excludes some attacks that are commonly thought of as terrorism, however, including assassination of prominent civilian figures. It is an open question whether such attacks have the same causes or effects as deliberately indiscriminate attacks on random civilians. While it makes sense for some research purposes to include these discriminate attacks on civilians, I exclude them here for two reasons. First, there is no bright line between such attacks and those on “ordinary” collaborators, making it difficult to distinguish discriminate terrorism from this more ubiquitous type of violence. While I use the term “terrorism” here as shorthand for deliberately indiscriminate terrorism, I leave it to other researchers to decide for their purposes whether discriminate forms of terrorism should be included in their own definitions and measures. Second, some weapon-of-the-weak claims hinge specifically on the (in)ability of a group to discriminate. Focusing only on indiscriminate violence thus makes for an easier test of the conventional wisdom.

The definition used here is based on the tactics used by organizations, that is, the types of attacks they carry out, not the cause for which they fight. Rebel groups may be “terrorists” and “freedom fighters” simultaneously.

Terrorism in the Context of Civil Wars

Civil conflicts provide a valuable testing ground for arguments about terrorism. Rebel groups constitute a universe of comparable cases – all motivated by serious political grievances, organized, and willing to use violence against the existing political order – across which the phenomenon of interest varies.

However, a drawback to this approach is its selection of organizations strong enough to appear in data sets on armed conflicts. The very weakest groups are thus omitted. This study ameliorates this problem, first, by employing as low a threshold of violence as possible with existing systematic data. All groups involved in conflicts causing as few as 25 battle deaths in a year are included. Second, it examines groups that eventually meet this threshold even in years they do not. An examination of groups who never meet this threshold is not possible because we do not have an appropriate comparison category. Existing databases of terrorism allow us to identify the very weakest groups who do use terrorism, but not those who do not.⁷ Nonetheless, this limitation poses an important scope condition for the conclusions of this study.

Hypotheses: What Does it Mean to Say that “Terrorism is a Weapon of the Weak”?

The phrase “weapon of the weak” is bandied about in the literature, but not everyone means the same thing by it. Below, I unpack the main ways the concept is used, drawing out their empirical implications.

Weakness is baked into some conceptions of terrorism, particularly in attempts to distinguish it from civil war by defining or labeling as terrorism only low levels of violence conducted by groups too weak to wage full-blown insurgency or to control territory (Cronin 2006, 31-32; Guelke 1995, 30-31; Merari 1993, 217, 225; Schmid and Jongman 1988, esp. 13-18; Silke 1996; Sambanis 2008; Sánchez-Cuenca and de la Calle 2009, 34). As Merari (1993, 245) puts it, “One might say, that all terrorist groups wish to be guerillas when they grow up.” If terrorism is defined by weakness, then any attempt to explain terrorism with reference to group strength is tautological. By these definitions, we also could not refer to deliberately indiscriminate attacks by groups like ISIS at the peak of their strength, or the LTTE in Sri Lanka, as “terrorism.” Excluding states from the definition of terrorism contributes to the tautology problem. Non-state militant organizations are almost always weaker than the governments they oppose. If only non-state actors use terrorism, then it is a weapon only of the weak almost by definition (Jackson 2008, 27).⁸

However, non-tautological arguments can be drawn from the literature. In its most general sense, the weapon-of-the-weak notion refers to terrorism as a tactic employed by groups less powerful than their opponent. “Terrorism is the poor man’s airforce” as the saying goes. Sometimes this is a justification for terrorism (Chaliand and Blin 2016). For example, Sayeed Siyam of Hamas (quoted in Sontag 2002) justified suicide bombings: “We do not own Apache helicopters ourselves, so we use our own methods.” Or as Peter Ustinov quipped: “Terrorism is the war of the poor, war is the terrorism of the rich.” While this may be an accurate description of how terrorist groups justify their actions to themselves and others, it does not explain variation in the use of terrorism by rebel groups, because virtually all opposition groups are considerably weaker than the governments they face. Vanishingly few rebel groups own Apache helicopters, yet not all resort to terrorism. Moreover, the only recent rebel group to wield air and naval forces, the LTTE, was notorious for its use and innovation of terrorist tactics.

Nonetheless, there is variation in the extent to which rebel groups are weaker than their opponents. The logic of the weapon-of-the-weak argument is often only implicit. In its most basic version, it suggests that because it is cheaper and easier to hit civilian targets than military ones, and easier to target indiscriminately than selectively, weak groups can carry out terrorism, even if they are less capable of other types of attacks (Bueno de Mesquita 2013; Crenshaw 2011, 41-2; Merari 1993, 225-226, 245).

The weaker rebels are militarily, relative to the government, the more they employ terrorism.

However, others (Wood 2014a; Asal et al. 2009; Enders and Sandler 1993) suggest a countervailing logic in which organizations with greater military strength are more capable of carrying out more attacks or killing many civilians. Both dynamics could be true:

H1a

Stronger rebels are less likely to use terrorism in the first place, but carry out more attacks and kill more people if they do so.

Some focus on state capacity, arguing that we observe terrorism in stronger states because in weaker ones the opposition can mount a (non-terrorist) guerilla campaign (Sánchez-Cuenca and de la Calle 2009, 39; Hendrix and Young 2014). Others speculate that poorly equipped government forces allow rebels to wage effective guerrilla campaigns, obviating the need for terrorism (Laitin and Shapiro 2008, 213).⁹

Rebels facing more capable states employ more terrorism.

By a similar logic, Laitin and Shapiro (2008, 213) argue that groups who enjoy favorable conditions for insurgency, such as rough terrain, should be less likely to resort to terrorism.

Rebels fighting on rough terrain use less terrorism.

Sánchez-Cuenca and de la Calle (2009, 32) argue that terrorism is more likely by groups who lack territorial control, and less likely against weaker states that lose territory to rebels. (See also de la Calle and Sánchez-Cuenca 2012; 2015; Cronin 2009, 147; and on territorial control and indiscriminate violence more broadly, Kalyvas 2006).

Rebels who control territory are less likely to use terrorism.

Others argue that groups with fewer supporters do not have “safety in numbers” and so turn to underground tactics such as terrorism rather than other forms of dissent (DeNardo 1985, 230; McCormick 2003, 483; Crenshaw 2011, 41-2; Bueno de Mesquita 2013; Lake 2002). Insurgency is also an “underground” form of dissent, so this argument, like many in the terrorism literature, may explain civil conflict rather than terrorism per se. But it is plausible to think that unpopular rebels turn to terrorism out of desperation because insurgency requires more humanpower (Lins de Albuquerque 2014; Sánchez-Cuenca and de la Calle 2009, 44; Della Porta and Tarrow 1986, 620 & 628), or because reliance on popular support restrains groups from targeting civilians (Fortna, Lotito, and Rubin 2018).

The less popular support a rebel group enjoys, the more terrorism it uses.

