LITERATURE REVIEW

Over the past 15 years, academic interest in nonviolent civil resistance campaigns has grown due to the combination of the Color Revolutions, the Arab Spring, and seminal contributions by Chenoweth and Stephan (Reference Chenoweth and Stephan2008; Reference Chenoweth and Stephan2011). Recent research has analyzed domestic (Brancati Reference Brancati2016; Butcher and Svensson Reference Butcher and Svensson2016; Chenoweth and Ulfelder Reference Chenoweth and Ulfelder2017; Dahlum Reference Dahlum2019; Groshek and Christensen Reference Groshek and Christensen2017; Tucker Reference Tucker2007) and international (Braithwaite, Braithwaite, and Kucik Reference Braithwaite, Braithwaite and Kucik2015; Gleditsch and Rivera Reference Gleditsch and Rivera2017; Karakaya Reference Karakaya2018; Murdie and Bhasin Reference Murdie and Bhasin2011) factors driving major nonviolent protests (such as diffusion), as well as the determinants of their success in realizing their primary objectives (Chenoweth and Stephan Reference Chenoweth and Stephan2011; Huet-Vaughn Reference Huet-Vaughn2013; Klein and Regan Reference Klein and M. Regan2018; Nepstad Reference Nepstad2011; Ritter Reference Ritter2015a).

The limited experimental research on domestic public attitudes toward protests reveals two consistent patterns. First, protests described as nonviolent usually elicit more positive responses than their violent counterparts (Feinberg, Willer, and Kovacheff Reference Feinberg, Willer and Kovacheff2020; Muñoz and Anduiza Reference Muñoz and Anduiza2019; Simpson, Willer, and Feinberg Reference Simpson, Willer and Feinberg2018). Second, despite their nonviolent nature, protests led by marginalized groups tend to be viewed more negatively (Davenport, McDermott, and Armstrong Reference Davenport, McDermott, Armstrong, Bonner, Seri, Kubal and Kempa2018; Edwards and Arnon Reference Edwards and Arnon2019), although Manekin and Mitts (Reference Devorah and Mitts2022) find that this negative perception varies by protest goals. Notably, the sole experimental study to date on environmental protests finds that violent environmental protests are seen in a substantially more negative light than nonviolent ones (Thomas and Louis Reference Thomas and Louis2014).

THE IMPACT OF FOREIGN INTERVENTION ON PUBLIC OPINION ON PROTEST MOVEMENTS

We argue that the perception of foreign intervention on behalf of protest movements reduces public support due to two primary reasons.Footnote ⁶ First, protest movements strive to signal high commitment and widespread support for their cause to third parties by showcasing their size or protest activities. When organizers succeed in bringing a significant number of people to their protests, they can plausibly claim widespread support among the non-protesting general population. Similarly, when peaceful protesters attend multiple demonstrations on the same topic and with the same set of demands, and/or repeatedly disrupt daily life, these activities indicate the willingness of protesters to dedicate significant resources to their declared cause (Denardo Reference Denardo1985; Klein and Regan Reference Klein and M. Regan2018).

However, the presence of foreign assistance could cast doubt on the protesters’ commitment. To the broader uninvolved public, the motivation for repeated demonstrations and disruptions could be seen as driven by foreign monetary compensation rather than genuine conviction. Furthermore, resources from foreign powers may enable protest movements with low support to artificially inflate the ratio of protesters to latent popular support within the general population (through better organization by main activists such as free busing services). This dynamics mirrors how well-organized unions can bring thousands or tens of thousands of their members for demonstrations even when their demands have limited or no support among the general public.

Second, the perception of foreign support might cause the protest movement to be perceived as a foreign conspiracy and a threat to national security. Both domestic elites and the public understand that foreign powers may back non-state groups to serve their interests, interests potentially conflicting with the protest organization’s objectives or targeting the public’s interests. As a result, the goals or policy solutions proposed by groups receiving foreign assistance may be seen as serving the foreign power more than the public. The foreign power’s assistance could also allow them some control over the protest group’s activities, reinforcing the perception that the group is a “tool” or even a “fifth column” of the foreign power. This divergence of preferences and possible foreign control amplifies the perceived security threat posed by protest groups receiving foreign aid.

