1. Introduction
In June 2021, Israel's president Isaac Herzog publicly condemned economic boycott actions against Israel as a form of “economic terrorism”. This terminology, prompted by Ben & Jerry's decision to stop selling ice cream in Israeli West Bank settlements, is neither new nor unique. In Canada, local politicians blamed COVID-19 demonstrators blocking a central road for exercising economic terrorism (Morden, Reference Morden2022). Canada itself was accused of economic terrorism by a US congressman after imposing tariffs on American farmers (Dale, Reference Dale2018). The term “terrorism” is used in other unorthodox ways: Russia denounced foreign rumors as “information terrorism” (Walters, Reference Walters2022), the Missouri Attorney General was accused of practicing “litigation terror” (Kuang and Shorman, Reference Kuang and Shorman2021), and language laws in India (Roy, Reference Roy2022), criminal behavior by Muslim youth in the UK (McKinstry, Reference McKinstry2017), and confederate statue removal in the US (Larimer, Reference Larimer2016) were described as “cultural terrorism”.
What stands out about these examples is their reference to nonviolent actions. By invoking terrorism, this rhetoric deviates from the concept's colloquial and formal definitions, which assume actual or threatened violence (Chenoweth, Reference Chenoweth2013; Schmid and Jongman, Reference Schmid and Jongman2017). Yet do ordinary citizens accept nonviolence as akin to terrorism? This question is highly important: perceiving nonviolent acts as terrorism can justify aggressive state repression and reduce the menu of legitimate tactics in conflictual settings. Recent research on popular perceptions of out-group contention considers different types of violent or physically threatening actions (Huff and Kertzer, Reference Huff and Kertzer2018; Edwards and Arnon, Reference Edwards and Arnon2021; Manekin and Mitts, Reference Manekin and Mitts2022; Norman, Reference Norman2022). Less attention is given to a broader range of nonviolent acts.
The paper explores this understudied question. We hypothesize that contrary to formal definitions, popular perceptions of terrorism do not strictly exclude nonviolent actions. Instead, they are part of a spectrum subject to similar logic as violent acts (Huff and Kertzer, Reference Huff and Kertzer2018): such perceptions of nonviolence should increase by the threat embedded in the action, elite labeling, and the target audience's ideological predispositions.
Using novel data from a preregistered survey-experiment conducted in Israel, we examine whether different nonviolent actions by the Palestinians—economic sanctions, legal petitions, and illegal construction—can be seen as forms of terrorism. To distinguish substantive classification of terrorism from symbolic terminology, we examine both general disapproval and actual willingness to employ anti-terrorism repression. We find that many Israeli Jews rank nonviolent Palestinian actions close to violent terrorism, deem them strongly illegitimate, and justify the use of security forces to stop them. These perceptions vary by ideology: right-wingers judge nonviolence most harshly, whereas left-wingers are more sensitive to the action's type. Automated textual analyses of open answers suggest that tangible threat and different partisan moralities are key mechanisms for these differences. Meanwhile, explicit terrorism labels primarily affect the way centrists, who hold swing positions on the conflict, perceive nonviolent resistance. Hence, such rhetoric seems especially effective in swaying swing audiences toward hawkishness. These heterogeneous patterns replicate with a convenience sample of Americans asked about the lower-salience US–Iran conflict, indicating generalizability to other cases. Nevertheless, Americans' judgments of nonviolence are less severe in absolute terms, suggesting conflict salience matters.
Our findings make several contributions to the growing debate about public perceptions of terrorism and outgroup behavior (e.g., Huff and Kertzer Reference Huff and Kertzer2018; Edwards and Arnon Reference Edwards and Arnon2021; Manekin and Mitts Reference Manekin and Mitts2022; Norman Reference Norman2022). First, by comparing violence and nonviolence, we expand past research to a fuller spectrum of adversary actions. Recent research has begun mapping different types of nonviolent resistance from the protestors' perspective (Cunningham et al., Reference Cunningham, Dahl and Frugé2017), but focuses less on target audiences' public perceptions. For the latter, we show, the distinction between violence and nonviolence is more fluid than assumed. In a salient conflict, nonviolence by a known adversary is often perceived as borderline terrorism and justifies use of force. This holds even for appeals to internationally acceptable fora and self-regarding acts with little direct harm. Thus, future research on terrorism and contentious politics should be mindful of the full repertoire of rival actions and their perceptions by target audiences.
Second, our analysis contributes to ongoing debates about the effects of top-down labels on public perceptions (D'Orazio and Salehyan, Reference D'Orazio and Salehyan2018; Baele et al., Reference Baele, Sterck, Slingeneyer and Lits2019; Dolliver and Kearns, Reference Dolliver and Kearns2022). Our findings provide new evidence that labeling nonviolence as terrorism can legitimize hawkish security policies, particularly in the eyes of centrist individuals. This carries real political implications: if nonviolent resistance can be labeled, perceived, and suppressed as terrorism, some adversaries may conclude that violence remains the only course of action. Our analysis suggests more work is needed in this vein.