Relatedly, many argue terrorism serves to mobilize support (Pape 2003; 2005; Arblaster 1977, 422), particularly in the early stages of a conflict, before “graduating” if possible, to guerrilla, and ultimately conventional, warfare (McCormick 2003, 485; Bueno de Mesquita 2013, 20; Cronin 2006, 32).¹⁰ Others suggest terrorism is used early in a conflict to destabilize the government and prepare the way for revolution (Crenshaw 2011, 118; Neumann and Smith 2005, 577ff; Thornton 1964 90ff.).

Terrorism is more likely in early stages of a conflict.

Note that if terrorism is used to mobilize support, the causal arrows between strength and terrorism run both ways: if initial weakness leads to terrorism, but terrorism, assuming this mobilization strategy works (an open empirical question), increases strength (Lake 2002).

Finally, some argue that groups who are losing turn to mass killing and civilian targeting to signal resolve and raise the opponent’s cost for winning (Hultman 2007; Wood 2014b), out of desperation in the face of defeat, or later in the conflict, after other methods fail (Downes 2008; Crenshaw 2011, 113; Valentino et al. 2004; Della Porta and Tarrow 1986).

H7a

Terrorism is more likely by groups on the verge of defeat.

H7b

Terrorism is more likely in later stages of a conflict.

Surprisingly little empirical work has evaluated the relationship between terrorism and strength, conceived of in any of the ways outlined above. Those studies that do examine the conventional wisdom empirically (sometimes only in passing) come to contradictory conclusions (Balcells and Stanton 2021, 54-55).

Stanton (2013) and Metelits (2010) provide empirical support for the general argument (H1) that terrorism is a weapon of the militarily weak, but Salehyan et al. (2014, 650-651) find strength increases civilian targeting, while Goodwin (2006), Fortna (2015) and Fazal (2018, 201-202) find no effect. Results are similarly contradictory on the effect of state capacity (H2), with Chenoweth (2010) providing indirect support, Hendrix and Young (2014) finding opposite effects of military capacity and bureaucratic capacity, and Coggins (2015) finding no relation to state failure.¹¹ Research on economic development (sometimes used as a proxy for state capacity) and terrorism is similarly mixed: some (Abadie 2006; Gassebner and Luechinger 2011) find no effect, while Enders and Hoover (2012) find a curvilinear relationship.

The rough terrain hypothesis (H3) has not been explored in detail, though Fortna (2015) found little effect. Territorial control has been found to decrease both civilian and indiscriminate targeting (H4) (Salehyan et al. 2014; Kalyvas 2006), but also to increase terrorism lethality (Asal and Rethemeyer 2008).

To my knowledge, there are no cross-national empirical studies of the relationship between popular support and terrorism (H5). Contra H6, Findley and Young (2012a) find less terrorism in the period before civil war than during, and Lewis (2020) finds civilian targeting rare in initial stages of insurgent group formation. In line with H7, Hultman (2007) and Wood (2014b) both find battlefield losses associated with greater civilian targeting; Polo and Gonzalez (2020) find that the effect of battlefield losses on terrorism is conditional on government repression and in-versus out-group dynamics. Valentino et al. (2004) argue that groups turn to mass killing when conventional means fail. However, Stanton (2016) finds that with few exceptions, groups who use terrorism as a systematic tactic do so throughout the conflict.

The lack of consistent findings suggests that the relationship between rebel strength and terrorism requires further analysis. Moreover, many of the most systematic studies listed above examine civilian targeting writ large. The empirical literature is further hindered by a tendency to conflate the effects of strength on terrorism specifically with effects on political violence more generally. By examining indiscriminate terrorism in the context of civil wars, this study isolates the effects of strength on tactical choice more cleanly. This should make support for weapon-of-the-weak arguments easier to observe, as negative effects of rebel strength on terrorism will not be crowded out by positive effects on discriminate violence or civil conflict.

Beyond the lack of clear empirical support, there is a theoretical gap in the link between strength and terrorism. If weak groups employ terrorism because it is cheap and easy, why would strong groups not also do so? Terrorism must either be less costly or more beneficial for weak groups than for strong ones. Most weapon-of-the-weak arguments have not made such a case. If terrorism entails high legitimacy costs (Fortna 2015), perhaps only those desperate in the face of defeat (H7) risk it in a gamble for resurrection? Perhaps groups with little popular support to begin with are less constrained by the legitimacy costs of terrorism? (Wood 2010, 602). However, this contradicts the notion that groups use terrorism to generate popular support when it is lacking.

The terrorism literature is littered with assertions that terrorism is a weapon of the weak, but the theoretical logic of this link is not well spelled out, nor is there clear empirical evidence for the claim.

Data and Research Design

I evaluate terrorism by rebel groups in intrastate armed conflicts from 1970-2013, using the Terrorism in Armed Conflict data set (TAC) (Fortna, Lotito, and Rubin 2018; 2020).¹² TAC covers 409 rebel groups in 166 conflicts in 97 countries. The main unit of observation here is the active group-year, for an N of 2,022.¹³ I include, but drop in robustness tests, the conflict between the US and al-Qaida even though it is unlike any other cases of intrastate conflict in the UCDP data and is a clear outlier (see TAC Codebook). Doing so biases in favor of the weapon-of-the-weak hypotheses.

Dependent Variable

How we conceptualize the “amount” of terrorism can significantly change findings (Young 2019). I examine the effects of rebel weakness on both the number of deliberately indiscriminate terrorism incidents and the total number of fatalities in those attacks. In robustness checks, I exclude the many incidents in GTD in which no one is killed, and subsequently limit to mass fatality incidents only.

TAC measures of terrorism exclude attacks against military or government targets, and use information on specific attack and target type and subtype as proxies to identify the intentionally indiscriminate attacks on which I focus here. I examine the robustness of results to both TAC’s less and more restrictive measures of deliberately indiscriminate attacks (see Fortna, Lotito, and Rubin 2020 for details). Attributing terrorist incidents to rebel organizations is often not straightforward. TAC provides a systematic and flexible way for researchers to vary inclusion levels when attributing GTD incidents to particular rebel groups. I use several TAC versions of this attribution process, focusing here primarily on two: TAC versions B and E. Version B attributes to a rebel group perpetrators deemed direct matches, armed wings, and factions or umbrella organizations. Weak groups may more likely commit attacks for which GTD is unable definitively to identify a perpetrator (because weaker groups are less likely to take responsibility by claiming an attack,¹⁴ or because they are more obscure and less reported on). I therefore also examine version E, which includes the many generic descriptor perpetrators in GTD that apply to the group in question (e.g., “Kurdish separatists” vis-a-vis the PKK in Turkey). In robustness checks, I also examine version A (direct matches only) and version F, which includes attacks by unknown perpetrators that could plausibly be attributed to a rebel group. The latter is extremely noisy, including many “false positives,” but does not discard the majority of attacks in GTD, for which no perpetrator is listed.