Unsurprisingly, peaceful protest groups who receive foreign assistance are more likely to face repression. A study comparing violent and nonviolent uprisings found that protests receiving foreign support were more likely to be targets of mass killings (Perkoski and Chenoweth Reference Perkoski and Chenoweth2018). For this reason, during the October 2000 Bulldozer Revolution, Otpor, a nonviolent protest group known in Yugoslavia for receiving Western assistance, emphasized in its messaging to the Yugoslav security forces that it was not a Western “fifth column” to prevent a potentially violent crackdown (Binnendijk and Marovic Reference Binnendijk and Marovic2006, 417–8). Therefore, our primary hypothesis is as follows:

We also expect that perceived foreign assistance to protest movements would harm them more severely if a large share of their membership is composed of out-groups (e.g., ethnic, racial, and/or religious minorities). In-groups tend to see protests by out-groups, regardless of their actual nature, in a more negative light and as motivated by harmful intentions. This leads in-groups to be more supportive of the use of repression against out-group protesters (Arnon and Edwards Reference Arnon and Edwards2019; Manekin and Mitts Reference Devorah and Mitts2022).

Furthermore, in-groups frequently believe that members of the out-group are less patriotic than members of the in-group, regardless of the circumstances (Kornweibel Reference Kornweibel2002; Kunst, Thomsen, and Dovidio Reference Kunst, Thomsen and Dovidio2019). Thus, the involvement or potential involvement of a foreign power with a domestic movement largely composed of members of an out-group is seen as more threatening than such foreign involvement with a majority in-group domestic movement.

Not surprisingly, studies on the causes of severe mistreatment of ethnic/national minorities have found that minority groups are more likely to be mistreated when they receive, or are expected to receive, external assistance from a foreign power. Governments perceive actual or potential foreign assistance to out-groups as a greater national security risk due to fears that these minorities will be used by foreign powers as armed “fifth columns” (Downes Reference Downes2006; Mylonas Reference Mylonas2013). While foreign assistance to such nonviolent movements rarely includes armaments, mass demonstrations can pressure governments into making major policy concessions, sap the government’s strength, and, in a few cases, even bring down both democratic and authoritarian leaders (Chenoweth and Stephan Reference Chenoweth and Stephan2011). Thus, preexisting negative perceptions of the out-group compounded with perceived foreign interference are expected to have a multiplier effect: The public will be far less willing to make concessions. This will also lead to greater support for repression against such foreign-backed protesters, leading to our second hypothesis.

Perceived foreign backing of protests can play two non-mutually exclusive roles: instigating protests or “oil spilling” to escalate ongoing protests. The timing of foreign assistance may affect public perceptions. If foreign support comes after protests start, the public may dismiss interference as representing genuine protester interests rather than foreign agendas.Footnote ⁷ In this case, the public may prefer concessions, since foreign hijacking poses a greater risk. However, evidence of foreign backing before protests begin could undermine protester legitimacy. Thus, if foreign backing is perceived as instigating rather than hijacking protests, it will have stronger delegitimizing effects. Thus, our third hypothesis is as follows:

We also argue that the type of foreign assistance to a protest movement may also matter. Perceptions of foreign assistance in the form of funding should elicit more negative reactions than other forms of assistance. There is a widespread belief that control over monetary resources by any actor provides them with control over society, as expressed in the famous American saying “whoever has the gold makes the rules.” Funding is assumed to give funders continuous leverage over recipients’ actions and priorities. This view is reinforced by the understanding that when recipients rely on external funding to operate, a cutoff could force them to scale back or cease activities.

Therefore, even domestic NGOs and protest groups with entirely domestic financing sources are subject to constant media scrutiny, government investigations, public debates, and legal disclosure requirements. When these groups’ domestic funding sources are discovered to originate from supposed or actual “problematic” domestic sources, it sparks significant political controversy (e.g., see Mayer Reference Mayer2010; Soskis Reference Soskis2017).