Third, we illustrate the strong influence of ideological predispositions toward the adversary, an understudied aspect in many recent works about public perceptions of terrorism and outgroup contention (see Norman Reference Norman2022). Our analysis finds this heterogeneity incredibly important: right-wing Israeli respondents consider all Palestinian actions forms of terrorism, whereas left-wingers maintain clearer hierarchies by severity. This pattern persists in the US, indicating that Israel is not unique. Moreover, we demonstrate that the greatest susceptibility to terrorism labels lies in the center, where many have weaker ideological convictions than in the extremes. If labeling nonviolence as terrorism makes hawks of centrists, it can establish broader public coalitions in favor of harsh security policies and against concessions. Thus, exploring the implications of such labeling practices in nonviolent domains and their heterogeneous influence is key to understanding processes of conflict escalation and resolution.
The paper proceeds with an overview of relevant research about popular perceptions of terrorism and nonviolence. We subsequently suggest several hypotheses on how perceptions of terrorism should apply to nonviolent actions. We then introduce our case study and experimental design, followed by our findings and an exploration of several key mechanisms. We conclude with a few broader takeaways.
2. Public perceptions of terrorism: what we know
Terrorism has been closely studied for decades (Schuurman, Reference Schuurman2020). Despite conceptual debates, formal definitions of terrorism emphasize violence against civilians for political motivations. Chenoweth (Reference Chenoweth2013, 356), for example, defines terrorism as “the deliberate use or threat of force against non-combatants by a non-state actor in pursuit of a political goal.” Similarly, the Global Terrorism Database conceptualizes it as “the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious, or social goal through fear, coercion, or intimidation” (Dugan et al., Reference Dugan, LaFree, Cragin and Kasupski2008).Footnote 1 Terrorism, accordingly, is used to affect public opinion both by its perpetrators (Leeman, Reference Leeman1986; Kydd and Walter, Reference Kydd and Walter2006; Enders and Sandler, Reference Enders and Sandler2012) and as a pejorative label against rivals' actions (Kapitan and Schulte, Reference Kapitan and Schulte2002; Meier, Reference Meier2020). Thus, public perceptions and reactions to this concept are highly important.
In a seminal exploration of this issue, Huff and Kertzer (Reference Huff and Kertzer2018) test how violent action characteristics affect the likelihood that Americans classify it as terrorism. Consistent with formal definitions, the severity and type of violence and whether it is motivated by politics or group hatred are particularly influential. Nevertheless, the perpetrator's identity and affiliation also have an independent influence. They conclude that folk definitions of terrorism largely follow formal ones in focusing on violence and harm, but, in many cases, deviate from these boundaries.
Newer studies expand on the role of such additional factors, finding that judgments vary with ingroup and outgroup affiliation and political views. In the US, violence by one's outgroup or against favorable targets is more likely to be considered terrorism and morally unjustifiable than identical actions by one's ingroup or for supported ideals (Norman, Reference Norman2022). This is especially true when the outgroup is racially stereotyped as connected to terrorism, as with Islamophobic audiences and Arab Americans (D'Orazio and Salehyan, Reference D'Orazio and Salehyan2018; Baele et al., Reference Baele, Sterck, Slingeneyer and Lits2019). This tendency manifests differently across the ideological spectrum: American liberals are more likely to view all political violence as terrorism, including for liberal causes, whereas conservatives judge violence by either camp less harshly (Norman, Reference Norman2022). Public discourse, biased reporting, and patterns of media consumption also correlate with assessments of violence as terrorism (Baele et al., Reference Baele, Sterck, Slingeneyer and Lits2019; Kearns et al., Reference Kearns, Betus and Lemieux2019; Dolliver and Kearns, Reference Dolliver and Kearns2022; Norman, Reference Norman2022).
However, in keeping with accepted definitions of terrorism, these studies focus only on physical violence. Nevertheless, nonviolent actions are important to this debate. Nonviolent tactics are an alternative route for political change, often with greater odds of success (Chenoweth and Stephan, Reference Chenoweth and Stephan2011; Braithwaite and Braithwaite, Reference Braithwaite and Braithwaite2018; Chenoweth, Reference Chenoweth and Johnston2019; Wasow, Reference Wasow2020). Chenoweth and Stephan (Reference Chenoweth and Stephan2011) suggest that nonviolence has a “participation advantage” that facilitates supporter mobilization and makes its repression harder. For this to succeed, however, target audiences must distinguish nonviolence from violence.
Recent studies call this premise into question. Experiments conducted in Israel and the US find that nonviolent protest is perceived as more violent, justifies greater state repression, and less likely to succeed when held by outgroup members that invoke negative stereotypes (Edwards and Arnon, Reference Edwards and Arnon2021; Manekin and Mitts, Reference Manekin and Mitts2022). Nevertheless, these works measure nonviolence using contentious mass protests, which can more easily seem violent and threatening. Yet active disputes across the world display a variety of strictly nonviolent resistance tactics, including economic, social, and political actions, typically analyzed from the protestors' perspective (Cunningham et al., Reference Cunningham, Dahl and Frugé2017). To the best of our knowledge, target audiences' public perceptions of nonviolent acts have not been similarly explored.
To conclude, meaningful advances notwithstanding, current research on popular perceptions of terrorism exclusively studies violence or physically threatening protests. In this domain, audiences are more likely to classify actions as terrorism the more harmful they are, but also by respondents' group affiliations, perceptions of the perpetrators, partisan ideology, and action framing. This leaves open several questions about the full spectrum of actions by disliked outgroups and their perception as illegitimate terrorism. Do audiences distinguish between violence and nonviolence in their definition of terrorism? Do perceptions differ by the type of action? And how are such views influenced by ideology and elite labeling? In what follows, we propose and test several preregistered hypotheses regarding these questions.