Coefficient plots below present the results across the eight versions of the dependent variable: combining versions B and E; more and less restrictive proxies of deliberately indiscriminate attacks; for incidents and fatalities. Also shown are a selection of the many robustness checks run to ensure that decisions about how to measure terrorism, or other modeling choices, are not driving results.

As Figure 1 shows, counts of terrorist incidents and fatalities are extremely skewed. Contrary to popular perception, deliberately indiscriminate terrorism is more the exception than the rule in civil conflict. In 54% of active group-years there are no incidents; in 62% there are zero fatalities from terrorism. When it occurs, it tends to be at low levels. In three-quarters of group-years, fewer than five incidents occur, and fewer than ten people are killed. But the tail of the distribution is very long.¹⁵ The maximum number of incidents is 291 (Syrian insurgents in 2013).¹⁶ Other high incident counts are ISIS in Iraq in 2013 (274), and Sendero Luminoso in Peru in 1984 (278). In the fatalities distribution, Al-Qaida in 2001 is a massive outlier at 2,793.

Contrary to the conventional wisdom, many groups that do not use terrorism are quite weak (e.g., FARF and MDJT in Chad, NDF in Yemen, and EPDM in Ethiopia), while some relatively strong groups (the NPFL in Liberia, RPF in Rwanda, the LTTE in Sri Lanka) do.

Independent Variables

To test the most straightforward weapon-of-the-weak argument (H1), I use two variables from Cunningham, Gledistsch, and Salehyan’s (CGS) Non-State Actor data (2009; 2013). The first measures rebel weakness relative to the government as a composite of three factors: relative ability to mobilize supporters, to procure arms, and fighting capacity. I use the original ordinal scale from much weaker to much stronger, and separately, dummy variables to compare much weaker groups, and the few groups (7%) at parity or stronger, to the modal category of weaker groups. Second, is the fighting capacity component on its own, again both ordinal and a dummy comparing no/low to moderate/high. (See online appendix for descriptive statistics).

Following Hendrix and Young (2014), who find support for the weapon-of-the-weak argument when state capacity is conceived of as military capacity, but not as bureaucratic/administrative capacity, I use the natural log of the Correlates of War (COW) CINC measure of the state’s share of global military capabilities (v.5.0) (Singer 1987) to test H2. I also examine COW’s measure of military personnel per capita (logged). US doctrine holds twenty troops per population thousand as a rule of thumb for effective counterinsurgency (COIN). I thus also use a dummy variable for state militaries meeting this threshold.¹⁷ Higher values indicate relatively weaker rebels.

I invert two conflict-level measures of rough terrain from PRIO to study terrain that is less rough or more accessible, thus undermining rebels’ advantage (H3): the proportion of the conflict area that is flat terrain (not mountainous), and the proportion that is unforested (Tollefsen et al. 2012; Hallberg 2012; Buhaug and Gates 2002).¹⁸ H4 is tested with CGS’ dummy measure of territorial control, inverted so the higher value denotes weakness.¹⁹

Two variables indicate rebels’ popular support (H5): first, CGS’ mobilization capacity measure, based on the size of the population rebels claim to represent, and connection with the local population, as opposed to dependence on external support²⁰ (measured ordinally, and with dummies comparing low and high to moderate levels); second, ACD2EPR’s coding (Vogt et al. 2015) of whether a majority of the relevant ethnic population supports the rebels (available for ethnic conflicts only, constituting 58% of the total).

To test whether terrorism is more likely in early stages of a conflict (H6), I include years since the first battle-related death, inverted so higher values denote younger conflicts.²¹ If a group targets (almost) only civilians at the start of its struggle (as a strong version of this hypothesis and the terrorists-hope-to-grow-up-to-be-insurgents argument would suggest), these early years will not be captured because civilian targeting is excluded from the 25 battle death threshold. In other words, the terrorist phase of the conflict may occur in the years before those analyzed, potentially biasing results away from H6. To remedy this, I also examine TAC data for years before active conflict. For all groups, I include the years between the first battle death and the year the 25 battle death threshold is crossed (423 group-years). In conflicts with an obvious match between a UCDP group and a GTD perpetrator before UCDP records the first battle death, I include observations starting in the year of the first GTD incident, adding 322 prewar group-year observations (for 46 groups). Because this latter set of observations are added only for groups that used terrorism at some point before UCDP records the start of conflict, this “selects on the dependent variable.” Analyses of the prewar years are thus biased toward finding support for H6.

Measuring whether a group is on the verge of defeat (H7a) is difficult without fine-grained data on battle outcomes or the trajectory of conflicts. As a rough proxy, I code group-years within 2 years of a rebel defeat (and alternatively, those within 1 year, or years of defeat only). I create two measures of rebel “defeat” from Kreutz’ (2010) Conflict Termination dataset: one counts only the 85 government victories; the other also includes wars ending in “low activity” as this often denotes rebels who have been largely defeated though not eliminated outright (312 cases fit this more expansive definition of defeat).²² If terrorism is a successful gambit for groups otherwise on the verge of defeat, it should make that defeat less likely. This measure will thus miss the very cases where it succeeds. However, empirical examples of this phenomenon are lacking. An alternative (though also imperfect), measure: time from the end of the conflict (inverted so higher values denote later stages), tests whether terrorism is more likely in late stages of a conflict, after other types of tactics have failed (H7b).²³

Control Variables

I control for factors that might drive both the relative strength of rebels and the prospect of terrorism. These include democracy and the extent to which human rights are protected, which may affect the state’s ability to counter terrorism, and motives for terrorism.²⁴ External support for rebels directly affects their relative strength, and their incentives to turn to terrorism (Salehyan et al. 2014; Fortna, Lotito, and Rubin 2018). Incompatibility, whether the war is fought over government or territory, may affect both the resources the state contributes to the fight, and the tactics rebels use. Stronger groups are likely to be involved in larger and more intensive conflicts, which are also more likely to see more incidents of all kinds, including terrorism; failure to include a measure of intensity would thus bias away from weapon-of-the-weak arguments. The number of rebel groups involved in a conflict may affect the relative strength of each one (how fractured the movement is), while competition among groups is thought to lead to outbidding through terrorism (Bloom 2004; but see Findley and Young 2012b). Finally, I control for time period, first, to account for the possibility that the Cold War shifted the relative resources available to governments and rebels (Kalyvas and Balcells 2010) and that terrorism may have decreased after the Cold War (Enders and Sandler 1999; Chenoweth 2010); second, because GTD changed its data collection procedures as the project moved across institutions.²⁵ (Control variable measurement details in the online appendix.)

Some of these variables arguably affect rebel strength and are affected by it, which could lead to attenuated results for the effects of rebel weakness on terrorism. I thus divide the control variables into those that are plausibly “post-treatment,” for which this may be a problem, and those that are not. The latter category includes controls for democracy and human rights protection (both measured ex ante), time period, and incompatibility.²⁶ For the three control variables plausibly post-treatment: external support, intensity, and the number of rebel groups, I test whether post-treatment bias is driving null results by dropping these from the model and adding them back in, one at a time (results in online appendix).