When funding comes from perceived foreign sources, these dynamics may be magnified. Foreign funding of activist and protest groups is usually problematic by definition, as publics know foreign powers assist non-state actors in promoting divergent interests. Not surprisingly, there are widespread complaints in many countries in non-protest settings, supported by some academic research (Englund Reference Englund2006: Stiles Reference Stiles2002), that foreign INGOs and foreign governments that fund domestic NGOs have undue influence over them. The discovery of a protest movement funded by a foreign power, along with implied leverage, leads to especially negative reactions.

Governments that seek to discredit protest groups focus their fire on the foreign funding that protest organizers allegedly received. For example, the Serbian protest movement Otpor received multiple forms of assistance from Western governments after its creation, including training in nonviolent protest methods and financial funding. Nevertheless, the Milošević regime, in its efforts to discredit the protesters as “Madeline [Albright] Youth,” emphasized in its messaging the financial assistance that Otpor received, even going to the extent of posting wall posters in which Otpor activists were shown with pockets stuffed with U.S. dollars (Sharp Reference Sharp2005, 320–4).

In contrast, other forms of foreign assistance to protesters are not seen as giving the foreign power similar leverage over the protest movement. For example, training in protest and mobilization tactics by a foreign power does not prevent the protest movement from continuing to operate as before with the skills that they acquired until that point in time. Accordingly, the public should be less wary of such foreign support. This leads to our final hypothesis.

RESEARCH DESIGN

To test our hypotheses, we conducted two separate surveys using the online survey firm Lucid. We surveyed 2,563 U.S. adults in August 2020 and 2,664 Canadian adults (preregistered) in March–April 2021. We employed quota sampling to recruit participants, stratifying by age, gender, geographic location, and racial/ethnic representation of the U.S. and Canadian populations.Footnote ⁸ This approach ensures that opt-in Internet surveys are nearly as representative, or at least not worse, than probability sampling through random digit dialing (Kennedy et al. Reference Kennedy, Blumenthal, Clement, Clinton, Durand, Franklin and McGeeney2018; Macinnis et al. Reference Macinnis, Krosnick, Ho and Cho2018). In fact, studies by Coppock and McClellan (Reference Coppock and McClellan2019) have found that Lucid samples perform as well as American National Election Studies.

However, recent studies have found data quality problems with Lucid in which inattentive respondents, or respondents who incorrectly answer attention check questions, could introduce attenuation bias (Ternovski and Orr Reference Ternovski and Orr2022). In their comparison of pre-pandemic and pandemic era samples of Lucid, Peyton, Huber, and Coppock (Reference Peyton, Huber and Coppock2022) not only find that evidence of less attentive respondents leads to smaller estimated treatment effects but also find that these conditions did not change the signs or significance of findings. To remedy possible data quality problems that may affect the representativeness of our sample, we used attention checks and analyzed our results with both attentive and inattentive respondents and found that our treatment effects remain unaffected by these problems (see Supplementary Figure A5.1). We also used quota sampling to ensure that our American and Canadian samples broadly reflect demographics with recent U.S. and Canadian census data (see Supplementary Table A2.1). We did not weigh our sample as quota sampling for online convenience samples is found to be more representative (see Miratrix et al. Reference Miratrix, Sekhon, Theodoridis and Campos2018 and Supplementary Appendix A2 for more details). For the Canadian survey, we preregistered hypotheses and analysis plans in the Open Science Framework repository.

Before describing the survey, we explain our rationale for choosing an environmental protest scenario in the US and Canada. Our case selection criteria needed two key properties for context hypothetical survey experiments: experimental realism and country comparability. Realistic scenarios, based on public priors, help elicit opinions that mirror real-life political dynamics (McDonald Reference McDonald2020). Thus, in the survey experiments, we use real-world names whether they be the identity of the countries interfering or the identity of the marginalized groups participating in the protests. Fielding an environmental scenario in the US and Canada satisfied these properties. Both countries are major fossil fuel producersFootnote ⁹ and have recently experienced large protests over gas and oil pipelines, making our experimental setting plausible and salient to both publics. Additionally, unlike the US, Canada is not a major world power and has not had recent negative experiences with foreign meddling in its domestic affairs. Comparing these countries allows us to identify possible differences in reactions due to these other factors, increasing the wider applicability of our results.