3. Terrorism and nonviolence: hypotheses
Our core argument posits that in conflictual contexts, which involve a known and disliked adversary, a broader range of nonviolent actions can be perceived as close to terrorism. Such perceptions, we suggest, rely on similar criteria applied to physically violent actions. We propose several key considerations: the level of violence and nonphysical harm to the ingroup, ideological predispositions about the conflict and adversary, and the act's public labeling. For focus, we hold constant the adversary's identity and negative affect, a well-established finding in the literature, although we later explore the conflict's salience as a potential mechanism.
First, while nonviolent actions may be perceived as forms of terrorism, we hypothesize that violence remains a key differentiating criterion. The literature on public perceptions of terrorism consistently indicates that audiences are sensitive to the act's nature, especially its degree of violence. Although the definition of violence is itself contested, we adopt the common sense of intention to cause physical harm (Pressman, Reference Pressman2017).Footnote 2 Hence our first hypothesis:
H1.a (Violence Hypothesis): All else equal, violent actions by an adversary will be perceived more strongly as terrorism than nonviolent actions.
Nonviolence, however, can still cause nonphysical harm (Pontara, Reference Pontara1978; Pressman, Reference Pressman2017). Some nonviolent tactics can be more harmful to targeted audiences, e.g., by disrupting transportation, causing financial and property damage, or inducing other types of discomfort. Conversely, other nonviolent acts are primarily nonharmful, e.g., by symbolic actions or self-serving tactics promoting the perpetrators' interests. Nonetheless, where intergroup rivalries are perceived as a zero-sum game, even nonharmful actions can seem belligerent. For instance, obtaining contested goods implicitly reduces available resources for other groups and may appear offensive. Hence, we hypothesize that the degree of harm also matters. Whereas violence occupies its own category, harmful nonviolence may seem closer to terrorism than nonviolence with zero-sum harm:
H1.b (Harm Hypothesis): All else equal, more harmful actions by an adversary will be perceived more strongly as terrorism than less harmful actions: physical violence will be perceived more strongly as terrorism than nonphysical harm, and nonphysical harm more strongly than self-regarding actions with zero-sum harm.
As noted, past research finds that ideology and group identities affect perceptions of terrorism and adversary actions. Multiple studies show that right-wing views are associated with greater outgroup resentment and threat (Jost et al., Reference Jost, Glaser, Kruglanski and Sulloway2003; Kam and Kinder, Reference Kam and Kinder2007; Duckitt and Sibley, Reference Duckitt and Sibley2010; Peffley, Hutchison, and Shamir, Reference Peffley, Hutchison and Shamir2015) and with more hawkish, distrustful, and realist attitudes on foreign affairs (Sulfaro, Reference Sulfaro1996; Brewer et al., Reference Brewer, Gross, Aday and Willnat2004; Kertzer and McGraw, Reference Kertzer and McGraw2012). These differences reflect a conservative tendency to prioritize group status and authority compared to liberal sensitivities to fairness and harm avoidance (Graham et al., Reference Graham, Haidt and Nosek2009; Kugler et al., Reference Kugler, Jost and Noorbaloochi2014). They also indicate greater ideological distance from the adversary's positions. Based on these moral and ideological tendencies, we expect that right-wing individuals will perceive nonviolent adversary contention as closer to terrorism compared to left-wingers:
H2 (Ideological Heterogeneity Hypothesis): All else equal, nonviolent actions by an adversary will be perceived more strongly as terrorism by ideologically right-wing individuals.
Apart from an action's objective parameters, its perception as terrorism may also depend on its portrayal. Elite framing can change how people judge various political issues by increasing the salience of certain aspects over others (Chong and Druckman, Reference Chong and Druckman2007; Betus et al., Reference Betus, Kearns and Lemieux2021). Labeling goes beyond framing by attaching a specific category, like “terrorism” or “hate crime”, to an action (Baele et al., Reference Baele, Sterck, Slingeneyer and Lits2019). Some scholars find that “terrorist” or “Islamist” labels increase negative perceptions of violence and support for harsher policy responses (Baele et al., Reference Baele, Sterck, Slingeneyer and Lits2019). Others, however, argue that labeling has a negligible effect on perceived threat compared to emotionally heightened depictions of violence (Feick et al., Reference Feick, Donnay and McCabe2021). The labeling effect of “terrorism” thus remains an open question (Huff and Kertzer, Reference Huff and Kertzer2018; Edwards and Arnon, Reference Edwards and Arnon2021).
Part of this vagueness, we suggest, reflects an embedded link between violence and terrorism regardless of explicit labels. Hence, labeling may play a more significant role in the gray area of nonviolent actions that fall outside classic definitions of terrorism:
H3 (Labeling Hypothesis): All else equal, nonviolent action by an adversary labeled as terrorism will be perceived more strongly as terrorism than an action not labeled as such.
Like judgments by action type, labeling, too, may have heterogeneous effects by ideological preconceptions. Nevertheless, political ideology may interact with labeling in contrasting ways. Terrorism labels might have a stronger effect on those predisposed to negatively judge the adversary's actions. Conversely, if baseline views are already negative, an explicit terrorism label may add little and be most effective on left-wingers with more room for change. These possibilities establish two competing hypotheses:
H4.a (Labeling and Ideology—Propensity Hypothesis): All else equal, labeling nonviolent actions by an adversary as terrorism will cause a greater increase in their perception as such by right-wing individuals.