As additional robustness checks I, first, drop the US-al Qaida case. Second, I drop civil conflicts that began as coup attempts. These uprisings by former members of the military are likely relatively strong militarily, and they are empirically associated with significantly less terrorism. Retaining them in the primary analyses thus biases the analysis toward finding support for the weapon-of-the-weak hypothesis. Third, I include controls for region because GTD is thought to have focused on some regions more than others, with less coverage of Africa in particular (Lutz and Lutz 2013).

Model

Because the dependent variables are counts (of incident or fatalities) for which by far the most common value is zero, I employ a zero-inflated negative binomial (ZINB) model.²⁷ ZINB regression consists of two parts: an inflate model estimating whether the outcome is zero, and a count model estimating the number of incidents or fatalities. This allows us to separate effects on whether terrorism is used at all from how much it is used, which may be generated by different processes.

I present coefficient and 95% confidence interval plots, for incidents and fatalities, for each version of the dependent variable discussed above: versions B and E, less and more restrictive, and a sample of robustness checks. Lines correspond to the models identified in Table 1. Fuller results for primary analyses are available in supplemental materials.

Table 1.

Key to Coefficient/Confidence Interval Plots.

Line of figure	TAC attribution version	TAC proxy of indiscriminate targeting
1	B	Less restrictive
2	E	Less restrictive
3	B	More restrictive
4	E	More restrictive
5	A	Less restrictive
6	F	Less restrictive
7	B	Less restrictive	Drop US-al Qaida case
8	B	Less restrictive	Drop coups
9	B	Less restrictive	Add regional controls
10	B	Less restrictive	Include only fatal incidents*
11	B	Less restrictive	Include only mass incidents*

Lines in the plots correspond to models combining different TAC attribution versions and less and more restrictive measures of deliberately indiscriminate terrorism, as well as other robustness tests. Attribution Version A is direct matches; Version B includes armed wings, factions, umbrellas; Version E includes generic descriptors; Version F includes unknown perpetrators.

*Robustness tests in lines 10 & 11 apply only to incident measures, not number of fatalities.

Note that in ZINB models, positive inflate coefficients mean positive associations with zero incidents, i.e., negative effects on terrorism; while positive count coefficients indicate positive associations with terrorism. Except for two sets of categorical variables (rebel strength, and mobilization capacity) in which high and low dummies are compared to a middle category, the independent variables are all coded such that larger values denote weaker rebels. Therefore, weapon-of-the-weak hypotheses expect negative coefficients in inflate models and positive coefficients in count models. Visually, this means that weapon-of-the-weak arguments would be supported by coefficients toward the inside of each count/inflate pair in the graphs below. The same is true of the weakest or lowest value of the categorical variables, and the opposite holds for the strongest/highest category (marked ^§). However, variant H1a expects negative coefficients in both models.

Results

Figure 2 shows only very partial support for the most straightforward weapon-of-the-weak hypothesis (H1): that the lower the military capability of the rebels, relative to the government, the more terrorism is likely to be used. Coefficients for the ordinal measure of rebel weakness (top left) are generally insignificant, and sometimes not even in the direction expected. While coefficients for the inflate models are usually in the expected (negative) direction, they are not statistically significant. Count model coefficients are not always positive. Only in the robustness check for Version F of the TAC data, which noisily matches even unknown perpetrator attacks to a rebel organization (line 6), is there a significant positive count coefficient.

Breaking this categorical variable into dummies (top right) shows no consistent or significant difference between weaker (the omitted mid-level category) and much weaker rebels. In some robustness checks, the very weakest groups may even be responsible for less terrorism, with coefficients flipping negative in some count models and positive in some inflate models. However, we see more evidence of the expected relationship when we compare the rare rebel groups as strong or stronger than the government, with the mid-level category (bottom part of this top right graph). This strongest category is responsible for lower incident counts, significantly so in some models. And these strongest rebels are less likely to use terrorism at all (positive inflate coefficients), again significantly so in some models. The picture is similar for fatalities, though here, the negative effect in count models is less robustly significant than the positive effect in inflate models (indicating a lower likelihood of killing anyone through terrorism). For this small set of extremely strong rebels, some evidence thus supports the weapon-of-the-weak hypothesis.

The lower half of Figure 2 examines fighting capacity alone rather than the composite measure. Whether ordinal or dichotomous, this measure provides little support for H1. Coefficients in the inflate portion of the models generally show the expected (negative) sign, but are almost never significant at conventional p ≤ 0.05 levels (though they come close or just barely cross this threshold in some models). Coefficients in the count portion of the models are rarely significantly positive (as H1 would expect), and for fatalities are sometimes even significantly negative. In sum, we see a surprising lack of consistent or robust support for the primary weapon-of-the-weak hypothesis (H1).

We see partial support for H1A, which expects weaker groups to be more likely to employ terrorism but to kill fewer people (i.e., negative coefficients in both models). However, this pattern emerges only for terrorism fatalities, not for incidents, nor the more general measure of weakness. Nor is it strongly robust across models.

Hypothesis 2 concerns the military strength of the state (Figure 3). When this is measured using CINC scores, coefficients are generally but not always in the expected direction. We see significant coefficients in the correct direction in a few specific robustness tests (e.g., in the inflate model for terrorist incidents when regional controls are included, or for version F), but support is scarce. That coefficients are significant both here and above for relative military strength with version F, which includes any unknown perpetrator attacks in the country, suggests this may be an artifact of terrorism by other groups or individuals in large powerful countries, rather than weakness of a particular rebel organization.

Evidence for H2 when state strength is measured with military personnel per capita is no less tepid. Coefficients are negative, as expected, in the inflate models, but these are rarely significant, and coefficients in the count models, while occasionally positive, are often negative, especially for terrorism fatalities. There is even less support for H2 when we use the twenty troops per thousand rule of thumb for measuring government counterinsurgency capacity.

Hypothesis 3, that rebels who do not enjoy the benefits of rough terrain will be more likely to use terrorism, is not supported (top of Figure 4). There is no consistent or significant effect of flatter terrain on terrorism. Rebels fighting in unforested terrain are also no more likely to employ terrorism. Indeed, here there is more support for the opposite conclusion. Positive and often significant coefficients for inflate models, especially for terrorism incidents, indicate that rebels disadvantaged by fighting on open terrain are, if anything, less likely to use terrorism.

We see no effect of territorial control (H4) on terrorism, undermining the conventional wisdom. Coefficients are generally (but not always) in the expected direction and consistently insignificant.

Figure 5 shows results for rebels’ popular support. Using mobilization capacity as a proxy, whether an ordinal measure or dummies, we see no evidence for hypothesis 5. Coefficients are insignificant or in the wrong direction, sometimes significantly so. For example, lower mobilization capacity is associated with a lower likelihood of using any terrorism, significantly so when wars that emerge from coups are dropped from the analysis. Comparing low to medium mobilization capacity, we generally see the expected negative inflate coefficients for fatalities, but positive coefficients for incidents. Count coefficients are often positive and sometimes significant, but not robustly so. For high mobilization capacity (bottom part of this top right graph), effects are generally the opposite of what hypothesis 5 would expect. The most popular groups are, if anything, more likely to use terrorism, not less (recall that here, coefficients are expected to be toward the outside of the plot).

