Extremism and Terrorism: Rebel Goals and Tactics in Civil Wars

Extremism and terrorism are thought to go hand in hand in civil wars. Yet do they? Are rebel groups with more extreme goals more likely than moderate ones to use terrorism, as commonly assumed? Arguments linking extremism to terrorism are often circular: groups are tagged as extremist because they do extreme things. Our article addresses this problem by articulating a novel conceptualization of extremism as the distance of group goals from the status quo. Understanding the relationship between what groups want and how they try to achieve it has obvious theoretical and policy implications. We theorize mechanisms that might connect extremist goals to terrorism and use new data on rebel group aims in civil wars from 1970 to 2013 to examine the empirical relationship between extremism and terrorism in a nontautological way. The results show that some extremist goals are associated with terrorism but not others. Groups with goals that involve changing the political ideology of the state or transforming political power across identity groups are more likely to use terrorism or to use more of it. Secessionist groups, however, are no more likely to produce terrorism than are those with less extreme territorial aims such as autonomy.

attacks on civilians, focusing on those that are deliberately indiscriminate, regardless of the political goals or the cause for which the group fights.We elaborate our conceptions of both extremism and terrorism later in the article.
The terms "extremism" and "terrorism" have pejorative connotations, and some advocate not using them at all (Moore 2015).We do so because they are used by policy makers and by scholars in literatures with which we hope to engage.2However, we try not to let our own moral judgments shape our conceptualization or coding of extremism (Jackson 2019). 3For example, the African National Congress in South Africa is coded as extremist for aiming to end apartheid, even though we favor that goal.Whether one approves of extreme change depends on one's view of the status quo.Similarly, we code groups as using terrorism based on their actions, not on our view of the legitimacy of their aims.
We follow a growing number of studies that examine terrorism in civil wars (Belgioioso 2018; Findley and Young  2012b; Fortna 2015; Fortna, Lotito, and Rubin 2018;  Stanton 2013; Thomas 2014).Rebel groups fighting civil wars constitute a useful set of cases in which to examine the causes and consequences of terrorism because they provide variation in the phenomenon of interest (Fortna 2015, 521).By definition, all the rebel groups we consider seek to alter the status quo; not coincidentally (Staniland 2021), all are engaged in violence.Therefore, aims and tactics are correlated to some extent for groups selected into our study.But among this set of cases, some rebel groups have more extreme aims than others, and not all use terrorism.By focusing on this universe of cases, we aim to investigate the relationship between extremism and terrorism specifically, rather than the broader question of extremism and violence of all kinds.This broader question is important but much less tractable empirically.We restrict our analysis, for empirical and theoretical reasons described later, to extremism and terrorism on the part of rebels, rather than of the states they fight. 4e use the Terrorism in Armed Conflict dataset (TAC; Fortna, Lotito, and Rubin 2022) and an original Rebel Extremism Dataset (RED), coding group goals for nonstate actors involved in internal armed conflicts from 1970 to 2013.After reviewing the literature on extremism and terrorism, we define each term and articulate a conception of extremism that focuses on the extent of desired change from the status quo.We differentiate this from alternative conceptions of extremism, including the geographic scope of group aims and commitment to the cause, and discuss the relationship between extremism and ideology.Next, we draw on scholarship on the causes of terrorism to discuss theoretical mechanisms that link extremism with terrorism and develop testable hypotheses on terrorism in civil war.We then present our research design and data, followed by a discussion of the results.
We do not find a blanket relationship between extremist goals and terrorism, as the conventional wisdom would expect.Instead, the relationship is more complex.We find that among groups fighting for control of the central government there is an association between aiming for more radical change and terrorism, but it is a nuanced one.Meanwhile, among groups fighting over territory, those with extreme secessionist goals are no more likely than those fighting over autonomy to turn to terrorism in their struggle.We attribute this to countervailing pressures in the legitimacy costs of terrorism for groups seeking recognition from the international community.Data limitations mean that our empirical conclusions can only be tentative.However, by presenting a nontautological conception of extremism and terrorism, articulating mechanisms that might link them theoretically, and beginning to explore the relationship empirically, we aim to contribute to understanding of this important question.

Literature on Extremism and Terrorism
Despite a vast literature on terrorism, surprisingly little scholarship examines the relationship between extremism and terrorism directly.Some work simply conflates extremist and terrorist groups, treating them as synonymous (Chowdhury and Krebs 2010; Dalacoura 2006;  Juergensmeyer 2017; Nasser-Eddine et al. 2011, 9).Other arguments infer extremism from terrorism-terrorist behavior reveals extremist preferences (Abrahms 2006;  Libicki, Chalk, and Sisson 2007, 25).
Some studies draw implicit connections between the two without explicitly theorizing their relationship.For example, Hoffman's (2006) discussion of religious fundamentalism suggests that something about groups' extremist ideology is connected to their use of terrorism.Similarly, DeNardo (1985) argues that "purists" who remain in a struggle after moderates and pragmatists are selected out by their willingness to compromise are most likely to resort to terrorism but does not specify why this might be.Others simply assert a relationship without theorizing or testing it (e.g., Schmid 2014). 5ydd and Walter (2006) argue that extremist groups often use terrorism to derail an impending settlement between moderate groups and the government in a strategy of "spoiling." 6Lake (2002) argues that extremists, who by his definition lack strong support among the population, use terrorism in a strategy of provocation.However, although both articles explicate strategies of terrorism that might appeal to extremists, they also delineate other strategies of terrorism that less extreme groups might use (Lake explicitly notes that less extreme groups will use terrorism in other ways).Wintrobe (2006) argues that groups with more extreme goals are more likely to use extreme tactics such as terrorism because they are more likely to have indivisible goals, but his argument explains violence, not terrorism per se.Piazza (2009) distinguishes between groups with limited "strategic" goals and those with more extreme "universalist/abstract" goals, arguing that the latter use terrorism to demonstrate commitment to their cause and the purity of their struggle.Because these groups play to an amorphous and transnational audience, they are "much less inhibited when planning attacks," resulting in more lethal terrorism (65). 7 Iannaccone and Berman (2006,  125) contend, however, that religious extremism makes groups that choose violence and terrorism particularly effective, but it does not necessarily make them more likely to turn to terrorism.They state, "To equate religious extremism with religious militancy is a serious error" because many extremist religious groups eschew violence and even political mobilization. 8n theory, the PCVE literature distinguishes between extremism of ideas and of behavior such as terrorism.The term "violent extremism" implies that extremists could be violent or nonviolent; otherwise, the expression is redundant.In practice, however, the PCVE literature generally conflates extremism and violence, often using the terms interchangeably (Nasser-Eddine et al. 2011, 9; Stephens,  Sieckelinck, and Boutellier 2021, 347).
Despite a strong sense in the terrorism literature and in the popular imagination that extremism and terrorism go together, we lack a coherent theoretical or empirical investigation of how extreme goals might lead to extreme tactics.Before undertaking such an investigation, however, we need working definitions of both concepts that avoid tautology.
We follow Fortna, Lotito, and Rubin (2022) in defining terrorism by rebel groups as intentional violence against public civilian targets to influence a wider audience and in focusing within this definition on indiscriminate terrorism, rather than selective targeting. 9Selective violence is a ubiquitous strategy of "control" in civil wars to deter defections and induce cooperation (Stanton 2016)  10 and is the focus of much of the civilian targeting literature (Kalyvas 2006;  Weinstein 2007).We focus instead on deliberately indiscriminate attacks for several reasons.First, the theoretical logic of selective targeting, we posit, is likely different from the logic of indiscriminate attacks.Second, because they are ubiquitous, selective attacks to control the populace yield little variation across rebel groups.Third, the link between extremism and terrorism is arguably strongest for indiscriminate attacks.Intentional killing of random civilians, such as bombing marketplaces or public transportation, is the least legitimate and thus most "extreme" type of violence.We leave to future analysis whether extremism of goals is related to discriminate forms of terrorism such as assassinations or punishing suspected collaborators.Most important for our purposes, the definition of terrorism is emptied of any discussion of the causes for which groups fight.