H4.b (Labeling and Ideology—Ceiling-Effect Hypothesis): All else equal, labeling nonviolent actions by an adversary as terrorism will cause a greater increase in their perception as such by left-wing individuals.
4. Context: the Israeli case study
We test our hypotheses using original data collected in Israel. Given its salient violent conflict with the Palestinians and long-standing democracy, Israel is a paradigmatic case study for public opinion in conflictual contexts (Phillips and Greene, Reference Phillips and Greene2022; Godefroidt, Reference Godefroidt2023). For our purposes, it offers several meaningful advantages. First, the conflict involves a known and disliked adversary. Indeed, Israeli-Jewish society maintains a stably negative public ethos about Palestinian goals and motivations (Oren, Reference Oren2019). This allows us to hold constant exogenous variation in the adversary's identity, their attributed motivation, and common stereotypes and biases (Huff and Kertzer, Reference Huff and Kertzer2018; Edwards and Arnon, Reference Edwards and Arnon2021; Dolliver and Kearns, Reference Dolliver and Kearns2022; Manekin and Mitts, Reference Manekin and Mitts2022).
Second, internal ideological differences regarding the conflict are well-developed and salient, form Israel's primary partisan axis (Shamir and Arian, Reference Shamir and Arian1999; Yakter and Tessler, Reference Yakter and Tessler2023), and strongly align with perceptions of Palestinians and minorities (Shamir and Shikaki, Reference Shamir and Shikaki2010; Peffley et al., Reference Peffley, Hutchison and Shamir2015). Hence, Israeli partisanship is a straightforward measure for individual-level preconceptions of the adversary, where the right marks the hawkish end of the spectrum and the left its dovish side.
Third, the Israeli context increases the plausibility of our research design. The conflict with the Palestinians includes multiple repertoires of contentions, including violent, diplomatic, legal, economic, and civil resistance tactics. Moreover, Jewish-Israeli elites from both ideological sides have referred to Palestinian nonviolence as terrorism, rendering such labeling believable. Examples include denotation of economic pressure, legal measures, and even construction in contested territories as forms of terrorism (Zarchin, Reference Zarchin2009; Schaeffer Omer-Man, Reference Schaeffer Omer-Man2014; Baruch, Reference Baruch2016; Lis Reference Lis2021).
These advantages serve our goal of exploring causal effects by newly-generated hypotheses but also force a trade-off with generalizability (Gerring, Reference Gerring2007). Whereas our analysis leverages the protracted and salient nature of the Israeli-Palestinian conflict, its patterns may be weaker or more complex in shorter or lower-intensity conflicts. Relatedly, the explicit labeling of nonviolent outgroup actions as terrorism may seem more outlandish in less salient contexts.
To counteract these limitations, we collected additional data regarding a low-salience conflict in the US. Our main findings replicate although with milder intensity, a point we discuss later. Moreover, past comparative research shows similar patterns of popular responses to violence in Israel and other Western democracies (Kibris, Reference Kibris2011; Christensen and Aars, Reference Christensen and Aars2017; Brouard et al., Reference Brouard, Vasilopoulos and Foucault2018; Canetti et al., Reference Canetti, Hirsch-Hoefler, Rapaport, Lowe and Muldoon2018; Nussio, Reference Nussio2020; Edwards and Arnon, Reference Edwards and Arnon2021; Manekin and Mitts, Reference Manekin and Mitts2022; Godefroidt, Reference Godefroidt2023). Our introductory anecdotes further demonstrate that labeling nonviolence as terrorism is not unique to Israel. Hence, with proper caution, our findings seem sufficiently generalizable and instructive to other conflictual contexts.
5. Research design
5.1 Sample
Our empirical analysis uses data from a preregistered survey-experiment conducted on a sample of Israeli Jews, the primary group engaged in the conflict with the Palestinians.Footnote 3 The survey was fielded by Israeli online polling firm Midgam on May 17–18, 2022, using quota sampling representing the party-vote distribution of Israeli Jews in the 2021 election. The final sample includes 2005 respondents of 13,127 invited panelists. Demographic distributions, detailed in Section A.1 in the Supplementary Information (SI), find the sample representative in terms of gender and slightly younger, more educated, and more secular than the adult Israeli-Jewish population. While there were some increased tensions in the months leading up to the survey, it was conducted at a calmer time.Footnote 4
5.2 Experimental design
Our experiment showed respondents a short sentence describing a fictional but realistic Israeli condemnation of a certain Palestinian action. We manipulated two elements in this sentence: (1) the type of Palestinian action and (2) whether it was labeled explicitly as terrorism. For action type, we randomly assigned one of four actions:
(1) Violence: “Palestinian attempts to kill and injure Israeli citizens.”
(2) Economic sanctions: “Palestinian attempts to promote an international economic boycott of Israel.”
(3) Legal action: “Palestinian petitions against Israel to the international court at the Hague.”Footnote 5
(4) Illegal construction: “Palestinian attempts to build illegally in Area C territories in the West Bank.”Footnote 6
These actions are designed to test our first two hypotheses. To evaluate the influence of violence (H1.a), we contrast a violent act with three nonviolent actions that fall outside formal definitions of terrorism. To assess differences by harm (H1.b), these actions pose varying degrees of threat to Israelis. Economic and legal actions threaten with nonphysical harm: the former harms ordinary citizens' material welfare and the latter combatants' legal status and Israel's international standing. Illegal construction, meanwhile, is self-regarding: it advances Palestinian well-being without targeting Israelis but can seem harmful in a zero-sum game over territorial expansion and rule enforcement. A manipulation check, detailed in SI Section A.5, verifies that the action type affected respondents' emphases on violence and harm.