Environmental protests against oil and gas pipelines are common in the US, Canada, and other Western countries. Notable examples include the 2016–2017 Dakota Access Pipeline protests opposing construction through Native American lands, which could endanger water supplies (Estes Reference Estes2019; Gilio-Whitaker Reference Gilio-Whitaker2019). In Canada, Indigenous peoples heavily participated in protests against the Trans-Mountain Pipeline expansion between Alberta and British Columbia (Campbell Reference Campbell2018). Environmental issues like the Greta Thunberg-led climate strikes are a common global protest cause, making them an ideal starting scenario for our hypothetical survey.

These protests often face public accusations from senior officials and media outlets in the US, Canada, and other countries that foreign powers have interfered in environmental protest movements to harm North American and European oil and gas production. For instance, in 2012, Canadian Environment Minister Peter Kent alleged that Canadian environmental NGOs and charities were “laundering” funds from unspecified foreign powers “for inappropriate use against Canadian interests.” Similarly, a 2018 U.S. House of Representatives science committee report claimed that Russia, in an attempt to damage the expansion of fracking in the US, funneled money to unwitting environmental groups opposed to fracking and pipeline expansion in the US, including supporting the Dakota Access Pipeline protest movement (Majority Staff Report 2018). Jason Kenney, head of Alberta’s Conservative Party, also claimed that anti-pipeline protests in Canada were bankrolled by Russia (McElroy Reference McElroy2018).

European leaders have also accused foreign actors of supporting domestic environmental protests to hinder European energy independence from Russia. Former North Atlantic Treaty Organization (NATO) Secretary General Anders Fogh Rasmussen claimed Russia conducted disinformation campaigns supporting environmental groups and protests to maintain European dependence on Russian energy. Officials in Romania and Lithuania made similar accusations, linking local anti-fracking protests to Russian intervention (Higgins Reference Higgins2014). In 2019, a Belgian minister alleged intelligence services informed her that foreign powers directed Greta Thunberg’s climate protests (Boffey Reference Boffey2019). Such accusations are also common in U.S. local referendums on fracking and pipelines (Mooney Reference Mooney2018; Tempus and Horn Reference Tempus and Horn2014).

These claims of foreign meddling on behalf of environmental groups have been strongly denied, and to date, no conclusive evidence has emerged for any of them.Footnote ¹⁰ Indeed, in the Belgian example noted here, the accusation was outright denied shortly afterward by Belgian intelligence agencies. However, for the purpose of our survey experiment, multiple allegations over the past decade by various government officials in the US, Canada, and elsewhere make our scenario plausible and salient to American and Canadian publics.

We made the deliberate choice to emphasize salience and plausibility in our research approach, employing real-world names to identify the actors within our scenarios. Studies, including McDonald (Reference McDonald2020), have demonstrated that using familiar names lessens the cognitive load on respondents and enhances information recall, particularly when the actors are well-known. This can boost the likelihood of eliciting a respondent’s authentic attitudes toward the given actor in hypothetical situations (Brutger et al. Reference Brutger, Kertzer, Renshon, Tingley and Weiss2023). However, prominent actors may introduce some bias due to strong preexisting opinions (Druckman and Leeper Reference Druckman and Leeper2012; Linos and Twist Reference Linos and Twist2018). For example, Greenpeace may be viewed less favorably due to their perceived extreme methods, and the timing of our American survey, conducted 2 months after the George Floyd protests, might cause significant pretreatment effects in vignettes discussing Black rights protest movements.

We posit that any pretreatment effects related to the 2020 Black Lives Matter (BLM) protests were likely minor in the context of our survey. As Linos and Twist (Reference Linos and Twist2018) note, such pretreatment effects depend on the extent to which the experimental vignette mirrors real-world issues. Given the disparity between our hypothetical 2025 environmental scenario and the real-life George Floyd protests against police brutality, we expect significant mitigation of these possible pretreatment effects.