Defining and Conceptualizing Extremism
Our objective is to develop a definition of extremism based on a group's goals, independent of its tactics.Goals represent preferences over political outcomes. 11Extremism denotes a position toward the end of some continuum, of being further out toward the extreme-but conceptions of extremism differ in the precise content of that continuum.Ours considers how far a group wants to change things along a single policy space (scale) or, relatedly, the number of policy spaces over which the group desires changes (scope).We distinguish this from other conceptions in the literature that focus on the territorial extent of desired change and commitment to the cause and then discuss the relationship between extremism and ideology. 12tremism as Scale and Scope of Distance from the Status Quo We define extremism as the scale and scope of the distance between a group's stated goals and the political status quo.Those who want to change things more dramatically are more extreme than those pushing for less drastic change.This conception of extremism involves the location of a group's preferences over what the state and its politics should look like-that is, the desired political end.But a key question is the reference point against which to measure extremity.Many conceptions of extremism implicitly consider it relative to the distribution of preferences in the population.Group preferences become more extreme the further they are from the median preference.Lake (2002, 18) explicitly defines extremism as preferences that lie in the tail end of the distribution of a society's population.Walter (2017, 16) also defines ideological extremism "in relation to the majority opinion of the affected population."Essentially, extremists are those whose goals are not very popular. 13his definition of extremism relative to median preferences or mainstream views is problematic, however, if we want to compare groups in societies with differing levels of polarization or where the political status quo favors a minority. 14In highly polarized societies, a significant segment of the population may share the preferences of a group that nonetheless advocates something far from the status quo.Is such a group more or less extreme than one in a different society that aims for something closer to the status quo, but that fewer people in the society support?For example, if a majority ethnic group, such as the Hutus, wants to depose a minority government run by Tutsis, is this less extreme because more people want the same thing than when a minority, such as the Mohajirs in Pakistan, presses for more government jobs and economic opportunities?The issue is effectively moot for this empirical project.The fine-grained public opinion data that would be needed to measure the distribution of preferences are scarce in the best of circumstances; in war-torn countries, obtaining accurate, cross-national public opinion data about whether people share the goals of illegal and violent organizations is likely infeasible.Note also that this conception often characterizes only one tail of the distribution as extreme.Considering opinions about the role of Islam in politics, Walter (2017), for example, portrays ideologies that are more conservative and fundamentalist than the majority as extremist but does not portray ideologies that are more liberal or secular than the majority as similarly extreme.
Our definition of extremism shares the conception of a continuum of preferences over desired political outcomes but measures extremism relative to the political status quo.The greater the distance between a group's desired political outcome and the status quo, the more extremist it is. 15his conception of extremism is similar to those that define extremism relative to median preferences, except that the referent baseline (the political, social, or territorial status quo) is observable.This conception is also agnostic as to the underlying distribution of societal preferences.
To illustrate, imagine two possible distributions of preferences over a contested policy space.In figure 1a, preferences are fairly normally distributed, and a rebel group with extreme preferences relative to the status quo (R E1 ) is further out in the tail of the distribution than a more moderate group (R M1 ).If the political status quo reflects something close to the population's median preference, then measuring extremism as distance from the status quo or as distance from median preferences is equivalent, and the assumption that extremists' goals are relatively unpopular follows.The shaded gray area in each figure indicates "fence-sitters" (discussed later) who could tip into one camp or the other.In figure 1b, however, preferences are highly polarized and bimodal.Such bimodal distributions are likely in countries afflicted by civil war.A rebel group (R E2 ) with similar distance between its preferences and the status quo, as was true for R E1 in distribution 1, might be near the median preference of a sizable chunk of the population.If R E2 represents an oppressed majority, it might be close to the median preference. 16Here, a more moderate group (R M2 ), although closer to the status quo, might be in the tail of this part of the bimodal distribution; extremist groups might be even more popular than more moderate groups among the population they claim to represent.By distinguishing extremism from popular support, we can evaluate the relationships of each concept to terrorism separately. 17ur discussion so far makes the simplifying assumption that whatever is contested in a conflict falls along a single policy dimension.It is straightforward, however, to relax this assumption.Rebel groups aiming to change many things can be considered more extreme than those with more limited aims over fewer policy spaces.Ideologies that call for a bundled set of changes to the status quo over many aspects of society are extreme in this sense: the overall extent of the desired change, relative to the status quo, is greater.

Alternative and Related Concepts
Commitment to the Cause.One conception of extremism that is often conflated with our own is intensity of commitment to a group's goals, whatever they are.Extremism is often thought of as intransigence or an unwillingness to compromise (DeNardo 1985, 231).In this view, extremism is akin to resolve: a willingness to continue to bear the costs of a struggle to achieve one's desired ends. 18f we assume that settlements are more possible when the positions of the rebels and the government are not so far apart, then unwillingness to settle may be functionally similar to distance from the status quo.Theoretically, however, the conceptions are different.One could hold moderate preferences in our conception-that is, be close

