Message or Messenger? Source and Labeling Effects in Authoritarian Response to Protest

Authoritarian regimes in the 21st century have increasingly turned to using information control rather than kinetic force to respond to threats to their rule. This paper studies an often overlooked type of information control: strategic labeling and public statements by regime sources in response to protests. Labeling protesters as violent criminals may increase support for repression by signaling that protests are illegitimate and deviant. Regime sources, compared to more independent sources, could increase support for repression even more when paired with such an accusatory label. Accommodative labels should have opposing effects—decreasing support for repression. The argument is tested with a survey experiment in China which labels environmental protests. Accusatory labels increase support for repression of protests. Regime sources, meanwhile, have no advantage over non-governmental sources in shifting opinion. The findings suggest that negative labels de-legitimize protesters and legitimize repression while the sources matter less in this contentious authoritarian context.


Introduction
How do authoritarian regimes respond to the threat of mass protests?In the 21st century, regimes rely less on kinetic force such as repression (Davenport, 2007) and increasingly on controlling information flows and shaping citizens' beliefs about events (Guriev and Treisman 2019).Among the strategies of information control, particularly for mass threats, are censorship and distraction in which the regime attempts to prevent the spread of information about protests which could undermine the regime's authority (King et al., 2017;Roberts, 2018).Yet the regime also communicates information about contentious actions directly to its citizens through state media outlets and official statements (Baum & Zhukov, 2015;Peisakhin & Rozenas, 2018;Rozenas & Stukal, 2019).Despite the ubiquity of regimes' information-based responses to protest, we know little about their effects on public opinion.Existing research suggests that attributes of protests affect public opinion (Dahlum, Pinckney and Wig 2022;Hou & Quek, 2019;Manekin & Mitts, 2020;Wasow, 2020), but regimes also present additional information in the form of "editorializing" their responses to protest events which could shape opinion (Carter & Carter, 2021).
In this paper, we argue that protest events in authoritarian regimes receive labels which describe the perceived legitimacy of their participants.Labels of protests may be accusatory or accommodative, either defining the events as illegitimate and their participants as criminals or acknowledging protesters' underlying grievances (Baum & Zhukov, 2015;Cohen, 2011).Accusatory labels are expected to increase citizens' support for repression of protest by pitting them as deviants against the ordering influence of security forces.Accommodative labels, meanwhile, decrease support for repression.Given that regimes often deploy these labels in response to protest events, it could be the case that the regime complements and enhances labels' effects when serving as the source for a statement.Dictatorships' statements have several known effects: inducing compliance (Huang 2015a(Huang , 2015b;;Trinh & Truong, 2020), favorably shifting policy positions (Hou & Quek, 2019;Peisakhin & Rozenas, 2018), and sending signals of strength or trustworthiness (Frye & Borisova, 2019) that make citizens more likely to support or oppose repression depending on whether the regime accuses or accommodates protesters, respectively.
We test the empirical implications of the argument with a survey experiment in China, an authoritarian regime in which control of information is a central strategy in managing public opinion (Huang 2015a, 2015b, 2018, King et al., 2017;Roberts, 2018).It is also a regime which uses a variety of labels-and in which a variety of sources comment-in response to domestic protest events.We focus on different informational responses to environmental protests, events which have become a salient issue drawing significant media and scholarly interest over the past two decades. 1 In the experiment, respondents were randomly presented with vignettes which employed different labels-accusatory, accommodative, and a baseline neutral condition-about an environmental protest which originated from either a government source or an non-government scholar source.We evaluate respondents' post-treatment perceptions, attitudes, and stated behavioral intentions with respect to environmental protests.
Our experimental results reveal a stark contrast between the effects of labels and the effects of sources.On one hand, accusatory labels have substantively large effects on the support for repression of protests and decreased willingness to support protests.These effects are consistent across different estimation strategies.On the other hand, the regime has no advantage over a non-government scholar in shifting citizens' attitudes in the direction intended by the label's content: regardless of whether the label is accusatory or accommodative, the effects of a state media statement about the event are indistinguishable from the effects of a scholar's statement across our key outcome measures.Ultimately, our findings indicate that the message, rather than the messenger, shapes attitudes when authoritarian regimes issue informational responses to protests.