To increase plausibility, we selected actions rooted in the conflict's reality and labeled as terrorism at least once by real-world Israeli politicians. Rhetorically, this labeling typically attaches the action type as an adjective to “terrorism” (“economic terrorism”, “legal terrorism”, etc.). Accordingly, the second element of our manipulation, designed to test the labeling hypothesis (H3), embeds the action in one of two randomly assigned sentence structures:
(1) No label: “Israeli leaders recently condemned [Palestinian action].”
(2) Terror label: “Israeli leaders recently condemned the Palestinian [violent/economic/legal/construction] terror (expressed in [Palestinian action]).”
This design follows recent recommendations by Brutger et al. (Reference Brutger, Kertzer, Renshon, Tingley and Weiss2023) about the proper level of abstraction in survey experiments. First, given their finding that hypothetical and real-world scenarios produce similar results, we present general Palestinian actions rather than overly-concrete actual events. Second, we do not mention speakers by name to prevent inflated or biased effects due to partisanship. Third, our minimal design avoids rich contextual details with added noise and cognitive load that can weaken the effect.
The two conditions form a 4 × 2 factorial design with eight treatment arms, summarized in Table 1. Each treatment group includes 245–259 respondents. A balance test verifies that the randomization produced demographically equivalent groups. SI Sections A.2–A.4 detail the survey questionnaire, balance test, and power analysis.
Experimental design
5.3 Dependent variables: terrorism perceptions
Measuring perceptions of an action as terrorism is not straightforward. Denoting an adversary's act as terrorism can signify two meanings: (1) genuine classification as terrorism akin to paradigmatic violent acts, and (2) hyperbolic terminology signaling general disapproval of the action.
To overcome this ambiguity, we use three dependent variables. After reading the treatment, respondents were asked about their level of agreement with three statements, presented in a random order, on a scale of 1 (strong disagreement) to 10 (strong agreement). The first variable measures straightforward denotation as terrorism by gauging agreement that “the noted Palestinian actions are an act of terror against Israel.”Footnote 7 The two other questions help distinguish between the underlying substantive and declarative dimensions. We assume that substantive classifications entail similar policy solutions as violent threats, whereas symbolic declarations reflect mere disapproval. To evaluate the former, we gauge respondents' willingness to use repressive force against the action: “Israel is justified in using force, including by its security apparatuses (the IDF, Shin Bet, and Mossad), to stop and disrupt the noted Palestinian actions.” The reference to security agencies explicitly invokes anti-terrorism policies and weighs against symbolic interpretations. The third variable, by contrast, aligns better with symbolic disapproval by capturing the action's perceived Illegitimacy: “even if I disagree with them, the noted Palestinian actions are legitimate actions.” We reverse the scale so that higher values reflect lower legitimacy. Figure 1 summarizes the three variables' distribution and mean scores across all treatment groups.
Dependent variable distributions.
5.4 Additional covariates: partisanship and respondent attention
Our primary analysis includes two added variables. First, to assess heterogeneous effects by ideology (H4.a and H4.b), we measure respondents' partisanship by their recalled vote in the recent election. Given Israel's multiparty system, we classify parties into three ideological blocs—left, center, and right—based on their positions regarding the conflict.Footnote 8 In Israel, right and left are defined primarily by this domain: right-wingers espouse hawkish-conservative views and left-wingers endorse dovish-liberal positions on the conflict. Centrists, meanwhile, hold ambivalent views combining support for some territorial compromise with distrust of the Palestinians (Manekin et al., Reference Manekin, Grossman and Mitts2019; Yakter and Tessler, Reference Yakter and Tessler2023). Recalled party votes were recorded by the polling firm prior to the survey.
Second, our data contains a mid-survey attention check using an instructed-response item asking to mark a specific answer on a grid (Gummer et al., Reference Gummer, Roßmann and Silber2021). Since attention may correlate with politically relevant attributes, we include it as a control indicator instead of screening those who failed (Berinsky et al., Reference Berinsky, Margolis and Sances2014). 125 respondents (6.2 percent of our sample) did not pass the check, balanced across treatment groups. SI Section C summarizes the descriptive statistics of all key variables.
6. Findings
6.1 Action type
To test the influence of action type on its perception as terrorism, we pool the different treatment arms in two ways. First, to test the importance of violence (H1.a), we code a binary variable contrasting all nonviolent actions with physical violence. Second, to consider whether the level of harm exerts influence (H1.b), we create a categorical variable that indicates each of the four actions separately. We then regress each dependent variable on the binary and then the detailed action variables. In all models, violent actions serve as the baseline category.
Table 2 presents the results. Models 1, 3, and 5 strongly support our violence hypothesis (H1.a). Aggregately, nonviolent actions receive lower terrorism scores, seem more legitimate, and provide lower justification for repressive force than violence. Substantively, the average effect magnitude ranges from 1.3 points for illegitimacy to 1.8 points for terror denotation on a 1–10 scale. The consistent pattern across all three outcomes implies that stronger terrorism denotations reflect both higher illegitimacy and an increased willingness to use force against nonviolent acts. This lends some support to Chenoweth and Stephan's (Reference Chenoweth and Stephan2011) expected participation advantage.