This hypothesis fares somewhat better in identity conflicts where we can measure majority support for rebels among their ethnic group. We continue to see no consistent or significant effect on incidents. For fatalities, coefficients are generally in the correct direction (except in version F) and sometimes significant in count models, but not significant and sometimes veer positive in inflate models. At least in ethnic conflicts, rebels with low popular support are not necessarily any more likely to turn to terrorism or to carry out more attacks, but they may kill more people with terrorism than groups with higher levels of support. Even here however, results are not consistent across versions of TAC. It is notoriously difficult to measure popular support for groups that are illegal and often clandestine. But the common notion that militant groups who cannot mobilize widespread support are most likely to employ terrorism is, at best, only partially and inconsistently supported by the available data.

The evidence for hypothesis 6, that terrorism is used by groups early on, before they have built up strength is interestingly mixed (Figure 6). Coefficients are consistently positive in inflate models, consistently significant for fatalities, but less so for incidents. Groups are less likely to employ terrorism in their early stages as they build strength, the opposite of what hypothesis 6 would predict. But at least for fatalities, the count coefficients are also positive, sometimes, though not robustly, significantly so. This means that rebels are less likely to kill anyone with terrorism early in a conflict, but if they do so, they kill more people. This rather odd set of findings provides partial support for H6, but interestingly runs against the logic of hypothesis 1A (which expects weaker groups to be more likely to attempt terrorism, but less capable of killing many people).

Because the selection of active conflict years may bias against this hypothesis, I also compare levels of terrorism in the earliest years of a conflict to those after it crosses the threshold of 25 battle deaths. Lack of data for many control variables for preconflict years precludes multivariate analysis, but bivariate difference-of-means tests show no support for the notion that groups use terrorism before they “graduate” to insurgency. The right side of Figure 6 also shows means and confidence intervals, for incidents and fatalities, comparing active and prewar years. Levels of terrorism are significantly (p ≤ 0.0001) lower, rather than higher, in the prewar years, for all versions of the dependent variable.²⁸ That this is true despite the selection bias toward finding terrorism in prewar observations gives us more confidence that H6 does not hold water.

I find even less support for the common view that groups turn to terrorism in desperation when facing defeat (H7a). Looking first at whether a group is within 2 years of defeat by the government (Figure 7, left side), the consistently negative and generally significant coefficients in count models and generally positive coefficients in inflate model are the opposite of what this hypothesis expects. This finding debunking the desperation hypothesis is generally consistent across robustness checks, and when an alternate measure of defeat that includes cases that end in “low activity” (not shown) is used. Moreover, results (also not shown) are even more consistently significant in the wrong direction for this hypothesis if we look only at the year of defeat or that year and 1 year prior. As noted above, these measures are potentially problematic: they would miss cases in which rebels avoided imminent defeat by resorting to terrorism. This strategy would have to succeed more often than not to mask evidence for H7a. I am unaware of any such cases, let alone many.

The results are similarly dire for H7b (right side Figure 7), which expects more terrorism toward the end of conflicts, as groups turn to it after other tactics fail. Except for the very noisy version F, coefficients are never significant in the expected direction, and are often significant in the wrong direction. Together, the findings on H6 and H7b suggest, contrary to the weapon-of-the-weak hypotheses, that terrorism is greatest in the middle years of a conflict.

In sum, there is remarkably little evidence to support the weapon-of-the-weak hypotheses. For no conception of weakness is there consistent, significant, and robust support. What support there is is tenuous or partial at best. This is notable given that multiple modeling and data choices favored these arguments.

For some measures of strength, we see glimmers of support for the conventional wisdom. The rare rebel group rated as strong or stronger than the governments is less likely to employ terrorism. Among ethnic conflicts, groups with less popular support kill more people with terrorism, but are no more likely to turn to terrorism in the first place, nor responsible for a larger number of incidents. For most measures, we see only an occasional significant effect in the expected direction, but no consistent or robust support. We see absolutely no support for notions that rough terrain or territorial control reduce the use of terrorism, nor for the idea that rebels use terrorism early in their fight before graduating to other methods, or conversely, out of desperation on the verge of defeat or after other methods have failed.

Do Null Results Mean No Effect?

This article debunks the conventional wisdom that terrorism is a weapon of the weak based on null results. Is there really no true effect of rebel strength, or could tests simply be underpowered, with a small-N leading to null results? Because this analysis is based on a census, the full universe of cases of rebel groups involved in civil conflicts over more than 40 years, not a sample, “sample size” is of less concern. If measures of rebel strength are not strong predictors of terrorism in this universe of cases, it is hard to argue that deliberately indiscriminate terrorism is a weapon of the weak.

Nonetheless, it is worth investigating confidence levels in the null effects; to distinguish absence of clear evidence from clear evidence of absence. I do so using Rainey’s (2014) two one-sided t-test (TOST) method for “arguing for a negligible effect” (see also Gross 2015). This approach flips the burden of proof by switching the null hypothesis to a substantive effect rather than zero effect. The graphs for each variable measuring rebel weakness in Figures 8A and 8B plot the probability that the true coefficient lies within an equivalence range (-m to +m), for each version of the dependent variable. Following Hartman and Hidalgo (2018), I use 0.36σ, where σ is the standard deviation of the independent variable in question, as a baseline (indicated by vertical lines in each graph) against which to assess the equivalence range of a substantively negligible effect. The horizontal line toward the top of each graph marks 95% for reference. This is a conservative reference, as only effects away from zero in one direction would undermine the argument that terrorism is not a weapon of the weak.

Above we saw a non-negligible effect for some measures of weakness, but in the wrong direction – this is true of unforested terrain and verge of defeat, for example. A low probability of a negligible effect for these measures is thus unsurprising and provides no defense of weapon-of-the-weak arguments. For some measures where we found little evidence of a significant effect supporting weapon-of-the-weak hypotheses, we also see much more uncertainty about there being a truly negligible effect. For low fighting capacity, for example, the probability of a count model effect smaller than +/− 0.36σ is below 25% for terrorist incidents, between about 20 and 50% for fatalities, and even lower for inflate model effects. It is similarly low for flat terrain. The probability of a substantively negligible effect in the count models is above 50% but still well below 95% for measures such as mobilization capacity and relative rebel weakness (again slightly lower in inflate models). For early stages of a conflict, it is over 75% and for some versions of the dependent variable, over 95% in the count model (but under 50% in the inflate model). For CINC scores it is consistently over 95% in the count model.