Furthermore, the changes in public opinion related to one of our vignette groups, BLM, began to regress to their pre-protest levels within months of the initial protests. Our survey was conducted in late August 2020, approximately 3 months after the onset of the George Floyd protests (Pew Research Center 2020). This timing further supports our assertion of limited pretreatment effects.

Additionally, using real-world names for marginalized groups, such as Indigenous groups, raises ethical concerns. Although Indigenous peoples are not the object of our study and not subject to ethics review typically associated with biological data, historical artifact, or artistic work, the use of the “likeness” of Indigenous groups without their consent could potentially lead to “subject appropriation” if our hypothetical scenario gained public attention. This could inadvertently perpetuate harmful stereotypes and stigmatize Indigenous communities (Young and Haley Reference Young, Haley, Young and Brunk2009), a concern given their historical exploitation and mistreatment by researchers and institutions (Hayward et al. Reference Hayward, Sjoblom, Sinclair and Cidro2021). We acknowledge that such actions could potentially harm Indigenous peoples’ cultures and identities (Wolfe Reference Wolfe2006).

These potential harms arise from publicizing certain tribal identities, which could have real-world consequences. Orr, Sharratt, and Iqbal (Reference Orr, Sharratt and Iqbal2019) expressed similar concerns in their survey experiment study of Native American tribes in Oklahoma, where the issue paralleled an ongoing, real-world court case. To address these ethical implications, we have implemented two measures. First, our vignette is clearly delineated as a hypothetical future scenario set in 2025, reducing the likelihood of our respondents associating our protest groups with their real-world counterparts. Second, in a related approach to that of Orr, Sharratt, and Iqbal (Reference Orr, Sharratt and Iqbal2019), we use real Indigenous group names in the survey experiment but replace them with pseudonyms in the manuscript and the Supplementary Material. There is a delicate balance between generating useful research and respecting communities (Bird-Naytowhow et al. Reference Bird-Naytowhow, Hatala, Pearl, Judge and Sjoblom2017; Lavallée Reference Lavallée2009). We believe this approach strikes a balance, mitigating harm while preserving experimental realism.

Nonviolent protests serve as a vital tool for Indigenous communities in Canada and the US to protect their rights, especially on environmental issues. We believe our findings provide insights into potential government strategies of implicating these communities with claims of foreign meddling, insights that can be leveraged to defend Indigenous rights, anticipate governmental actions, and strategize suitable responses. To achieve this, our research design must maintain experimental control by ensuring treatment consistency in our scenarios (Brutger et al. Reference Brutger, Kertzer, Renshon, Tingley and Weiss2023). The use of abstract descriptions for marginalized groups, like “Indigenous rights group,” alongside real country names and other real entities, could compromise our experimental control and create inconsistencies for respondents. Using real-world names, in contrast, minimizes differences in treatment effects between real and hypothetical actors, which improves the external validity of our results (Croco, Hanmer, and McDonald Reference Croco, Hanmer and McDonald2021). For a more detailed discussion on the ethical implications of our research on Indigenous communities, see Supplementary Appendix A3.

EXPERIMENTAL DESIGN

The survey began with a hypothetical environmental protest event set in 2025, where protesters attempted to halt a pipeline project extending from Texas (Alberta) to New Orleans (British Columbia).Footnote ¹¹ After the introduction, respondents were informed that the protests were led by either an in-group or an out-group. The in-group was randomized to be a well-known environmental group in the US or Canada, while the out-group was randomized to be a Black civil rights group (BLM or National Association for the Advancement of Colored People (NAACP) in the US or an Indigenous rights group (names withheld) in Canada. As discussed earlier, we used pseudonyms for Indigenous groups and randomized multiple in-groups and out-groups to avoid treatment effects driven by group-specific public reputation. These groups were chosen based on pretest survey results (see Supplementary Appendix A1) relating correlations between groups and perceived racial affiliation (e.g., Greenpeace protesters are predominantly viewed as white).