Figure 1 Extremism and Population Preferences
to the status quo-but be dead set on achieving them no matter how high the cost.Or one could prefer greater change but be willing to settle for much less given the costs of fighting.Depth of commitment, or intransigence, does not automatically correlate with the location of preferences.
This conception avoids circular arguments between extremism and terrorism, but we do not focus on it here for two reasons.First, it differs analytically from our conception of extremism, and there are already other terms for it: commitment to the cause or resolve.Second, as a practical matter, commitment is difficult to observe and measure a priori.We cannot know until negotiations are attempted how willing rebels are to reach a deal.Settlement also depends on the preferences and strategy of the government, other parties in the conflict, and international actors (Walter 2002; Wood and Kathman 2015).Commitment is thus difficult to measure in a way that is not based on observed behavior and is susceptible to the tautologies we are trying to avoid.
Territorial Extent of Change.Another conception of extremism concerns the extent of the geographic area over which a group desires to enact change.For example, Buhaug (2006, 694) describes the goal of overthrowing the government as more extreme than the goal of autonomy or secession for only a part of the state.Universalist or transnationalist ideologies such as Marxism or fundamentalist Islamism are often described as extremist because of their geographic scope.Yet it is not universalism or transnationalism per se that makes an ideology extreme.Some "moderate" ideologies such as humanism are universalist, and in the West, the dominant liberal, democratic, capitalist ideology is also transnationalist (in ways sometimes used to justify violence). 19It is again with explicit reference to the status quo that extremism can be discerned.Of those who want to change the status quo, transnationalist and especially universalist groups want to enact change over a larger geographic or territorial scope.In this alternative conception, it is more extreme to want to change whatever it is you want to change throughout a bigger swath of the globe. 20deology and Religion.Ideology and extremism are often linked and sometimes conflated.Ideology, itself an ambiguously defined concept, refers to "a set of beliefs about the proper order of society and how it can be achieved" (Walter 2017, 15). 21In common parlance, "ideological" often implies extremism.But calling a group "ideological" cannot by itself distinguish among political organizations; all organizations have some ideology, although it may be implicit (Gutiérrez Sanín and Wood 2014, 214). 22Explicitly ideological positions are often characterized by a rejection of "dominant meanings" (Freeden 2004, 9) that are reflected in the political status quo but are no less ideological.Groups that share the overall ideology of the status quo need not specify it, so their ideology remains implicit.It is thus not ideology per se, but whether the rebels want to change the basic ideology of the state that makes a group more or less extreme. 23Thus, it is reference to the status quo that determines extremism: the ideology of liberal democrats is not extreme in Western democracies, but it is in Iran or Saudi Arabia where secular, feminist, and gay rights activists are extremist given the political status quo. 24The greater the ideological difference from the status quo, the more extreme the group.
Ideologies are not just the sum of preferences over different policy outcomes; they entail a coherent set of ideas and beliefs that bundle things together as "package deals," addressing multiple dimensions of political and social life. 25Bundled ideas are not a priori extreme, but if a group's ideology differs from the status quo, bundling suggests that change is desired for a larger set of issues, making it more extreme in scope.
Much recent discussion of ideological extremism and terrorism focuses on religious ideology (especially Islam).We argue that religion per se does not make for extremism; a group's religious ideology may align with that of the state.If a group seeks to implement a religious way of life and politics in a secular polity (or to change from a dominant religious ideology to a different one), this is again a matter of the scope of change from the status quo. 26ome argue that religious ideologies are different, and more extreme, because of the role played by a transcendental audience, of "answering to God" or of rewards in the afterlife (Juergensmeyer 2017).This may change the cost-benefit ratio of fighting, but these aspects of religiosity pertain to commitment to the cause, rather than extremism as we conceive it (Berman 2009; Walter 2017).Fundamentalist religious ideologies (whether Muslim, Christian, Jewish, Buddhist, or any other tradition) can also be considered extreme if their strict belief in the literal interpretation of religious texts and traditions differs significantly from the political status quo.However, those who believe in a literal interpretation but are not organized politically to contest the secular state are not extremist as we define it.It is politicized religion rather than fundamentalist religion that matters.
Ideologies that are extreme relative to the status quo may (or may not-that is the empirical question) be related to extreme tactics such as terrorism. 27This leads us to a final important element of ideology: the connection between ends and means.Ideologies often connect values or goals with a "program of action" for achieving them (Gutiérrez Sanín and Wood 2014, 215). 28A view of ideology that sees ends and means as going together points to an empirical relationship between them. 29But whether this yields a connection between extremism and terrorism is not clear.Some scholars argue that ideology legitimizes targets that would otherwise be off-limits (Drake 1998;  Ron 2001).However, by proscribing certain behaviors, ideology may restrain violence against civilians (Thaler  2012).For example, Gandhi's goals, which were extreme by our definition (and seen as such by the British), were connected to an explicitly nonviolent program of action.

Extremism in Civil Wars
Using our conception of extremism as the scale and scope of desired change from the status quo, what distinguishes more moderate from more extremist rebel organizations?All rebel groups seek to change the status quo, but some desire greater change than others.
The Uppsala Conflict Data Program's (UCDP) Armed Conflict Dataset distinguishes between wars fought for control of the government and those fought over the status of territory (Themnér and Wallensteen 2014).Territorial and governmental conflicts occur over very different policy spaces, and it is difficult to make relative assessments of extremism across these categories.However, within each category, rebel groups have goals that range from moderate to extreme.
Among rebels fighting for control of the state, the more that groups seek to transform society and government in some fundamental way, the more extreme their goals.The extremist category thus includes groups that aim to transform a capitalist society into a communist or socialist one (e.g., the FARC in Colombia), or vice versa (e.g., the Contras in Nicaragua), as well as groups that aim to transform a secular society into one governed by religious law (e.g., al-Gamaa al-Islamiyya in Egypt).The opposite -groups aiming to establish a secular state in a religious state-would also be considered extremist under this conception. 30Moderates include rebels fighting to obtain the reins of power for themselves but not to transform the basic political, economic, or religious structure of society (e.g., the Cocoyes or Cobras rebel groups in the Congo, who fought to oust a leader but had no broader societal agenda, or, more recently, the conflict between the military and the Rapid Support Forces in Sudan).
Political systems are also transformed by drastic changes to the internal balance of power.We count as extremist groups those that aim to upend and redraw political participation by ending or replacing a monopoly of power in the hands of one group (e.g., the ANC's goal of ending Apartheid or the Hutu rebels' goal of deposing the Tutsi in Burundi). 31Minority groups fighting for greater rights within a given political system, but not to change the system altogether, have more moderate goals (e.g., the UTO in Tajikistan, a coalition pushing for greater political involvement for marginalized groups).Identity-based groups are not necessarily extreme-only those that aim to change the fundamental system of political rights are.
Among groups fighting over territorial status, some aim only for autonomy within existing borders of the state (e.g., the United Wa State Army in Myanmar/Burma), whereas others aim to change international borders. 32Most commonly, this entails seceding to form an independent state (e.g., the PKK in Turkey); less commonly, this goal involves joining another existing state or irredentism (e.g., the WSLF in the Ogaden region of Ethiopia that desired to join Somalia).Secessionist or irredentist groups are more extreme than those fighting for autonomy within the existing state.Figure 2 illustrates how we conceive of extremism across the two conflict types.