What can we conclude from this? In general, there is more uncertainty about whether weakness affects the use of terrorism at all (inflate models), than whether it affects the amount of terrorism (count models). For many versions of the weapon-of-the-weak hypothesis, we cannot reject the standard null hypothesis of no effect, but neither can we with great confidence reject the flipped null hypothesis of a true effect. That is, while the conventional wisdom is thrown very much in doubt, we cannot conclude with certainty that there is no effect. For other measures, there is a strong preponderance of evidence against the conventional wisdom. And for a few, e.g., CINC scores, we can confidently conclude that weakness has a truly negligible effect on terrorism incidents and fatalities.

Caveats

Several caveats bear reiterating. First, while this study includes very low-levels of conflict, it excludes groups so weak they cannot mount a fight that ever meets the 25 battle death threshold. The very weakest of the weak are thus selected out. Among organized militant groups that cross this low threshold, we can conclude that weapon-of-the-weak arguments evidence little support. However, we cannot rule out the possibility that at the very lowest end of the strength continuum, terrorism might be more likely than in the portions of the continuum examined here. Assessing this claim requires systematic data on organizations (whether or not they use terrorism) below this low threshold, which does not currently exist. Both Findley and Young’s (2012a) finding that terrorism occurs predominantly in rather than outside the context of civil wars (that by definition cross this threshold), and Lewis’ (2020) study showing that groups rarely target civilians in their initial stages of formation before they have built up some strength, should give us greater confidence that this selection is not hiding a true weapon-of-the-weak effect.

Second, as is generally true in cross-national studies of civil conflicts, some measures and proxies are far from perfect. Even measures of relative military strength that ostensibly vary annually are more static than the reality they purport to measure.²⁹ Measurement error in independent variables can bias results toward null findings. Better data may yet prove the conventional wisdom correct. However, we do not see stronger evidence for weapon-of-the-weak arguments for the measures in which we might have more confidence (e.g., rough terrain or territorial control), nor can poor measures account for findings that suggest a negative relationship (as opposed to a negligible one) between weakness and terrorism.

Third, it is possible that other forms of terrorism not examined here, such as assassinations, are associated with weaker groups. However, selective forms of terrorism are more challenging than attacks on random civilians. It is indiscriminate attacks that theory suggests should be most associated with weakness.

Finally, but more obviously, this study is observational, not experimental. Some unobserved variable may drive both strength and terrorism, hiding a genuine causal connection. I have endeavored to control for factors that might create a spurious (non)relationship. However, it is neither possible nor ethical to manipulate rebel group capacities to test in a “causally identified” way the relationship between weakness and terrorism. While a systematic analysis is beyond the scope of this article, anecdotally, the most important exogenous shift in relative strength in civil wars, the end of the Cold War, provides no support for the weapon-of-the-weak argument. Rebel groups strengthened by the sudden loss of Soviet support to the governments they fought, such as UNITA in Angola and Renamo in Mozambique, increased their use of terrorism from 1989-1991, while levels of terrorism by rebels weakened by their loss of Soviet support, e.g., the FMLN in El Salvador or the URNG in Guatemala, either dropped or remained the same.

Conclusion

These caveats notwithstanding, we see astonishingly little empirical support for the deeply seated conventional wisdom that terrorism is a weapon of the weak. For some authors, this notion is true by definition. But if we take weapon-of-the-weak arguments to be non-tautological explanations for why some groups resort to terrorism while others do not, we should expect to see a positive relationship between various aspects of group weakness and the decision to target civilians in deliberately indiscriminate ways.

This article elicits what people mean by “weak” when they say that terrorism is a weapon of the weak. For some, it is a matter of relative military capability – that militarily weaker groups, relative to the governments they fight, should be more likely to resort to terrorism, as should those fighting more powerful and militarily capable states. There is limited and inconsistent support, at best, for these arguments. The rare rebel groups that reach parity with government forces are responsible for lower levels of terrorism, but even here there are exceptions, with relatively strong groups, like the TPP in Pakistan, and al-Shabaab in Somalia, employing terrorist tactics. And there are many cases of relatively very weak groups that do not use indiscriminate terrorism (CNR in Chad, KDP in Iraq, ABSDF in Myanmar, BLF in Pakistan, FAR in Guatemala are among dozens of examples). For the vast majority of groups, neither relative strength and fighting capacity nor state strength have any discernible effect. There is similarly only tepid and inconsistent evidence that unpopular groups use more terrorism.

Meanwhile, arguments that rebels who enjoy the advantages of rough terrain or who control territory use less terrorism are not at all supported in the data. Nor is there any evidence that terrorism is more likely early in a conflict, before a group gains enough strength to wage insurgency, or conversely, toward the end of conflict after other methods fail, or in desperation by a group on the verge of defeat.

For many measures of weakness, we cannot conclude with certainty that there is no effect, and more testing with better measures is clearly necessary. But this study casts serious doubts on a deeply held conventional wisdom. Knowledge of a group’s strength provides a surprisingly poor predictor for policy-makers or scholars attempting to ascertain whether rebel groups will turn to terrorism. Simply put, there is no clear evidence that deliberately indiscriminate terrorism is a weapon of the weak.

Supplemental Material

Supplemental Material - Is Terrorism Really a Weapon of the Weak? Testing the Conventional Wisdom

Supplemental Material for Is Terrorism Really a Weapon of the Weak? Testing the Conventional Wisdom by Virginia Page Fortna in Journal of Conflict Resolution

Footnotes

Acknowledgements

Special thanks to Nick Lotito and Mike Rubin for our work together on the data that made this article possible, to Laura Resnick for research assistance on war termination data, to Naoki Egami for advice on testing for negligible effects, and to Merlin Heidemanns for research assistance with TOST. Thanks also to discussants and seminar participants at ISA, George Washington, the University of Maryland, Princeton, Rice, and Notre Dame for comments on earlier versions of the paper, and to two anonymous reviewers whose suggestions have made this research much stronger. I am grateful to more people than I have space to thank here for their advice and support on the larger project of which this article is a part.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by Columbia University Institute for Social and Economic Research and Policy; Seed Grant on “The Causes and Consequences of Terrorism.”

Replication Materials

The data used in this article, as well as replication “.do” files for the data set construction and analysis, and an with additional information, will be made available on the JCR website.

ORCID iD

Virginia Page Fortna

Notes

References

Abadie

Alberto

. 2006. “Poverty, Political Freedom, and the Roots of Terrorism.” American Economic Review 96 (2): 50-56.

Arblaster

Anthony

. 1977. “Terrorism: Myths, Meaning and Morals.” Political Studies 5 (3): 413-24.

Asal

Victor

Rethemeyer

R. Karl

. 2008. “The Nature of the Beast: Organizational Structures and the Lethality of Terrorist Attacks.” Journal of Politics 70 (2): 437-49.

Asal

Victor

Rethemeyer

R. Karl

Anderson

Ian

Stein

Allyson

Rizzo

Jeffrey

Rozea

Matthew

. 2009. “The Softest of Targets: A Study on Terrorist Target Selection.” Journal of Applied Security Research 4 (3): 258-78.