In each scenario, protesters opposed the pipeline project due to the potential environmental harm it posed to a nearby community. Specifically, respondents viewed the following vignette, with italics highlighting the randomized protest group and the associated community:

Following this vignette, respondents were randomly assigned to one of three groups: no interference, interference (75% intel confidence), and interference (100% intel confidence). Participants in the no interference group received information indicating no interference had occurred. In the interference scenarios, participants were informed that U.S. or Canadian intelligence agencies had either 75% confidence or 100% confidence that a foreign country had interfered with the environmental protest. The 75% confidence treatment was included to mirror real-life statements from intelligence agencies that acknowledge some uncertainty in their assessments and to ensure that our observed effects are not solely due to portraying the foreign meddling as an unequivocally clear-cut case.Footnote ¹³ Specifically, respondents in the interference treatment groups read the following details, with the text in bracketed italics representing the randomized components and parenthetical text provided for the Canadian audience:

Type of support was either “financial assistance” or “organizational training,” the two most common methods of foreign assistance to protests (Chenoweth and Stephan Reference Chenoweth and Stephan2021, 15, 61). Country was drawn from Saudi Arabia, Norway, or Russia; timing of interference was either “six months before the protests started” or “several weeks after the protests started.” For participants who read this vignette, intel confidence, type of support, country, and timing of support were all randomized, resulting in 2 × 2 × 3 × 2 = 24 variations, which were crossed with 2 × 1 = 2 combinations of in-group or out-group protesters. In our analysis, we increased statistical power by analyzing some of the variations for our conditional effects (e.g., country) while averaging over others (e.g., intel confidence). Figure 1 summarizes the timings and survey instruments.

Figure 1. Experimental Vignettes

After assigning participants to interference or no interference group scenarios, we asked participants for their opinions on three sets of Likert-scale agreement questions. The first set asked whether the government should concede to protesters and whether it should employ the police to arrest protesters.Footnote ¹⁵ The second set investigated how foreign interference influences respondent opinions on national energy security and perceived protester commitment.Footnote ¹⁶ We chose energy security to represent national security to parallel how foreign interference frequently frames issues as national security. For example, the Russian intervention in the 2016 U.S. elections led to discussions about “election security.” These questions parallel this logic.

We analyzed the models using ordinal logit, considering only treatment variables as independent variables. We also recorded demographics and political attitudes that could affect protest support, for example, environmental policy, attitudes toward protests, and attitudes toward the specific interfering country. Robustness checks, descriptions, analyses, and discussions can be found in Supplementary Appendix A5.

CAUSAL MECHANISMS AND ROBUSTNESS CHECKS

We expect foreign interference to erode public support for protest movements in two ways: by associating the group as a threat to national security and by creating the perception that these protesters are less committed to their cause. To investigate these mechanisms, we used causal mediation analysis and structural topic modeling on open responses. For the mediation analysis, two questions served as our mediators. The first question asked respondents about their views on energy security, and the second question asked respondents to rate the commitment level of the protest group assigned in their vignette.

Figure 4 displays how these two mechanisms, perceived protester commitment and energy security, mediate how foreign interference affects public support for the government to concede to or repress protesters.Footnote ²¹ The top panels (a–d) confirm that foreign interference reduces public support for protests by directly undermining the credibility of the protest group in question. Specifically, the average causal mediation effect (ACME) through protester credibility is negative on concessions for Americans (ACME: $ -0.033 $ ; $ p<0.001 $ ) and Canadians (ACME: $ -0.03 $ ; $ p<0.001 $ ): that is, foreign interference reduces protester credibility, which decreases public support for concessions toward protesters. The commitment mediator accounts for roughly 18% (Americans) and 15% (Canadians) of the total loss in public support for concessions due to foreign interference. The ACME is positive with respect to repression for Americans (ACME: 0.015; $ p<0.001 $ ) with Canadians reacting more strongly (ACME: 0.032; $ p<0.001 $ ). Here, the commitment mediator accounts for 6% (Americans) and 9% (Canadians) of the total effect of foreign interference on public support repression. In general, as protesting groups lost credibility in the eyes of American and Canadian respondents, respondents became less willing to support government compromise (a–c) and more willing to tolerate repressive tactics by the government (b–d).