Following the results of our main hypothesis tests, we probe the mechanism through which labels affect citizens' response to protest and repression and consider alternative explanations.We show that accusatory labels deter respondents' sympathy with protests not by shifting attention away from the underlying issue motivating the protest, but rather by changing their perceptions of the protesters and their intended behavior.In particular, respondents do not shift attitudes toward local officials or the underlying policy stakes of the protest even when the central government is the source of the label.We also examine our null finding for source effects, inferring that both government and scholarly sources persuade respondents by showing that perceptions of source credibility and indicators of intimidation do not vary between the government and non-government scholar sources.Our results suggest that, as a common yet rarely studied response to protests in authoritarian regimes, accusatory protest labels de-legitimize protesters and legitimize repression.It also shows that manipulating the views about an existing protest may also prevent future protest.Moreover, while literature on propaganda focuses on effects of messages from state media (Adena et al., 2015;Pan et al., 2021;Peisakhin & Rozenas, 2018;Yanagizawa-Drott, 2014), we compare the content of the message between governmental and non-governmental sources in an autocracy and find the content plays a more important role in shaping citizens' perceptions.
By showing how statements about protest events in autocracies can divide and demobilize citizens, this paper builds a bridge between research on repression and on authoritarian propaganda.Recent scholarship has tended to divide into a focus on a "hard" repression involving state use of kinetic force to contain mass threats (Gohdes, 2020;Sullivan, 2016;Svolik, 2012) and a "soft" repression involving the use of state media and information control to shape citizen attitudes and behavior on issues unrelated to mass threat (Huang 2015b;Peisakhin & Rozenas, 2018;Rozenas & Stukal, 2019).The findings in this paper, particularly outcome tests showing a decrease in willingness to protest when presented with accusatory labels-consistent with Carter and Carter (2021)-builds on research studying how official responses to contention may complement a coercive response (Potter & Wang, 2022) and research showing that depictions of protester violence decrease public support (Dahlum, Pinckney and Wig 2022;Edwards & Arnon, 2021).
Finally, in linking hard and soft repression, this paper also contrasts protest labeling with more ostentatious propaganda designed to achieve social control through generating cults of personality, for example.The ostentatious efforts, described variably as "preposterous" (Huang, 2018(Huang, , 1035)), "phony" (Wedeen, 2015, 6), and "empty" (Kubik, 1994, 42), are designed to enforce compliance through signaling a ubiquitous and unchallenged state.The protest labeling we study, on the other hand, is a plank in 21st-century autocracies' strategy of controlling information through "leading citizens to believe" perspectives on events that favor the regime and disfavor dissidents (Guriev and Treisman 2019, 101).Such pro-regime belief modification may come through distraction (Munger et al., 2019), warnings (Trinh & Truong, 2020), salience (Pan et al., 2021), or persuasion.Our evidence suggests persuasion is the most likely-though not exclusive-channel through which protest labels affect beliefs.Our findings complement this recent research on different mechanisms explaining the effectiveness of regimes' more subtle 21st-century information control and manipulation strategies.

Information-Based Responses to Protests
Protest events-in which a group demands a change in status quo policy in a state 2 -can threaten the survival of a regime in the absence of official response.As these events grow into larger movements, they attract supporters and apply increasing political pressure to a regime (Chenoweth & Stephan, 2011;Tilly, 1978).A lack of official response to a movement only fans the flames of resistance.For example, as protests broke out in Soviet-dominated Hungary in summer 1956, the USSR dithered in its response.Soviet Premier Nikita Khrushchev took more than four months to decide to crush the uprising rather than accommodate its demands, in which time the protests grew into a full-fledged violent revolution (Taubman, 2003).A similar dynamic occurs in democracies.In 2011, the incumbent Socialist government in Spain struggled to respond to protests driven by economic grievances and dissatisfaction with the political system which drew hundreds of thousands of participants (Della Porta, 2015).The Socialist Party was roundly defeated in elections that year.
The challenge for governments is both deciding whether to respond to protest events, and deciding how to respond.Repression, as in the Hungarian Revolution, is one response option.Yet even authoritarian regimes are constrained in the use of repression, as using violence against civilians can both lead to military overthrow (Svolik, 2012) or backlash.Under backlash, with emotion and moral indignation when they witness repression, drawing them off the sidelines and against the government (Pearlman, 2018).Another option for government is to make policy concessions or attempt to co-opt protesters.However, the government may have already be constrained by concessions needed to keep political elites in line (Gandhi, 2008) and by the inefficiency of co-optation (Thomson, 2017).Constrained in their ability to act decisively when faced with protest, leaders respond where they are less constrained: words.