Theorizing a Connection between Extremism and Terrorism
As the scale or scope of distance between a group's goals and the political status quo increases, why might a tactic like terrorism, as opposed to other forms of violence, become more attractive?Once we move away from tautology, the answer is not immediately obvious.In this section, we draw on the terrorism literature to explore mechanisms that might make groups with more extreme goals more likely to use terrorism: provocation, the difficulty of achieving greater change, negotiation dynamics, and legitimacy costs associated with terrorism. 33We thus Provocation.Lake ( 2002) links extremism to terrorism through the mechanism of provocation.When the preferences of the population are fairly normally distributed (as in figure 1a), then extremist groups are unpopular and may use terrorism to provoke a government crackdown that redounds to their benefit by increasing support for them (Kydd and Walter 2006, 69-70).This is a way of trying to shift preferences closer to those of extremist rebels by polarizing the population (making the distribution look less like figure 1a and more like figure 1b).The benefits to extremists of provocation are relatively high in such scenarios.
However, arguments about terrorism used strategically for provocation assume that terrorism is more provocative -that is, more likely to elicit an overreaction-than other types of attack.Yet, several sources suggest this might not be the case.Carter (2016) finds that guerilla attacks were more likely to elicit a forceful state response than terrorist attacks.For the receiving public, there may not be a strong differentiation between attacks on civilians as opposed to military targets (Huff and Kertzer 2018).In Sri Lanka, the attacks that provoked the largest overreaction by the state and the majority population were attacks on soldiers, not terrorist attacks (Fortna n.d.).It is thus unclear whether extremism should make terrorism more likely, as opposed to violence more generally.Several other mechanisms more clearly connect extremism to deliberately indiscriminate terrorism.
Difficulty of Achieving Change.Extreme goals are more difficult to obtain and thus likely take longer to achieve.Groups aiming to move society further from the status quo may thus be more willing to "play the long game."Fortna  (2015) shows that terrorism makes achieving ultimate political goals less likely but lengthens the life span of rebel groups.This creates a dilemma for rebel organizations: terrorism is good for organizational survival but comes at the expense of the goals for which they fight.For groups expecting a long struggle, organizational survival may be relatively more important, making terrorism a more attractive strategy.
Negotiation Dynamics.If a compromise solution through a negotiated settlement with the government is possible, moderates will be more likely to prefer a settlement, whereas extremists will be more likely to value continued fighting.This follows logically from our conception of extremism as distance from the status quo along a continuum of preferences.For a given level of commitment, any possible compromise solution acceptable to extremists will also be acceptable to moderates, but the reverse is not true. 34The "win-set" of acceptable compromises is smaller between the government and extremists than between the government and moderates.This is related to the mechanism of playing the long game described earlier, but it can also lead to a spoiling dynamic.The organizational raison d'etre of extremist groups (or factions within groups) may be threatened by a negotiated settlement between the government and more moderate groups.Extremists thus have a strong incentive to scuttle talks, and terrorism is a good way to do so (Kydd  and Walter 2002; but see also Pearlman 2009).
Relatedly, extremist groups may use terrorism to signal their preferences for bargaining purposes.To be credible, signals must be costly (Fearon 1997).Terrorism is not costly in material terms-it is a relatively cheap way to inflict costs on the other side.However, it is costly in two ways that may help screen and therefore credibly reveal preferences.First, because it makes concessions from the government and negotiated settlement less likely (Fortna  2015; Kydd and Walter 2002), terrorism signals that rebels are not interested in compromises that more moderate groups might accept (Abrahms 2006).This makes it less likely that a bargain will be achieved, but if one is achieved, it is more likely to be more favorable to the rebels.
Second, because terrorism entails significant legitimacy costs, it is likely to drive potential supporters away.If terrorism drives away moderate supporters more than extremist supporters-that is, those whose preferences lie between the group and the government versus those whose preferences are even further out)-then terrorism shifts the distribution of preferences of the group's supporters.This could enhance the group's bargaining position by entrenching it.This strategy entails enduring the cost of shrinking the pool from which members and support can be recruited, while ensuring that members and supporters have more extreme preferences.It can thus serve as a reliable signal of preferences. 35egitimacy Costs.Targeting civilians in a deliberately indiscriminate manner is beyond the pale normatively, so it always risks undermining the legitimacy of a group and costing it support.But several factors may make the legitimacy costs of terrorism relatively lower for extremists than for more moderate groups. 36roups coded as extremist because they aim to implement a religious political system may face reduced legitimacy costs to terrorism relative to secular movements.Religion is a legitimating device, so terrorism may be less costly for groups that can frame it as necessary to protect divinely ordained political tenets.This may be especially true for religious ideologies that demonize nonbelievers (Juergensmeyer 2017), reducing the legitimacy costs of violence against them.Similarly, extremist nationalist ideologies seeking to alter power dynamics among identity-based groups and exclude other groups from power dehumanize the "other" in ways that justify (in their adherents' eyes) violence against civilians. 37overnments likely feel more threatened by groups with more extreme aims and may target them, and the populations for whom they claim to fight, more severely than they do moderate groups (Staniland 2021).This higher level of violence by the government, particularly when abuse is directed against civilians, can be used by extremist groups to legitimize responding by targeting civilians in kind. 38ote that when the legitimacy costs of terrorism are lower, using terrorism is relatively less useful as a signaling device to convey preferences.These dynamics are not contradictory, however.Terrorism creates higher legitimacy costs for all groups that use it than for groups that do not, but relatively less so when strong "othering" is at play.
In conflicts over territory, the logic of international, as opposed to domestic, legitimacy costs cuts the other way, however.All rebel groups would presumably prefer international support, whether material or diplomatic, but such support is crucial for secessionist groups.To achieve their aim of an independent state, secessionists require international acceptance-after all, the difference between de facto and de jure sovereignty is international recognition.Secessionists must therefore concern themselves with international legitimacy more than other rebels, making them more likely to exercise restraint and avoid tactics beyond the normative pale, including terrorism (Fazal  2017; 2018b).This creates a countervailing pressure, with the negative effect of relative international legitimacy costs for secessionists countering the positive effect of relative domestic legitimacy costs for extremists.It suggests that extremism will be more closely tied to terrorism in conflicts over government control than in those over territory.Thus, the effect of extremism on terrorism may be conditional on conflict type. 39potheses Although the available data preclude empirical adjudication of these potential mechanisms here, the preceding discussion suggests two testable hypotheses about the empirical relationship between extremism of a group's goals and the likelihood that it uses terrorism.We begin by testing the conventional wisdom of a blanket effect: H1: Rebel groups with extremist aims are more likely to employ terrorism than are those with more moderate goals.
Given our discussion of the countervailing effects of domestic and international legitimacy costs for secessionists, however, we hypothesize that the effects of extremism may be different for those fighting over territory (where extremism entails secessionism) and those fighting over government control (where extremism entails a fundamental change in the ideology of the political system or a fundamental reshaping of the political power of identitybased groups).Our second hypothesis thus tests a conditional relationship: H2: The relationship between extremism and terrorism will be less pronounced in territorial conflicts than in conflicts over government control.