Figure 4. Foreign Interference’s Impact of Energy Security and Protester Commitment on Public Support for Protest

Note: This figure plots the average causal mediation effect (ACME), average direct effect (ADE), and total effect. Each half of the figure is split between the American (A,B,E,F) and Canadian (C,D,G,H) results. For point estimates and confidence intervals from this figure, see Supplementary Table A6.1.

The second mechanism proposed, national security threat, investigates how foreign interference on environmental protests may trigger energy security concerns. Here, we observed divergent results between American and Canadian respondents. Americans are particularly concerned with energy security when exposed to the foreign interference instrument, but Canadians are not. On average, heightened concerns about foreign interference in domestic energy supplies decreased American support for concessions by 0.01 ( $ p<0.05 $ ), accounting for 6% of the total effect of foreign interference. At the same time, these concerns increased American willingness to support repressive tactics by 0.037 ( $ p<0.05 $ ), explaining 16% of the total effect of foreign interference. For Canadians, foreign interference did not change public opinion on energy security, which made it a poor mediator of foreign interference’s effect on public support for protests.Footnote ²² This inconsistency across samples may be driven in part by respondent views on the importance of national energy security.

Together, these results provide strong evidence that foreign interference reduces public support for protest movements by casting doubt on the commitment of protesters. The threat mechanism finds weaker support since this relationship affects Americans who are concerned about energy security in the face of foreign interference. However, this mechanism does not affect Canadians because they do not appear as worried about national energy security as their American counterparts.

To further understand the process behind these micro-mechanisms, we also asked in our Canadian survey two open-ended response questions asking participants to explain their choice on whether the government should concede to protesters and on whether the government should arrest protesters.Footnote ²³ We employed a structural topic model (STM) on these open-ended responses to generate topical content (words discussed in the selected topic) approach to compare how foreign interference affects respondents’ justification for their choices. Unlike structured machine learning used to generate topics, STM works more naturally for survey experiments where researchers can integrate the treatment covariate as a natural contributor to topic variance.Footnote ²⁴

We used STM on respondents’ open responses that justify their separate choices for the government to concede to and repress protesters leveraging the model to identify which topics appear more frequently in responses exposed to foreign interference than those that are not exposed to interference.Footnote ²⁵ Following Brutger and Kertzer’s (Reference Brutger and Kertzer2018) approach, we present the results by first comparing the difference in topical prevalence between interference and no interference in Figure 3. Then, we discuss a set of representative comments by respondents that were used to derive the corresponding substantive labels for topical prevalence (for more details, see Supplementary Appendix A7).

Figure 5a displays three distinct substantive topics on respondents’ thought processes behind whether the government should concede between those exposed to the interference treatment and those not exposed to the interference treatment, and Supplementary Table A7.1 shows representative responses associated with these topics. When protest movements do not receive foreign support, respondents focus on the legal basis on which the government should concede to protester demands. “We are a country of law and order,” one respondent claims, “If the government capitulates it will undermine the democratic approval process.” Another respondent concurs, writing, “Unlawful actions should not be tolerated. This project was approved and has been underway.” These comments, which stress the importance of procedurally handling the protests correctly, are significantly more likely to be associated with respondents who did not read about foreign interference.

Figure 5. Difference in Topical Prevalence Between Interference and No Interference

Note: This figure illustrates the shift in topic prevalence among respondents’ comments, from those indicating foreign interference (to the left of the dashed line) to comments suggesting an absence of interference (to the right of the dashed line). This is shown separately for the “concede” (A) and “repression” (B) outcome variables.

In contrast, for respondents who read about foreign interference, their comments focus on the need for Canadian energy security and question the motives of protesters. On the former, one respondent writes, “We need natural resources to be used. We shouldn’t be using foreign sources,” with another respondent elaborating, “We need to continue to harvest and develop our natural resources, to the best of our ability, to be competitive and support Canadian industry.” For the latter, one respondent raises a concern about protester motives, “Because it’s really suspicious, like it has a high chance of the Saudi Arabian government doing this to make no one dominate his spot.” Another respondent goes further and writes, “If the protest was genuine then the concerns should be checked but since it is just a make work project financed by the Saudi government the protested should be sent home.” Together, these comments are consistent with the causal mediation analysis on how foreign interference is mediated by respondent concerns regarding energy security and protester commitment.