Public statements by regimes are a quick and nearly costless responses to protest.Leaders use public statements to "craft their own narrative…in the face of heightened social unrest" (Barberá & Zeitzoff, 2019, 124).Leaders' statements about protests are an effective response because citizens "[seek] guidance from credible elites," delegating opinion formation on an issue to these elites (Druckman, 2001(Druckman, , 1045)).Political elites are especially likely to be trusted in their statements about issues when those issues are complex or vague (Nicholson, 2011).Protests are often fast-moving and multi-faceted events, involving many tactics, participants, and claims simultaneously.This provides political elites an opportunity to influence opinion about protests with their statements by strategically calling attention to certain tactics or participants in a protest which could build support for the government and/or undermine support for protesters (Edwards & Arnon, 2021). 3 One way in which the public statements influence opinion is the label given to a protest event and its participants.Labels are descriptive, evocative terms that elites apply to protests, and range from sympathetic and accommodative to hostile and accusatory. 4Tilly (2006) observes "the very labeling of a performance as one thing or another regularly has consequences for the participants.During the years of the Riot Act [in Great Britain], authorities who called a worrisome assembly a 'riot' assumed the right to use force against the assembled crowd" (47).O'Donnell (1988) makes a similar point: " 'Rebelliousness', 'subversion','disorder', and 'lack of discipline' are labels affixed to situations that threaten the continuity of what previously were assumed to be the natural attitudes and practices of the dominated classes" (25).Similarly, labeling a contentious event as "terrorism" delegitimates participants and justifies state repression (Huff & Kertzer, 2018).Labels of protests do not have unlimited influence, however.Public statements may be received poorly if they run against the audience's prior beliefs or come from a source with whom the audience disagrees (Boettcher & Cobb, 2009).As a result, unpopular governments could inflame the threat posed by a protest with their response.To sidestep this problem, governments can choose the source which issues a statement.Officials can make statements anonymously, or make them with the imprimatur of an institution, as through a press release or spokesperson (Barberá & Zeitzoff, 2019).
The breadth and function of sources for protest labels differ between democracies and dictatorships.In protest settings in democracies, both protesters and government sources freely deploy labels for contentious events, with each side seeking to persuade the public of their claims' validity (Benford & Snow, 2000).In authoritarian regimes where media freedom is constrained, however, labeling fits within the bounds of what the regime deems permissible.Benford and Snow (2000) illustrate this by showing how Chinese pro-democracy protesters in the 1990s "carefully fashioned and articulated reformist prognoses…consistent with traditional Chinese cultural narration of community devotion and self-sacrifice" (617).
Since the 1990s, labeling around contentious events in China has become more routinized.In "discursive accommodation" (Steinhardt, 2017), part of a broader governance approach known as "consultative authoritarianism" (Repnikova, 2017), scholars argue that the Chinese regime signals approval for protesters' claims through its statements, but not unconditionally.Moreover, the regime allows independent media sources to report on protest, collecting information about citizen grievances while also constraining reporting and limiting independent sources when protests are deemed threatening to regime stability (Lorentzen, 2013(Lorentzen, , 2014;;Repnikova, 2017).Such "uneasy partnerships" between press and regime obscure the distinction between independent media and state sources, meaning the information citizens consume results from selection through the process of "discursive accommodation."

Strategies of Sources and Labels
Despite the ubiquity of these labels and their use by dictatorships in response to protest events, we know little about their effectiveness.Just as coercive responses to protest can sometimes backfire and at other times deter resistance, so could statements about protest increase or decrease opposition to those protests.Generally, we argue that citizens follow a logic of proportionality in the repression they demand in response to protest (Armstrong et al., 2021).A label which portrays protesters harshly will increase citizens' demand for coercion, while a label which portrays protesters generously will decrease demand for coercion.A label which originates from a government source may also accentuate these effects.We outline these expectations in turn.

Accusation and Accommodation
Consider a protest event which makes a claim against the regime, attempting to change a status quo policy or making a maximalist demand such as a change in government (McAdam et al., 2001).Participants in the protest event use some mixture of non-violent and violent tactics, as these events often involve multiple groups of participants carrying out a spectrum of different actions (Pressman, 2017).Further, police used force to break up the protests in an act of repression. 5In a setting with some amount of media freedom, this sequence of events reaches public awareness.As part of media reporting, or in an official response from the government, articles, press releases, and other statements about the event reach print and digital platforms for citizens' consumption.