Universe of Cases and Selection Issues
We evaluate these hypotheses in intrastate armed conflict between 1970 and 2013. 40We use the TAC dataset (Fortna, Lotito, and Rubin 2022) to measure terrorist incidents attributable to rebel groups in the UCDP dyadic data on intrastate wars (Themnér and Wallensteen 2014).
The TAC data, which draw on the Global Terrorism Database (GTD; LaFree and Dugan 2007), cover 409 rebel groups in 166 conflicts in 96 countries.Our main unit of observation is the group-year (or dyad-year).
Studying extremism and terrorism in civil wars has both pros and cons.This universe of cases does not cover the full range of extremism or political tactics.If we included the full range of political organizations, such as unions, parties, policy advocacy groups, and so on, we would observe a much longer spectrum from moderate to extreme. 41Only groups with serious grievances with the status quo are likely to take up arms.Similarly, groups unwilling or unable to use violence and to take on the state militarily are excluded.Rather than consider the relationship between extremism and violence or tactics more broadly, we address a more specific question: Among groups with serious grievances that have decided to employ violent means-that is, to wage civil war-why do some turn to a particular form of violence, deliberately indiscriminate terrorism?Many arguments about the causes of terrorism conflate it with causes of violence or even conflict more broadly.We aim to isolate the effects of extremism on deliberately indiscriminate terrorism as the most puzzling form of violence.It is almost universally condemned normatively and has been found empirically to be ineffective (Abrahms 2006; Fortna 2015).Perhaps it is unsurprising that extremism is thought to explain terrorism.
There are also practical reasons to restrict our study to civil wars and the UCDP Armed Conflict Dataset (on which the TAC dataset is based).Other datasets that cover both violent and nonviolent movements are restricted in terms of group aims.To answer the broader question of the relationship between extremism and violence would require systematic data on the universe of political organizations, data that, to our knowledge, do not exist. 42e focus here on extremism and terrorism by rebel organizations rather than states for both theoretical reasons and practical reasons of data availability.On theoretical grounds, because we conceive of extremism relative to the status quo, and governments by definition defend the current status quo in a civil war, they cannot be extremist in the same way as rebels.Governments are not unitary actors and members or groups within governments can wish to change the status quo.Governments may also come to power intending to change the existing status quo and may try to alter prevailing political norms. 43In this case they are extremist by our conception, but the measurement of the relevant status quo needs to be amended.Notably, they attempt to effect change by using the levers of state power, not by rebellion.The mechanisms that would lead them to use deliberately indiscriminate violence are thus likely different than for rebel organizations.This is not to say that states do not commit indiscriminate violence against civilians; empirically, they are responsible for significantly more violence against civilians than are rebel organizations.But whether governments that attempt to change from some prior status quo are more likely than governments defending an entrenched status quo to engage in such violence is a question we leave to further research.Systematic data on deliberately indiscriminate terrorism are available only for rebel organizations, not for governments.
Dependent Variable: Indiscriminate Terrorism Our dependent variable is the use of deliberately indiscriminate terrorism by the rebel groups.The TAC data provide a flexible and systematic way to match groups from the GTD to the UCDP, accounting for the ambiguity in attributing many terrorist incidents to specific rebel groups.For our primary analyses, we use TAC's version A-the "more restrictive" method of filtering attack and target types as a proxy for deliberately indiscriminate attacks on civilians-and the total fatalities from terrorist attacks in a year (Fortna, Lotito, and Rubin  2022).We focus on fatalities as opposed to terrorist incidents because the mechanisms, including provocation, spoiling, and legitimacy costs, are most likely to be driven by deadly attacks and the overall number of deaths, rather than nonfatal attacks (the majority of incidents in the GTD kill no one). 44dependent Variable: Extremism Our main explanatory variables are measures of extremism, defined as the scale and scope of distance between the rebels' desired goals and the political status quo.The Rebel Extremism Dataset (RED) (Joyce & Fortna 2024), created for this project, draws on the UCDP's Conflict Encyclopedia and Cunningham, Gleditsch, and Saleyhan's (2009;  2013) Non-State Actor (NSA) dataset coding notes, supplemented with primary and secondary sources (including group manifestos where available), to code each group's stated goals at the conflict outset.The status quo is defined as the political situation at the conflict outset, no matter how recently it might have changed.
Coding goals at the beginning of the conflict helps control for potential endogeneity issues: rebels' demands reflect strategic responses to their environment and may grow with battlefield success or shrink with setbacks or negotiations with the government. 45The use of terrorism may also affect bargaining dynamics, making governments more or less likely to offer concessions, and thus affecting what rebels think they can achieve at the negotiating table.Measuring aims once fighting is underway is thus fraught with complications.It is also important to avoid inferring goals from actions or statements that occur after the conflict.Former armed groups, for example, may become participants in the political process and their platforms will change.The primary limitation of this approach is the static measure of extremism; our data are not well equipped to capture subtle changes in aims. 46e rely on stated goals as a proxy for group preferences, which are not directly observable.Neither we, nor their opponents, can ever know rebels' true aims with perfect accuracy.Stated goals may not reflect true aims, given that rebels are strategic actors who may have incentives to misrepresent.Rebels may present their goals as either more moderate or more extreme.They may, in classic bargaining strategy, exaggerate demands knowing that a settlement will entail a compromise from their initial positions.Alternatively, they may present their position as more moderate, hoping to gain modest concessions and then to demand more.Governments often seem to fear this latter tendency-that there is a slippery slope, such that if more reasonable concessions are granted, demands will only increase (e.g., if autonomy is granted, secessionism will follow). 47f rebels who exaggerate the extremity of their position as a bargaining ploy are also inclined to use terrorism to demonstrate their commitment and resolve, or if those who underrepresent their true desire for change to appear as reasonable bargaining partners are less likely to use tactics seen as beyond the pale such as terrorism, our analysis might suggest a stronger relationship between stated goals and terrorism than exists between true goals and terrorism if the former could be observed.But this dynamic depends on there being a true relationship between stated goals and terrorism-the question we assess here.Moreover, the alternative to using stated goals, inferring preferences from behavior, leads to the tautology we seek to avoid.
Although we conceive of extremism as a continuum, we unfortunately lack data that are fine-grained enough to code goals along a continuum and so must rely on dichotomous proxy variables for our empirical analysis.We create three indicators: two for conflicts over governmental control and one for conflicts over territory (as distinguished by UCDP).Among the former, binary variables indicate, respectively, whether the rebel group's goals include changes to the basic political ideology of the state (e.g., from capitalism/market economy to communism/collective ownership, or vice versa, or from secularism to religious law), or major identity status changes that would transform the political system. 48We do not treat identity and ideological extremism as mutually exclusive: a group can be either, both, or neither, depending on its stated goals.Among territorial conflicts, a binary variable indicates whether a group seeks to change international borders. 49Our distinction between secessionist and autonomy-seeking groups aligns with others' coding of maximalist or radical goals (e.g., Chenoweth and  Stephan 2011; Staniland 2021; Thomas, Reed, and Wolford 2016); however, our coding of whether a group aims to transform the political system is somewhat different, as detailed in the online appendix.To test H1, we create a binary variable that takes a value of 1 if any of the three indicators of extremism takes a value of 1 and use data that pools the two types of conflict.
Although comparing across sets of cases with different measures means we do not have a perfect test of H2, splitting the population by type of conflict provides some indication of whether, as H2 would predict, the relationship between extremism and terrorism is more muted in territorial conflicts.Accordingly, to test H2, we evaluate separately the relationship between extremism and terrorism in conflicts over government and territory.