Balcells

Laia

Stanton

Jessica A.

. 2021. “Violence Against Civilians During Armed Conflict: Moving Beyond the Macro- and Micro-Level Divide.” Annual Review of Political Science 24 (1): 45-69.

Berman

Eli

. 2009. Radical, Religious, and Violent: The New Economics of Terrorism. Cambridge MA: MIT Press.

Bloom

Mia M.

2004. “Palestinian Suicide Bombing: Public Support, Market Share, and Outbidding.” Political Science Quarterly 119 (1): 61-88.

Bueno de Mesquita

Ethan

. 2013. “Rebel Tactics.” Journal of Political Economy 121 (2): 323-57.

Buhaug

Halvard

Gates

Scott

. 2002. “The Geography of Civil War.” Journal of Peace Research 39 (4): 417-33.

10.

Chaliand

Gérard

Blin

Arnaud

, eds. 2016. The History of Terrorism: From Antiquity to ISIS. Berkeley: University of California Press.

11.

Chenoweth

Erica

. 2010. “Democratic Competition and Terrorist Activity.” Journal of Politics 72 (1): 16-30.

12.

Chenoweth

Erica

. 2013. “Terrorism and Democracy.” Annual Review of Political Science 16 (1): 355-78.

13.

Coggins

Bridget

. 2015. “Does State Failure Cause Terrorism? An Empirical Analysis (1999-2008).” Journal of Conflict Resolution 59 (3): 455-83.

14.

Crenshaw

Martha

. 1981. “The Causes of Terrorism.” Comparative Politics 13 (4): 379-99.

15.

Crenshaw

Martha

. 2011. Explaining Terrorism. New York: Routledge.

16.

Cronin

Audrey Kurth

. 2006. “How Al-Qaida Ends: The Decline and Demise of Terrorist Groups.” International Security 31 (1): 7-48.

17.

Cronin

Audrey Kurth

. 2009. How Terrorism Ends: Understanding the Decline and Demise of Terrorist Campaigns. Princeton University Press.

18.

Crozier

Brian

. 1960. The Rebels: A Study of Post-War Insurrections. London: Chatto and Windus.

19.

Cunningham

David E

Gleditsch

Kristian Skrede

Salehyan

Idean

. 2009. “It Takes Two: A Dyadic Analysis of Civil War Duration and Outcome.” Journal of Conflict Resolution 53 (4): 570-97.

20.

Cunningham

David E

. 2013. “Non-State Actors in Civil Wars: A New Dataset.” http://ksgleditsch.com/data/NSAEX_codebook.pdf

21.

de la Calle

Luis

Sánchez-Cuenca

Ignacio

. 2012. “Rebels Without a Territory: An Analysis of Nonterritorial Conflicts in the World, 1970-1997.” Journal of Conflict Resolution 56 (4): 580-603.

22.

de la Calle

Luis

Sánchez-Cuenca

Ignacio

. 2015. “How Armed Groups Fight: Territorial Control and Violent Tactics.” Studies in Conflict & Terrorism 38 (10): 795-813.

23.

Della Porta

Donatella

Tarrow

Sidney

. 1986. “Unwanted Children: Political Violence and the Cycle of Protest in Italy, 1966–1973.” European Journal of Political Research 14 (5-6): 607-32.

24.

DeNardo

James

. 1985. Power in Numbers: The Political Strategy of Protest and Rebellion. Princeton: Princeton University Press.

25.

Department of the Army . 2006. Counterinsurgency. Field Manual 3-24.

26.

Downes

Alexander B.

2008. Targeting Civilians in War. Ithaca: Cornell University Press.

27.

Enders

Walter

Hoover

Gary

. 2012. “The Nonlinear Relationship between Terrorism and Poverty.” American Economic Review 102 (3): 267-72.

28.

Enders

Walter

Sandler

Todd

. 1993. “The Effectiveness of Antiterrorism Policies: A Vector-Autoregression-Intervention Analysis.” American Political Science Review 87 (4): 829-44.

29.

Enders

Walter

Sandler

Todd

. 1999. “Transnational Terrorism in the Post-Cold War Era.” International Studies Quarterly 43 (1): 145-67.

30.

Fazal

Tanisha M.

2018. Wars of Law: Unintended Consequences in the Regulation of Armed Conflict. Ithaca NY: Cornell University Press.

31.

Findley

Michael G.

Young

Joseph K.

. 2012a. “Terrorism and Civil War: A Spatial and Temporal Approach to a Conceptual Problem.” Perspectives on Politics 10 (2): 285-305.

32.

Findley

Michael G.

Young

Joseph K.

. 2012b. “More Combatant Groups, More Terror? Empirical Tests of an Outbidding Logic.” Terrorism and Political Violence 24 (5): 706-21.

33.

Fortna

Virginia Page

. 2015. “Do Terrorists Win? Rebels’ Use of Terrorism and Civil War Outcomes.” International Organization 69 (3): 519-56.

34.

Fortna

Virginia Page

Lotito

Nicholas

Rubin

Michael

. 2018. “Don’t Bite the Hand That Feeds: Rebel Funding Sources and the Use of Terrorism.” International Studies Quarterly 62 (4): 782-794.

35.

Fortna

Virginia Page

Lotito

Nicholas

Rubin

Michael

. 2020. “Terrorism in Armed Conflict: New Data Attributing Terrorism to Rebel Organizations.” Conflict Management and Peace Science 39 (2): 214-236.

36.

Gassebner

Martin

Luechinger

Simon

. 2011. “Lock, Stock, and Barrel: A Comprehensive Assessment of the Determinants of Terror.” Public Choice 149 (3-4): 235-61.

37.

Gleditsch

Nils Petter

Wallensteen

Peter

Eriksson

Mikael

Sollenberg

Margareta

Strand

Håvard

. 2002. “Armed Conflict 1946-2001: A New Dataset.” Journal of Peace Research 39 (5): 615-637.

38.

Goodwin

Jeff

. 2006. “A Theory of Categorical Terrorism.” Social Forces 84 (4): 2027-46.

39.

Gross

Justin H.

2015. “Testing What Matters (If You Must Test at All): A Context-Driven Approach to Substantive and Statistical Significance.” American Journal of Political Science 59 (3): 775-88.

40.

Guelke

Adrian

. 1995. The Age of Terrorism and the International Political System. London: Tauris.

41.

Hallberg

Johan Dittrich

. 2012. “PRIO Conflict Site 1989–2008: A Geo-Referenced Dataset on Armed Conflict.” Conflict Management and Peace Science 49 (2): 219-232.

42.

Hartman

Erin

Hidalgo

F. Daniel

. 2018. “An Equivalence Approach to Balance and Placebo Tests.” American Journal of Political Science 62 (4): 1000-1013.

43.

Hendrix

Cullen S.

Young

Joseph K.

. 2014. “State Capacity and Terrorism: A Two-Dimensional Approach.” Security Studies 23 (2): 329-63.