The influence of action type on perception of terrorism (OLS Regression)
Standard errors in parentheses, † p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001.
Models 2, 4, and 6 break down nonviolence into distinct action types. The estimates show mixed support for the harm hypothesis (H1.b). The expected difference between nonviolent harm (economic and legal actions) and self-serving actions (illegal construction) is confirmed for terror denotation: economic action scores closer to violence than legal action (F 1,2000 = 4.51, p = 0.03), which in turn rates higher than illegal construction (F 1,2000 = 6.51, p = 0.01). However, these differences blur when judging legitimacy: economic sanctions seem less legitimate than legal petitions (F 1,2000 = 5.83, p = 0.02) as we expect, but legal and construction actions rank similarly (F 1,2000 = 1.65, p = 0.24). In addition to harm differences, the harsher judgments of economic actions may also reflect the public salience of the Boycott, Divestment, and Sanctions movement compared to other nonviolent actions (Thrall, Reference Thrall2018).Footnote 9 Meanwhile, state force is least justified against legal petitions compared with both economic sanctions (F 1,2000 = 3.08, p = 0.08) and illegal construction (F 1,2000 = 4.50, p = 0.03). Hence, we do not find a consistent harm-based hierarchy across all outcomes, only for terror denotation. Nevertheless, economic sanctions, which have more tangible costs and salience, are repeatedly judged more harshly.
These differences notwithstanding, the results in Table 2 exhibit strikingly high values across all actions and outcomes. On a 10-point scale, the average scores for denotation range from 5.9 to 8.1, for action illegitimacy from 6.6 to 8.3, and for use of force from 6.2 to 8.1. Hence, all nonviolent Palestinian actions are ranked closer to violent terrorism than not.
We also hypothesize that judgments should vary by ideology, such that right-wing individuals consider nonviolence closer to terrorism (H2). To test this hypothesis, we re-estimated our models while interacting the action type (violence/nonviolence) with respondents' ideological bloc.Footnote 10
Figure 2 plots the estimated predicted values.Footnote 11 The results corroborate H2: partisan ideology strongly influences perceptions of nonviolence. Right-wing respondents rank nonviolence close to violence in all aspects, including high willingness to apply force. As hypothesized, these belligerent perceptions of nonviolence diminish as we move leftward: left-wingers perceive the greatest difference between violence and nonviolence and centrists locate in between. Nevertheless, the absolute scores for all action types remain closer to terrorism than not. Only left-wingers score nonviolence below the mid-scale point (5.5), marked in dashed horizontal lines.
Predicted values of violent versus nonviolent actions by partisanship. The vertical lines mark 95 percent Confidence Intervals. The dashed line marks the mid-scale point.
6.2 Labeling nonviolence as terrorism
Are nonviolent actions more likely to be perceived as terrorism when explicitly labeled as such? We coded a dummy variable dividing all nonviolent treatments by whether they are labeled as terrorism. We then regressed our outcomes on this variable while controlling for respondent attention.
The results, shown in Table 3, largely support H3. Explicit labeling of Palestinian nonviolence as terrorism increases respondent agreement with this denotation. Nevertheless, the size of the labeling effect is relatively small, raising their average denotation score from 6.4 to 6.8 compared to unlabeled actions. Importantly, rather than a symbolic expression of illegitimacy, the effect is most pronounced in substantive willingness to employ force, albeit slightly below the 95 percent threshold (p = 0.088).Footnote 12
The influence of terror label on perception of terrorism (OLS Regression)
Standard errors in parentheses, † p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001.
Interacting labeling with ideology clarifies the picture. Figure 3 plots the predicted values of these estimations. We find that the small labeling effect for the full sample conceals meaningful partisan differences, although differently than how we expected. Our preregistered hypotheses suggested labeling should primarily affect right-wingers (H4.a) or left-wingers (H4.b), yet its influence is concentrated among centrists. Substantively, it shifts centrists' terror denotation score from 6.1 to 6.9 and their support for use of force from 6.3 to 7.2 on average.
Predicted values of terror label by partisanship. The lines outline 95 percent Confidence Intervals.
This shift is meaningful: labeling moves centrists—who resemble swing voters on the conflict—rightward in support for militant repression. To illustrate this point further, we coded a dummy variable indicating high support for use of force against an action (8–10 on the 10-point scale). We then ran a logit regression estimating the probability of such high support based on an interaction of labeling and partisanship. The results, detailed in SI Section D.2, estimate a 53.2 percent probability for high support among centrists when such a label is used compared to 38.4 percent without it, a 38.6 percent increase.
6.3 Robustness tests
Several tests, detailed in SI Section E, validate the robustness of our findings. First, to confirm that our results are not influenced by unmeasured demographic traits, we re-estimated our analyses controlling for sex, age group, religious identification, education, immigration, and geographic region.Footnote 13 The results remain substantively unchanged.
Second, although our treatment mentions unnamed “Israeli leaders”, we verified that unobserved perceptions of speaker partisanship do not bias the results. After showing the treatment and measuring our outcome questions, we also asked respondents to place the noted speakers on a left-right scale from 1 to 7. Including this covariate does not change our findings.
Third, to validate our measure of partisanship, we re-estimated our interaction models while using respondents' left-right self-identification on a 1–7 scale instead of voting blocs. This measure adds non-voters to the sample but levies a higher cost in statistical power. Since we find that labeling affects centrists most strongly—i.e., highest in mid-scale values—we interacted our treatments with the squared value of self-identification. The findings remain substantively similar.