Controls
To avoid spurious results, we control for factors likely to drive both extremism and terrorism.Government regime type is a well-known predictor of terrorism (Chenoweth  2013).Because democracies allow for accommodation of a wider set of preferences and compromise among differing groups, we may be more likely to observe civil wars where the regime's opponents hold extreme views.More moderate groups are thus more likely to be selected out of the civil war data in democracies than autocracies.
We also control for population size, a common control in the terrorism literature that could be connected to extremism if bigger countries are more likely to have a sizable enough group whose preferences diverge from the status quo to mount an insurgency.The population in the tails of the preference distribution might be larger if the overall population is larger.
Terrorism is commonly thought to be a "weapon of the weak."Although empirical evidence for this conventional wisdom is lacking (Fortna 2022), we control for relative military strength because strength may shape the goals for which rebels fight (Buhaug 2006; Werner 1999), as well as popular support, which may be related both to extremism and terrorism (again in a weapon-of-the-weak logic). 50e control for multiple rebel groups in a conflict because some argue that competition among groups drives rebels both to articulate more extreme goals (Kydd and  Walter 2002; Walter 2017) and to engage in terrorism in a bid for popular support and recruits (Bloom 2005; Kydd   and Walter 2006).Although recent studies have cast doubt on an empirical relationship between competition and terrorism through "outbidding" (Findley and Young 2012a), we include a dummy variable denoting conflicts with multiple rebel groups.
We control for conflict intensity by distinguishing years of major conflict (1,000 or more battle deaths) from years of minor conflict (25-999 battle deaths).Levels of terrorist violence may be higher in more intense conflicts simply because all forms of violence are more prevalent and the conflict may be more intensely fought when the parties' positions are further apart.This relationship may be reciprocal, with more intense fighting polarizing positions and making rebels even more extreme.Even if intensity is not causally prior to extremism, controlling for it helps separate the direct effects of extremism on terrorism from those that run through intensity.Doing so may bias away from finding an effect of extremism but enables us to isolate the effects of extremism on terrorism from effects on violence more generally.
We include a control for ethnic conflict.All societies engulfed in civil war are polarized, but we might expect that conflicts that pit identity groups against each other to be particularly so.The legitimacy costs of terrorism may be lower in deeply polarized conflicts.Extremist groups may be more likely to emerge under these conditions, and identity conflicts, with strong "us versus them" dynamics and a tendency to dehumanize the other side, may also be prone to terrorism.Note, however, that this measure differs from our measure of extremism based on a stated aim to change the political status of identity groups in a way that transforms the political system.Not all ethnic conflicts are extreme.
How a rebel group finances its fight may be related to terrorism (Fortna, Lotito, and Rubin 2018; Salehyan,  Siroky, and Wood 2014).It might also shape extremism if outside sources of financing allow groups to emerge that are more extreme than would otherwise be the case, for example, because they lack support among the domestic population (Lake 2002).Aims may also be endogenous to resources-the more resources a group has, the more ambitious its aims can be (Buhaug 2006).We include a dummy variable that marks whether a rebel group had external support or financed its fight through lootable resources such as gems, drugs, and so on. 51e also include several controls for the time period.The ideological struggle of the Cold War shaped the aims of rebel groups fighting during this period, both through ideological diffusion and because there were clear incentives for rebels to articulate strong Marxist or capitalist/ democratic aims to gain superpower patronage.Some scholars have noted a shift in the tendency for groups to use terrorism after the Cold War (see Chenoweth 2010;  Enders and Sandler 1999). 52everal of our control variables are plausibly endogenous to or have a reciprocal relationship with extremism.Including them may thus entail controlling for the effect of our independent variable in a way that mutes results.This is most clearly possible for popular support, if extremist groups have a harder time recruiting supporters.Through popular support, this may also be true of relative strength (of which support is a component) and rebel financing (if those with less popular support turn to external or lootable resource financing or, conversely, if the ability to find external financing is affected by extremism).There is also likely a reciprocal relationship between rebel group extremism and conflict intensity.We therefore test the robustness of our results when these plausibly endogenous control variables are lagged by one year.

Model Selection
Because our dependent variable, terrorism fatalities, is a count variable characterized by overdispersion and zeroinflation (i.e., many more observed zeros than expected by chance), and because the mechanisms we propose might have effects on the decision to employ any indiscriminate terrorism, on the overall number killed, or both, we use a zero-inflated negative binomial (ZINB) model.A ZINB regression model consists of two parts: a logistic regression (inflate model) that calculates a point mass at zero, and a negative binomial log-linear regression (count model) that calculates the number of nonzero events. 53hese two parts have different interpretations.In the inflate model, a positive coefficient indicates a positive association with zeros (i.e., a negative relationship with the use of any fatal terrorism), whereas a positive coefficient in the count model indicates a positive association with higher amounts of terrorism (more fatalities in a year).Because we are interested in the effects of extremism on the decision to use terrorism at all and on the amount of terrorism if it is used, it makes theoretical sense to use a ZINB model.All models include robust standard errors clustered by conflict; the results are robust to clustering by country or by dyad.

Results
We begin by examining all conflicts together; that is, pooling conflicts over territory and those over governmental control.Table 1 shows the relationship between terrorism and extremism across all conflict types, testing H1.To facilitate the interpretation of effect sizes, figure 3 provides marginal effects of a one-unit increase in the respective predictor on the predicted number of terrorism fatalities (count model) and the probability of zero fatalities (inflate model).Because the key independent variables are dummy variables, a one-unit increase is the difference between extreme and moderate cases.In the pooled results, we find that groups with extreme aims are more likely to employ terrorism (as indicated by the significant and negative coefficient in the inflate model), but not to kill more people in terrorist attacks (here the coefficient sign suggests the opposite, though it is not significant), offering mixed support for H1.This suggests that the conventional wisdom is too simple.
To examine whether the effect of terrorism differs by type of extremism (H2), we split the population, looking separately at conflicts over government control (table 2 and figure 4) and conflicts over territory (table 3 and  figure 5).Among the former, we also distinguish between two indicators of extremism: groups that aim to alter the basic political ideology of the state and groups that fight to change the status of identity groups in fundamental ways.
As table 2 indicates, both ideological and identity goals are associated with terrorism in conflicts over government; however, the relationship runs through different channels for the two measures of extremism.Groups that aim to change the ideology of the state are more likely to employ deliberately indiscriminate terrorism.However, the association operates primarily through the decision to use terrorism (as shown by the significant negative coefficient in the inflate model), rather than through the number of fatalities (as seen in the count model, where the coefficient is in the predicted direction but is not significant).On average, groups with ideological goals are nearly 22% more likely to use terrorism.Terrorism is also associated with groups seeking major identity status change, relative to groups with moderate or no identity-based goals.Here, the relationship is apparent in both the number of fatalities (count) and the decision to use terrorism (inflate), but the former is much more robust than the latter (see the later discussion).This difference potentially reflects "othering" dynamics in which-relative to ideological differencesidentity-based conflicts lower the legitimacy costs of killing large numbers of "enemy" civilians by dehumanizing the other side.The positive relationship between extremism and terrorism disappears in conflicts over territory.Table 3 shows no association between secessionist goals and terrorism.Figure 5 presents the marginal effects, which are small and close to zero for secessionism, relative to autonomy claims.This finding supports expectations in the literature that secessionist groups, motivated in part by their quest for international legitimacy, are more likely to exercise restraint in their choice of tactics (Fazal 2017; 2018b).It counters arguments that secessionist groups will use more terrorism (Pape 2005), however.
Together, these findings provide support for H2.We see a significant relationship between extremism and terrorism in conflicts over government control but not in conflicts over territory.Our test of H2 is imperfect and there is room for further research on this point, but it tentatively supports the notion that domestic and international legitimacy costs create countervailing tendencies for extremist groups that hope to gain international recognition for an independent state.

Robustness Checks and Additional Tests
We examine the robustness of these results using several tests.First, we examine alternative measures of terrorism, using the yearly number of terrorism incidents (rather than yearly fatalities) and less restrictive measures of terrorism (capturing a broader set of attack and target types) from the TAC data.Second, we incorporate additional covariates, including a lagged measure of terrorism fatalities and a measure of conflict duration.Third, we drop the measure of ethnic conflict because controlling for it may make results dependent on the few, arguably idiosyncratic, cases of territorial conflict not fought along ethnic lines.Fourth, we lag potentially endogenous control variables.Because doing so introduces missing data, as explained in the online appendix, we conduct this robustness test using multiple imputation.Fifth, we drop the few cases where we code a substantial change in goals after the war began to ensure that our decision to use a static, prewar measure of goals (to avoid endogeneity concerns) is not skewing our results.Sixth, we use negative binomial regression as an alternative model specification.Finally, we use alternative measures of ideology and independence (as a proxy for secession) from Braithwaite and Cunningham (2020).
The results, reported in the online appendix, are largely robust to these alternative specifications with a few exceptions.As noted earlier, the effect of groups aiming to transform the political status of an identity group on the decision to use terrorism (inflate models) is not particularly robust.It loses significance when we use terrorism attacks (including many in which no one was killed) instead of fatalities or a less restrictive measure of deliberately indiscriminate terrorism as the dependent variable, when we drop the control for ethnic conflict, or when we lag potentially endogenous control variables.Effects of identity status extremism on the amount of terrorism (count models) are generally robust, 54 as are results (both inflate and count) ideological extremism and the null results for secessionism.Including a lagged measure of the dependent variable affects some results (see the online appendix), although it is not clear that this variable has systematic causal value, and it may be artificially deflating the coefficients of primary independent variables.Finally, the online appendix reports an empirical exploration of an alternative conception of extremism discussed earlier: the territorial extent of change.