44.

Hoffman

Bruce

. 2006. Inside Terrorism. New York: Columbia University Press.

45.

Hultman

Lisa

. 2007. “Battle Losses and Rebel Violence: Raising the Costs for Fighting.” Terrorism and Political Violence 19 (2): 205-22.

46.

Jackson

Robert

. 2008. “An Argument for Terrorism.” Perspectives on Terrorism 2 (2): 25-32.

47.

Kalyvas

Stathis N.

2006. The Logic of Violence in Civil War. New York: Cambridge University Press.

48.

Kalyvas

Stathis N.

Balcells

Laia

. 2010. “International System and Technologies of Rebellion: How the End of the Cold War Shaped Internal Conflict.” American Political Science Review 104 (3): 415-29.

49.

Kreutz

Joakim

. 2010. “How and When Armed Conflicts End: Introducing the UCDP Conflict Termination Dataset.” Journal of Peace Research 47 (2): 243-50.

50.

Kydd

Andrew H.

Walter

Barbara F.

. 2006. “The Strategies of Terrorism.” International Security 31 (1): 49-80.

51.

Laitin

David D.

Shapiro

Jacob N.

. 2008. “The Political, Economic, and Organizational Sources of Terrorism.” In Terrorism, Economic Development, and Political Openness, edited by Keefer

Philip

Loayza

Norman

, 209-32. New York: Cambridge University Press.

52.

Lake

David A.

2002. “Rational Extremism: Understanding Terrorism in the Twenty-First Century.” Dialogue-IO 1: 15-29.

53.

Lewis

Janet I.

2020. How Insurgency Begins: Rebel Group Formation in Uganda and Beyond. Cambridge: Cambridge University Press.

54.

Lins de Albuquerque

Adriana

. 2014. “Violent or Nonviolent Means to Political Ends.” PhD Dissertation, Columbia University.

55.

Lutz

Brenda J.

Lutz

James M.

. 2013. “Terrorism in Sub-Saharan Africa: The Missing Data.” Insight on Africa 5 (2): 169-83.

56.

Mao

Tse-Tung

. 1961. (1937). On Guerilla Warfare. Samuel B. Griffith, trans. New York: Praeger.

57.

McCormick

Gordon H

. 2003. “Terrorist Decision Making.” Annual Review of Political Science 6: 473-507.

58.

Merari

Ariel

. 1993. “Terrorism as a Strategy of Insurgency.” Terrorism and Political Violence 5 (4): 213-51.

59.

Metelits

Claire

. 2010. Inside Insurgency?: Violence, Civilians, and Revolutionary Group Behavior. New York: New York University Press.

60.

Neumann

Peter R.

Smith

M. L. R.

. 2005. “Strategic Terrorism: The Framework and Its Fallacies.” Journal of Strategic Studies 28 (4): 571-95.

61.

Pape

Robert A.

2003. “The Strategic Logic of Suicide Terrorism.” American Political Science Review 97 (3): 343-61.

62.

Pape

Robert A.

2005. Dying to Win: The Strategic Logic of Suicide Terrorism. New York: Random House.

63.

Polo

Sara M. T.

González

Belén

. 2020. “The Power to Resist: Mobilization and the Logic of Terrorist Attacks in Civil War.” Comparative Political Studies 53 (13): 2029-60.

64.

Rainey

Carlisle

. 2014. “Arguing for a Negligible Effect.” American Journal of Political Science 58 (4): 1083-91.

65.

Salehyan

Idean

Siroky

David

Wood

Reed

. 2014. “External Rebel Sponsorship and Civilian Abuse: A Principal-Agent Analysis of Wartime Atrocities.” International Organization 68 (3): 633-61.

66.

Sambanis

Nicholas

. 2008. “Terrorism and Civil War.” In Terrorism, Economic Development, and Political Openness, edited by Keefer

Phillip

Loayza

Norman

, Cambridge: Cambridge University Press.

67.

Sánchez-Cuenca

Ignacio

de la Calle

Luis

. 2009. “Domestic Terrorism: The Hidden Side of Political Violence.” Annual Review of Political Science 12: 31-49.

68.

Schmid

Alex P.

Jongman

Albert J.

. 1988. Political Terrorism. Revised. Amsterdam: North Holland.

69.

Silke

Andrew

. 1996. “Terrorism and the Blind Man’s Elephant.” Terrorism and Political Violence 8 (3): 12-28.

70.

Singer

J. David.

1987. “Reconstructing the Correlates of War Dataset on Material Capabilities of States, 1816-1985.” International Interactions 14: 115-32.

71.

Sontag

Deborah

. 2002. “The Palestinian Conversation.” New York Times Magazine, February 3.

72.

Stanton

Jessica A.

2013. “Terrorism in the Context of Civil War.” Journal of Politics 75 (4): 1009-22.

73.

Stanton

Jessica A.

2016. Violence and Restraint in Civil War: Civilian Targeting in the Shadow of International Law. Cambridge: Cambridge University Press.

74.

START . 2016. GTD: Global Terrorism Database. National Consortium for the Study of Terrorism and Responses to Terrorism. https://www.start.umd.edu/gtd

75.

Thornton

Thomas Perry

. 1964. “Terror as a Weapon of Political Agitation.” In Internal War: Problems and Approaches, edited by Eckstein

Harry

, 71-99. London: Free Press.

76.

Tollefsen

Andreas Forø

Strand

Håvard

Buhaug

Halvard

. 2012. “PRIO-GRID: A Unified Spatial Data Structure.” Journal of Peace Research 49 (2): 363-74.

77.

Valentino

Benjamin

Huth

Paul

Balch-Lindsay

Dylan

. 2004. “‘Draining the Sea’: Mass Killing and Guerrilla Warfare.” International Organization 58 (2): 375-407.

78.

Vogt

Manuel

Bormann

Nils-Christian

Rüegger

Seraina

Cederman

Lars-Erik

Hunziker

Philipp

Girardin

Luc

, 2015. “Integrating Data on Ethnicity, Geography, and Conflict: The Ethnic Power Relations Dataset Family.” Journal of Conflict Resolution 59 (7): 1327-42.

79.

Weinstein

Jeremy

. 2007. Inside Rebellion: The Politics of Insurgent Violence. New York: Cambridge University Press.

80.

Wood

Reed M.

2010. “Rebel Capability and Strategic Violence against Civilians.” Journal of Peace Research 47 (5): 601-14.

81.

Wood

Reed M.

2014a. “Opportunities to Kill or Incentives for Restraint? Rebel Capabilities, the Origins of Support, and Civilian Victimization in Civil War.” Conflict Management and Peace Science 31 (5): 461-80.

82.

Wood

Reed M.

2014b. “From Loss to Looting? Battlefield Costs and Rebel Incentives for Violence.” International Organization 68 (4): 979-99.

83.

Young

Joseph K

. 2019. “Measuring Terrorism.” Terrorism and Political Violence 31 (2): 323-45.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.13 MB