Finally, to verify that our 10-point scale does not inflate variation artificially, we re-estimated our models with a collapsed 5-point scale. The results remain robust.
7. Mechanisms
While our design focuses on causal effects, automated textual analysis and additional data collection suggest three central mechanisms: the primacy of physical threat, partisan moralities, and conflict salience. The first mechanism helps explain why most respondents rank violent actions closer to terrorism than nonviolent acts. Following Huff and Kertzer (Reference Huff and Kertzer2018), we expect that physical threat against civilians, a staple of formal definitions, is a primary folk criterion when judging violence and nonviolence. To gauge this mechanism, we included a follow-up open question asking respondents to briefly explain their chosen terrorism-denotation score. Using these answers, we estimate a relative frequency analysis to identify the words used most distinctively in the violence versus nonviolence conditions.Footnote 14
Figure 4 displays the fifteen most distinctive words used by respondents in each group. The bars represent each word's keyness score (χ 2), indicating how relatively more frequent it is in each group of interest. We omit rare words appearing in fewer than five observations. The results support the expected mechanism. Respondents in the violence conditions, who judged the actions more harshly, were more likely to mention their lethality and severity (“murder”, “crime”, “dead”, “killing”, “wounded”, “axe”, “violence”), civilian targets (“civilians”, “person”, “Jewish”, “population”), and political motivation (“terrorism”, “nationalist”, “goal”, “point”, “reason”), all included in accepted terrorism definitions. Nonviolent actions, conversely, prompted more neutral action descriptions (“economy”, “boycott”, “area”, “legal”, “construction”, “territories”, “law”, “The Hague”) and vaguer goals and threats (“Israel”, “country”, “attempting”, “took over”, “world”).
Relative frequency analysis of word in the violence and nonviolence treatments.
Second, our findings showed that right-wing respondents rank nonviolence closer to violence than left-wingers. Different partisan moralities may help explain this difference. Past research finds that right-wing conservatives tend to value ingroup status and protection from outgroup threats, whereas left-wing liberals prioritize fairness and care (Jost et al., Reference Jost, Glaser, Kruglanski and Sulloway2003; Graham et al., Reference Graham, Haidt and Nosek2009; Kugler et al., Reference Kugler, Jost and Noorbaloochi2014). To gauge this mechanism, we estimate a relative frequency analysis of the most characteristic words in each partisan group when treated with nonviolent actions.
Figure 5 compares the fifteen most distinctive words used by right-wing and left-wing respondents. The results are consistent with different partisan moralities. When prompted with Palestinian nonviolence, right-wingers are more group-centric: they are more likely to invoke group labels (“Jewish”, “Israel”, “Arab”), ingroup sovereignty (“land”, “country”, “territories”), and outgroup threat (“took over”, “killed”, “dangerous”, “murderer”, “encouraging”, “change”). Conversely, left-wingers are more reserved (“is not”, “unsure”), descriptively refer to the actions (“legal”, “building”, “area”, “action”, “active”), and allude to the other side's perspective (“right”, “should”, “settler”, “home”).
Relative frequency analysis of words by left-wing and right-wing respondents in the nonviolence treatments.
Finally, the conflict's high salience may serve as a mechanism explaining the high absolute scores assigned to nonviolent actions. Higher conflict salience can deepen the sense of outgroup threat, entrench negative stereotypes, and provide regular cues sustaining these impressions and their cognitive availability. In such contexts, any adversary action may seem harmful by default.
To consider this mechanism, we replicated our experiment in a lower-salience conflictual context: American attitudes about Iranian actions.Footnote 15 According to a Gallup survey from February 2022, 84 percent of Americans see Iran unfavorably but only 2 percent noted it as the country's greatest enemy, indicating an existing but low-level conflict. The experiment was fielded online by the Harvard Digital Laboratory for the Social Sciences (DLABSS, see Strange et al., Reference Strange, Enos, Hill and Lakeman2019) on August 11–30, 2022. The convenience sample includes 1135 Americans (144–152 respondents per treatment) and is somewhat older, more educated, richer, and more male-dominated than the adult US population. SI Sections B.3–B.4 summarize the sample's demographic distributions, balance, and power analysis.
The design is near-identical to the Israeli survey, with two notable changes. First, since illegal construction does not apply to the US–Iran context, we replace this condition with “financial terrorism”: “Iranian attempts to violate US sanctions on trade.” Like illegal Palestinian construction, it underscores self-regarding gains and unapproved behavior rather than nonviolent harm. Second, we measure ideology using respondents' self-identification as liberals, conservatives, or independents. Independents indicating ideological lean are classified into the other two groups accordingly. SI Section B.2 details the full US questionnaire.
We examine two outcomes: Figure 6 displays the predicted values for violence versus nonviolence and Figure 7 compares the predicted values of labeling, both broken down by ideology (comparable with Figures 2 and 3, respectively).Footnote 16 Supporting the salience mechanism, the results replicate our findings but with lower magnitude. Like in Israel, nonviolent acts are ranked lower than violence but less so among right-wing conservatives. Furthermore, labeling effects are again strongest (and substantively similar) among independents, who shift from liberal-like to conservative-like evaluations. Yet the intensity of these judgments is lower: Americans rank nonviolent actions on the bottom half of the terrorism scale, though still above its lowest value. These data indicate that conflict salience affects the severity of judgments regarding an adversary's nonviolent actions. Still, it does not change their relationship with the action's nature, respondent ideology, and labeling, which generalize across contexts.