Conclusion
Is there a nontautological relationship between extremism and terrorism in civil wars?The answer to this question is not an easy "yes" or "no."It depends on the type of conflict and, we hypothesize, whether there are countervailing legitimacy pressures.In conflicts over governmental control, more extreme goals are associated with terrorism, whereas there is no clear relationship in conflicts over territory.The conventional wisdom that extremism leads to terrorism is too simple.Once the tautology of equating terrorism with extremism is unpacked, a more nuanced relationship can be seen.
To investigate this question, we develop a conception of extremism that focuses on the distance between a group's stated political aims and the political status quo.We discuss several mechanisms that might link extremism of goals to the strategic use of deliberately indiscriminate terrorism.These include incentives to provoke a government backlash that will make otherwise unpopular extreme positions more popular (though this should only lead to terrorism if terrorism is more likely than other forms of violence to provoke an overreaction, which is an open empirical question); a higher value placed on organizational survival for groups that anticipate a long fight; incentives to spoil the possibility of peace between the government and more moderate groups; a need for credible signals of preferences; and lower domestic legitimacy costs to terrorism in polarized societies, among groups pursuing religious ideological aims, or among populations targeted more harshly by the government.We hypothesize that the relationship between extremism of goals and terrorism will be muted in the cases of secessionist rebels, however.Here, the mechanisms that link extremism to terrorism are offset by secessionist rebels' particular need for recognition from the international community, which makes the legitimacy costs of terrorism higher for them than for other extremist groups.
We test these hypotheses with new data on rebel extremism.In conflicts over governmental control, we find that extremist groups that seek to transform either the political ideology of the state or political power among identity groups are more likely to use terrorism, although these results are nuanced.Ideological goals are strongly associated with whether groups use terrorism but not with the number of fatalities from terrorism.Identity goals, in contrast, appear to be related to the number of people killed through terrorism, although effects on the decision to use it in the first place are not particularly robust.We find no such relationship between extremism and terrorism among conflicts over territory.This is consistent with our hypothesis that competing dynamics cancel each other out: extremism pushes toward terrorism, while the need to behave responsibly to attract international recognition of independence pushes against it (Fazal 2018a).We acknowledge, however, that extremism may have no effect in this set of conflicts.Determining whether there is no effect or countervailing effects that cancel each other out is an important question for future research.
Although our findings point to a relationship between extremism and terrorism in civil wars, our data do not yet allow us to adjudicate between the different mechanisms that might be at play. 55Future research should further explore these mechanisms.The alternative conceptions of extremism also invite further investigation.Preliminary tests (in the online appendix) suggest only mixed support for a relationship between terrorism and extremism conceived of as the geographic scope of aims.More research should also investigate the relationship between commitment to the cause and terrorism, although there are empirical challenges to measuring commitment a priori.
Further research might also explore whether different types of ideological extremism have different relationships to terrorism.Is distance from the status quo along the religious/secular dimension different from that along the dimension of rights for linguistic groups 56 or along the left/right dimension, for example?We deliberately conceive of distance from the status quo in either direction as extremism.This is important to avoid labeling only goals we dislike as extremist, as is often done in the literature.But the relationship to terrorism may not be symmetric.Ideologies are "packages" of ideas that often link ends with prescribed means.Future work might explore how the content of certain ideologies promotes or proscribes violent tactics such as terrorism. 57We separate these conceptually to avoid tautology, but they may be correlated empirically.Are groups pushing for liberal change in an illiberal state less likely to use terrorism?They appear to be less likely to use violence and hence are selected out of our data.This selection helps us isolate explanations of terrorism from explanations of violence more generally but makes our data inappropriate for investigating whether liberal and illiberal ideological extremism differ.Similarly, more could be done to examine the role that transnational movements might play in spreading tactical choices (such as whether to use terrorism) across conflicts.Ideologically aligned groups influence and sometimes even train each other (the PLO is thought to have provided training for the LTTE, for example).This may be less about the content of ideology than about network effects among rebel groups that are in some ways less territorially constrained than are national governments.
Finally, our research engages with a topic of importance to policy.Government programs to prevent or counter "violent extremism" as part of counterterrorism efforts tacitly posit a relationship between extremism and terrorism.But to know whether such a relationship exists, we need clear and distinct conceptions of these phenomena.This article offers a framework for defining extremism and draws on existing arguments in the civil war and terrorism literatures to explore, in a nontautological way, how extremism is related to terrorism in civil war.Our findings suggest that PCVE efforts (including programs to counter extremist ideology) may be more productively focused on conflicts over government, where the association between extremism and terrorism is strongest.Our findings further suggest that the international community may have more sway over the use of terrorism with secessionist groups, which need recognition if they are to achieve sovereignty, than with other rebel groups.The role of domestic legitimacy costs in shaping the pressures to use terrorism suggests that domestic constituencies and the government also have some leverage in influencing rebels' choice of tactics.Once the tautology between extremism and terrorism is undone, it becomes possible to examine the relationship theoretically and empirically.Doing so reveals that the relationship is not as simple as the conventional wisdom suggests.