For the repression open responses (Figure 5b and Supplementary Table A7.1b), two clear topics emerged for respondents who read about protests without foreign interference: environmental protection and the right to protest. When discussing whether the government should arrest protesters, one respondent writes, “I think that would be too severe of justice for a people trying to protect there home and environment.” Another respondent speaking about the motivation of protesters writes, “They are trying to protect the environment which we all have a responsability to do for future generations.” Similarly, the prospect of mass arrests evoked the following comment from one respondent: “The government should not step in because protesters have the right to protest, peacefully.” Another underscored the fundamental right of protest, writing, “Freedom and Rights of Canadians. It is an essential right to protest.”

For respondents who read about foreign interference (Figure 5b and Supplementary Table A7.1), a single prevalent topic emerged: treason. In these comments, one respondent, displaying their own and unfortunately xenophobic personal view writes, “This is treason and the violators, even and including [Indigenous Group 1], should be jailed and treated as any one committing treason.”Footnote ²⁶ Another respondent believes that protesters “should be treated as a foreign force trying to undermine Canadian sovereignty.” The national interest is invoked: “The protesters are being influence by foreign powers to hinder the national interests of the country,” worries another.

Structural topic modeling is an inductive approach, which contrasts from supervised methods that use our prior expectations on foreign interference’s impact on public support for protests. Yet the patterns that we have uncovered with STM are consistent with our theoretical priors about how foreign interference raises public fears of national security and undermines the credibility of protesters themselves. Moreover, these comments suggest that foreign interference reduces public focus on the environmental costs or even the legal nature of protests, leading instead to a singular focus on foreign threat and protester commitment.

HETEROGENEOUS EFFECTS

Our analysis has thus far focused on examining public reactions to a specific protest group. Foreign collusion may also undermine the causes championed by these groups. For example, Tomz and Weeks (Reference Tomz and Weeks2020) find that foreign meddling erodes public trust in democracy. As such, we also investigated whether foreign intervention changes public preferences for environmental policy regarding fracking and oil pipeline expansion and whether respondent preferences blunt or exacerbate the negative impact of foreign interference. Although we did not formally preregister hypothesesFootnote ²⁷ related to this aspect, the significant substantive and policy relevance of potential conditional effects prompted us to report these exploratory findings.

Figure 6 displays the marginal treatment effect of foreign interference on government repression of protesters conditioned by respondents’ environmental preferences (a–c) and level of patriotism (b–d).Footnote ²⁸ In each subfigure, the vertical axis displays the conditional marginal treatment effect of foreign interference with the horizontal axis displaying the respondent response. It shows that foreign interference increases public support for the repression of protesters, and this effect increases with both more environmentally conscious and nationalistic American and Canadian respondents. For example, foreign interference increases support for repression for respondents in the 20th percentile compared to the 80th percentile of environmentalism by reducing respondents who strongly or somewhat disagree with repression from 2% (8) to 13% (14) and increasing those who strongly or somewhat agree with repression from 4% (11) to 12% (14). This indicates that environmentally conscious respondents feel betrayed by groups that do not genuinely advocate the cause. In fact, the negative effect of foreign interference is only insignificant among a small minority of respondents who are extremely indifferent to the environment (14% of Americans and 8% of Canadians) and who are accordingly less likely to feel betrayed in the first place. Similarly, an increase of nationalism from the 20th percentile to the 80th percentile enlarges support for repression by increasing the percentage of respondents who “strongly agree” with repressive actions from 3% (7) to 15% (11) for Americans (Canadians).

Figure 6. Heterogeneous Marginal Treatment Effects (Repression)

Note: These figures depict the marginal treatment effect of foreign interference conditional on respondent environmental preferences (A–D) and nationalism (E–G) for Americans (circle) and Canadians (diamond). For point estimates and confidence intervals from this figure, see Supplementary Table A4.1.