Predicted values of violent versus nonviolent actions by ideology, US sample. The vertical lines mark 95 percent Confidence Intervals. The dashed line marks the mid-scale point.
Predicted values of terror label by partisanship, US Sample. The lines outline 95 percent Confidence Intervals.
8. Conclusion
Although terrorism is commonly defined by actual or threatened violence, the term is often invoked to decry nonviolent actions by adversaries. However, little is known about the willingness of audiences to consider nonviolent campaigns a form of terrorism. Exploring this understudied question, we argue that subjective perceptions of terrorism are a spectrum that includes nonviolence too. Whether nonviolence seems akin to terrorism and justifies repression depends on the action's nature, the audience's ideology, and elite labeling. This fluidity differs from formal terrorism definitions, which strictly distinguish violence from nonviolence.
We support this argument using a preregistered survey-experiment in Israel, an active conflictual setting with violent and nonviolent repertoires of contention, and a complementary US replication. In terms of categorization as terrorism, Jewish Israelis consider nonviolent resistance tactics by Palestinians relatively close to violence. Moreover, they perceive them as illegitimate, and, like terrorism, justify their forceful repression. These perceptions vary with the action's degree of physical threat and respondents' ideology and partisan moralities. Ideology also determines susceptibility to explicit terrorism labels: hawkish and dovish respondents are relatively unaffected, but labeling sways centrists to judge outgroup nonviolence more hawkishly. These relationships replicate in a low-salience conflictual context in the US, albeit with milder absolute judgments of nonviolence, implying broader generalizability.
These results are important for several reasons. First, they demonstrate that public perceptions of terrorism are not restricted to violence. Thus, target audiences are willing to diverge from formal definitions of terrorism further than previously suggested. More research is warranted into what other nonviolent actions, and under which conditions, audiences are willing to accept as terrorism and justify the use of coercive force.
Second, our findings carry normative implications. Part of the motivation for the renewed interest in nonviolence is the moral premise that less violence is better and harder to defeat. Our findings imply that leaders who wish to suppress nonviolent campaigns can invoke the pejorative sense of terrorism without falsely claiming that they are violent, as Edwards and Arnon (Reference Edwards and Arnon2021) worry they would. Further, if nonviolent contention is portrayed, perceived, and quelled like violence, the resisting side may find actual violence more effective. Thus, labeling nonviolence as terrorism may become a self-fulfilling prophecy. Our evidence from the Israeli-Palestinian case lends credence to this concern but establishing whether it can indeed occur merits follow-up exploration.
Third, our analysis underscores the pivotal role played by swing audiences in such processes. The different perceptions of adversary actions by dovish and hawkish audiences reflect deep-seated moralities that deem them less swayable by elite labels. It is the centrists and independents, who have weaker ideological convictions, that are most susceptible to such efforts. This aligns with recent research showing that nonpartisans are the most affected by international actors when choosing between violent and nonviolent contention (Shelef and Zeira, Reference Shelef and Zeira2022). Hence, competition over the center seems critical for ideologues on both sides who wish to establish broader dovish or hawkish coalitions in conflictual contexts. Our analysis indicates that labels and rhetoric are key tools in this arena.
Finally, we identify several avenues for future research. The first should explore additional contextual factors determining negative perceptions of adversary nonviolence. Our mixed findings regarding the internal hierarchies of nonviolent actions by harm imply that other factors are also at play (e.g., conflict type, media attention, adversary characteristics, conflict history, and others). The second should examine what it means for an act to be classified as violence to begin with. Our analysis underscores the centrality, even if not exclusivity, of such classifications to judgments of terrorism, rival legitimacy, and support for repression. Third, the nontrivial impact of labeling on centrists emphasizes the need for more research on partisan heterogeneity and swing voters in these contexts. Whereas the underlying morals and attitudes of liberals and conservatives have been studied extensively, we need a better understanding of the factors underlying centrist attitudes and behavior on conflictual issues. A final avenue for future research is to generalize the patterns that we find to other rhetorical domains. For example, labels such as “Enemy”, “Nazi”, or “Fascist” are similarly invoked in various partisan and international contentions, raising comparable questions about their public acceptance and implications.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2023.22. To obtain replication material for this article, please visit https://doi.org/10.7910/DVN/ZFHS3J.
Acknowledgements
We wish to thank Daniel Arnon, Victor Asal, Stephen Chaudoin, Erica Chenoweth, Ryan Enos, Josh Kertzer, Noam Lupu, Christoph Mikulaschek, Ethan Miles, Justin Pottle, Nicole Samuel, Bradley Smith, Kevin Troy, Oguzhan Turkoglu, Chagai Weiss, the Harvard Digital Labs for Social Sciences (DLABSS), and the PSRM editor and two anonymous reviewers for their helpful comments and suggestions. Earlier versions of this article were presented at the 2022 annual meetings of the European Political Science Association (EPSA) and American Political Science Association (APSA), at the Online Peace Science Colloquium (OPSC), and at the Harvard International Relations Graduate Workshop. The project was supported by the Israel Science Foundation (Grant No. 2976/21) and the Harvard Center for Jewish Studies. The article preregistration plan is available at https://osf.io/fj9du.
Open access