The content and tone of these statements in response to the event may vary.An accusatory response questions the underlying motives of the protesters and asserts the protests are illegitimate on the grounds of law-breaking (Baum & Zhukov, 2015).This response could be justified through claiming protesters are criminals, terrorists, or thugs.For example, as the 2011 Arab Spring protests in Egypt gained traction, Egyptian government officials took to state media to label participation in the protests as "dangerous" on account of the presence of agitators stirring up resistance to the Mubarak government (Lindsey, 2012).The regime's ultimately futile effort was to persuade Egyptians to support the regime's initially repressive response to the protests.In Argentina, during the country's Dirty War from 1976 to 1983, the military dictatorship referred to the unarmed victims of state repression-of whom there were up to 30,000-as "subversive elements," "delinquents," and "criminals" (Feitlowitz, 2011).These labels were part of a regime strategy to persuade Argentines that repression was justified.Cohen (2011) provides an authoritative account of how labels which accuse their targets of criminality and delinquency shape public attitudes.Beginning with a judgment of those deemed to be "deviant," such labels are taken up by "moral entrepreneurs" on media platforms and come to acquire "descriptive and explanatory potential" about their targets (38).Leveraged in this way, accusatory labels link targets to broader societal problems and thus infuse the targets with additional negative connotations for the audience.What begin as ambiguous social situations become threatening, creating a division between the targets and those presented as "the real heroes"-police and security forces (108).The end product is "public support for the use of violence against [alleged] criminals," particularly violence from the police (182).Accusatory labels should function similarly in the context of protests in authoritarian regimes: increasing public support for the use of repression and decreasing sympathy with the protesters.Hou and Quek (2019) show that reports on protester violence may generate more support for repression of ethnic minority groups in China through the suggestions that protesters are criminals or terrorists.It implies that citizens may be persuaded that a strong state response is necessary when they are exposed to protest which appear to threaten the government.This produces the following observable implication: Hypothesis 1a: When a protest is labeled with accusations, citizen support for repression of the protest will increase compared to when the protest is labeled neutrally.
In contrast from accusatory labels which gin up public hostility, accommodative labels validate the underlying motive for the protest, even if statements fall short of making concessions or endorsing protesters' methods.While these words are typically not in themselves credible commitments to action when originating from the regime, they can mollify protesters and buy time for the government until concrete steps can be taken to remedy grievances.A famous historical example is King Richard II's response to the Peasants' Revolt of 1381.Confronted by rebels in London, the king expressed sympathy and promised to abolish serfdom.Satisfied, the rebels dispersed (Acemoglu & Robinson, 2006). 6More recently, in Egypt, after labeling protests as "dangerous" failed to prevent their escalation, dictator Hosni Mubarak held a press conference in which he announced that he was "attached to the suffering of the Egyptian people" and pledged to resolve the crisis (C-SPAN, 2011).
Accommodative labels trigger a reverse process from accusatory labels, defining protesters not as deviants but as having legitimate claims for material benefits or policy change.By thus minimizing the social distance between protesters and the audience of citizens, the citizens become more supportive of the protesters and less willing to support the use of state violence against them (Manekin & Mitts, 2020).The next implication follows: Hypothesis 1b: When a protest is labeled with accommodation, citizen support for repression of the protest will decrease compared to when the protest is labeled neutrally.

Regime Sources
Besides the content of the label, its source could also influence opinion.The government, especially in authoritarian regimes with state-dominated media sectors and with censorship of opposing viewpoints (Guriev and Treisman 2019), is a frequent source for statements about protest.Officials may hold a televised press conference in which they are visibly associated with a statement, they may visit the site of a protest, or issue a written statement in their name through state media outlets.This involvement usually benefits the authoritarian regime.Issuing official statements about events, particularly when coupled with empirical evidence, increases trust in the government and its official narrative (Huang 2015a(Huang , 2015b) ) while also raising the salience of policy issues important to the regime (Pan et al., 2021).Messages deployed through state media tend to cast the regime in the most favorable light possible (Rozenas & Stukal, 2019), while shifting observers' views in the regime's preferred direction-particularly on contentious political issues (Peisakhin & Rozenas, 2018).
The process of attributing response to protest-even at the local level-to the regime establishes an official narrative about the events and links them with the regime for the audience.Evoked by the regime's use of official statements is its added effectiveness in shaping public opinion about the underlying events when serving as the source of the statement.In contrast with statements issued by non-governmental sources about events, such as subjectmatter experts or commentators unaffiliated with state media, government sources by their regime connections are freighted with extra meaning, which may accentuate the effects of the messages' content.