Supplementary Material
To  5 Crenshaw (2011, 69-70) suggests that all terrorist groups have extreme goals but not all groups with radical aims use terrorism.6 Pearlman (2009) argues it is not only policy preferences but also internal politics within a movement that drive spoiling.7 Piazza (2009) examines lethality per attack and compares this indicator only among terrorist groups, but his theory seemingly applies to the decision of whether or how much to use terrorism.See also Masters  (2008).
8 Berman (2009, 17) gives examples of Mennonites, Hasidim, and others.We hear most about the radical religious groups that use violence because violence generates attention.9 We include "categorical terrorism" in which attacks target identity groups but not particular individuals within them.For a fuller discussion, see Fortna (n.d.). 10 Kydd and Walter (2006, 66) refer to this as "intimidation."On civilian cooperation and noncooperation with rebels, see Arjona (2017).11 We are agnostic as to whether rebel groups form first and then choose goals strategically (e.g., Walter 2017) or whether goals emerge organically from the grievances of a societal group (e.g., Bormann, Cederman,  and Vogt 2017; Petersen 2001).12 These dimensions are not mutually exclusive.13 The idea that unpopular views motivate terrorism underlies some arguments about terrorism as a "weapon of the weak" (e.g., Crenshaw 1981, 384).For an empirical critique, see Fortna (2022).14 Lake (2002, 18) explicitly assumes that preferences are "randomly distributed" within society.15 See also Vogt (2019).Chenoweth and Stephan  (2011) and Thomas, Reed, and Wolford (2016)  distinguish maximalist goals to fundamentally alter the political order from more limited goals.Schmid (2014) considers extremism relative to the political status quo in Western liberal democracies but not how that baseline shifts elsewhere.Staniland (2021) also considers opposition groups "radical" if their goals differ more strongly from that of the state.16 If the aggrieved population is a small minority, the bump on this side of the distribution will be relatively small, but it could represent a majority of the country's population if an ethnic majority is politically oppressed (as in Apartheid South Africa).17 Popular support is not only difficult to measure but is also likely endogenous to many factors of the conflict, including group goals, tactics, strength, the government's tactics, and so on.2019).25 Ideologies often come in a "distinct and pre-structured form, such as liberalism, conservatism, socialism, feminism, or fascism" (Freeden 2004, 7).26 That Islam is considered particularly extreme or threatening reflects the status quo bias of the predominantly Judeo-Christian West.27 Crenshaw (2011, 73, 98) notes that explanations of terrorism often focus on ideology, but many terrorist groups' commitment to ideology is weak or inconsistent.28 Tokdemir et al. (2021) distinguish ideologies from demands.They include tactics as part of the former and note that groups can differentiate themselves from other groups both in terms of their ideology and their demands.Wood and Thomas (2017) show that a group's ideology affects who fights, specifically the participation of women in combat roles.29 On the relationship between ideology and mass killing, see Straus (2015) and Leader Maynard (2022).30 There are no examples of this in our data.31 White nationalist groups, such as the Atomwaffen Division (AWD) in the United States that aim to (re) create a monopoly of power, are similarly extremist.32 Autonomy aims are rare among territorial conflicts in the UCDP: more than 80% of rebel groups in such conflicts have stated an intention to redraw borders, whereas 13% seek only autonomy (there is mixed or indirect evidence of secessionist or irredentist intent in the remainder).33 These mechanisms could, in theory, operate under other conceptions of extremism as well.
34 This is theoretically different from, but may be observationally equivalent to, a conception of extremism as commitment to the cause: holding preferences constant, more resolved groups are more willing to pay the continued costs of fighting.35 A parallel mechanism may link terrorism to resolve if only those who are highly committed continue to support an organization that uses terrorism (Berman  2009; Piazza 2009).36 In highly polarized societies, terrorism may also be less costly for opposition groups, no matter their aims, than in less polarized societies.Potential supporters of a rebel group may be more likely to believe that the "ends justify the means."There may also be fewer "fence-sitters" (see figure 1) likely to be alienated by terrorism.Finally, a polarized population is more likely to mobilize around the rebels' cause in response to a government backlash than to shift toward a government from which it is already alienated.37 Stanton (2013, 1013) argues that rebels who seek gain for their own identity group over others are more likely to use high-casualty terrorism because they need not worry about backlash among the wider population.However, groups seeking autonomy are also primarily concerned with support from their "own" identity group.38 We thank an anonymous reviewer for suggesting this mechanism.39 This logic runs counter to Pape's (2005) suggestions that suicide terrorism is motivated by the goal of national liberation.40 Data are missing for all groups in 1993 because the GTD is missing data for that year.Some of our control variables are not coded through 2013, thus limiting the number of cases in our main analyses.41 Although some extremist groups eschew violence (Vogt 2019), the vast majority of moderate political organizations operate through nonviolent and usually legal means.42 The NAVCO dataset includes only groups with "maximalist" (i.e., extremist) goals (Chenoweth and  Lewis 2013), whereas data on self-determination movements include a wider range on extremism but are restricted to conflicts over territory (Cunningham  2014; Sambanis, Germann, and Schädel 2018).43 Staniland's (2021) theoretical starting point and focus are on the ideological project of the state.He notes that the state's political project can change, but dramatic changes are quite rare.44 Note that the mechanisms we propose earlier could affect either a binary decision to use (fatal) terrorism at all, the overall number and lethality of attacks, or both.
The statistical model we use, described later, allows us to investigate this question.
45 Goals at conflict onset are endogenous to prior political processes and may affect whether a conflict crosses this threshold, creating a selection effect.On strategic demands, see Buhaug (2006); Thomas, Reed, and  Wolford (2016); and Werner (1999).46 RED codes larger shifts in extremism.Such shifts are relatively rare, however, occurring in just 35 cases.See the online appendix.47 Anecdotal evidence suggests that rebels' stated goals align with their true goals or that they have difficulty backing away from their stated goals.For example, the LTTE in Sri Lanka maintained and fought for its stated goal of independence, even though this stance cost it international support and led to military defeat.48 The online appendix includes descriptions of variables, coding procedures, and data sources for controls.We do not code as extremist those groups aiming to change the methods of selection of executive leadership unless accompanied by other extreme goals.
Although we use binary variables for the analyses presented here, RED codes 14 variables on group aims (including changes to leadership selection), providing scholars with options for different analyses.49 If there is ambiguous evidence of whether a group seeks to change borders, we code a group as secessionist only if we find additional evidence of an intention to create an independent state.An example of an ambiguous case without a clear declaration of independence is the Popular Movement of Azawad that fought for a Taureg homeland in Mali.50 Rebel strength and popular support may also be related to the credibility of stated aims (Thomas, Reed,  and Wolford 2016).51 We have data on rebel financing only through 2006, limiting our main analyses to 1970-2006 (see the online appendix).52 Other period controls account for the GTD's modification of data collection procedures as the project was transferred among institutions in 1998 and again in 2008.Coefficients are not reported here because of a lack of space.53 ZINB models assume zeroes result from two separate zero-generating processes: one "structural" and one "sampling."In our context, structural zeros come from groups that do not (or cannot) use terrorism.Conversely, "sampling" zeros reflect groups that might have used terrorism but did not do so in a given year.We relax this assumption in robustness tests using a negative binomial regression model.54 However, effects drop to only marginal significance when we lag potentially endogenous controls.55 Future research might explore whether mechanisms that run through others' perception of extremism (e.g., if governments target groups that they perceive as extremist more violently than other groups) are more or less supported than mechanisms that run through a group's own assessment of the extent of its demands (e.g., a perceived need to play the long game).We thank an anonymous reviewer for raising this point.56 Borman, Cederman, and Vogt (2017) argue that linguistic cleavages are no more prone to violence than religious cleavages, but whether they are more prone to terrorism is an open question.57 Some ideologies (e.g., Fascism and some strands of Islamism) seem to valorize violence, but it is an open question whether the valorization of violence leads to or follows from a decision to employ it.Whether the content of some ideologies makes the particular form of violence that we study here, terrorism, more likely also remains unanswered.

Figure 2
Figure 2Conflict Types, Group Aims, and Extremism Figure 3 Marginal Effects for Extremism and Terrorism in Civil WarsExtreme aim

Table 2
Extremism and Terrorism in Conflicts over Government

Table 3
Extremism and Terrorism in Conflicts over Territory view supplementary material for this article, please visit http://doi.org/10.1017/S1537592724000598.Previous versions of this paper were presented at the American Political Science Association annual meeting, the Conference on Ideology and Political Violence at the Bush School of Government, Texas A&M University, the Political Violence Workshop at Washington University, St. Louis, and the Monday International Relations Thoughts Series Colloquium at the University of California, Berkeley.We thank Jessica Braithwaite, Libby Wood, Carly Wayne, Kathleen Cunningham, Julia Raven, other panel members and workshop participants, and the anonymous reviewers for very helpful feedback.Thanks also to Laura Resnick Samotin for research assistance on data collection, and to data science specialist Noah Greifer at the Institute for Quantitative Social Science, Harvard University, for statistical support.Any and all errors are our own.