If the regime is the source of an accusatory message about protest events, the regime associates itself with law and order by labeling protests as threatening social stability.In authoritarian regimes in particular, incumbents who challenge the legitimacy of protests remind the audience of the regime's authority and, especially, its power to repress.Citizens reminded of the regime's authority and coercive power are more likely to express public support for regime repression of protests (Truex & Tavana, 2019) while also becoming less willing to protest (Carter & Carter, 2021;Huang, 2018).These effects from the source complement and exceed the effects of a negative label, which, without a regime source, influences citizens' beliefs only about the protest rather than the regime.
Non-government sources, by contrast, are less likely to evoke the authority and coercion of the state with their use of labels.While citizens may believe, due to selection effects from a controlled information environment, that a nongovernment source is sympathetic with the government with respect to protests, this source still does not speak in an official capacity nor signal that a policy response such as repression could directly result from their statement.

Hypothesis 2a:
When the regime provides an accusatory message about protest response, the accusatory message increases citizen support for repression of the protest more than the case in which the message is provided by a non-government source.
If the regime is the source of an accommodative message about protest events, the regime associates itself with the possibility of concessions.Regime accommodation has the obverse effect as regime accusation.When the government takes such a "soft-line" response to protests, citizens-both regime supporters and opponents-interpret the label in a way which increases sympathy for the protesters.Among citizens who sincerely support the regime, receiving an accommodative message from the regime triggers motivated reasoning: they become more likely to follow the message provided them because of their prior loyalty.Among citizens who are indifferent or even opposed to the regime, an accommodative response to protest can signal the regime's permissiveness toward political challengers, increase these citizens' trust in the regime and make them more receptive to its message of protest accommodation (Frye & Borisova, 2019).
These effects work together to decrease support for a repressive response for the protest when the regime signals accommodation, more so than the moral and factual beliefs about a protest an accommodative label changes by itself: Hypothesis 2b: When the regime provides an accommodative message about protest response, the accommodative message decreases citizen support for repression of the protest more than the case in which the message is provided by a non-government source.
Note the combination of Hypothesis 1 with Hypothesis 2 has theoretical significance.For example, it is possible that giving a protest an accusatory label (H1a) affects support for repression, yet the source of that label (H2a) does not condition the effect of the label.If this is the case, then labeling effects dominate source effects.Whether the regime responds to the protest, or citizens receive information from any other type of source, is inconsequential.If the reverse is true and source effects dominate labeling effects, then the regime's strategy to issue responses becomes more important (Barberá & Zeitzoff, 2019;Steinhardt, 2017).If both sources and labels have effects, then this would suggest the two strategies are complements for regimes.

Research Design
We test our hypotheses using a survey experiment fielded in China and designed around a salient topic within Chinese politics: environmental issues.China is a prototypical case for studying protest labeling, with variation in protests, media coverage of those protests, and variation in labels used.By the 2010s, the number of local protests in China increased substantially (Zhang & Pan, 2019).Steinhardt (2017) shows that participation in protests in China expanded from under .4% of the population in 2002 to 2.3% in 2010.As the number of protests and protesters has increased significantly, this has reinforced the process of "discursive accommodation" in which protesters recognize constraints on dissent, seek to achieve only limited goals centered around specific local policy issues such as environmental degradation, and demobilize as soon as grievances are addressed (Lorentzen, 2013;O'Brien & Li, 2006;Weiss, 2014).
Protests in China meet wide variation in government responses.An accommodating response occurred when environmentalists in Sichuan Province won a victory after massive protests, both violent and non-violent, against a copper smelting complex.The US$1.6 billion project was permanently canceled in response to a long protest campaign (Bradsher, 2012).People's Daily, the Chinese central government's official media that "enjoys hegemony in shaping Chinese public opinion" (Wu, 1994, 195), reposted a commentary showing compassion for protesters and criticizing the lack of transparency and participation during the local policy-making process. 7In contrast, protesters against several destructor plants in Jiangxi Province were arrested and labeled as "collectively disturbing social order." 8Another prominent example of an accusatory response relates to the 2019 protests in Hong Kong, in which mostly non-violent protesters were labeled thugs, criminals and violent gangs by the central government (Myers and Mozur 2019).Interestingly, even within this escalating protest environment, the local police chief opted for an accomodative response in ordering that his police officers not label demonstrators as cockroaches, regardless of protesters' actions (Dixon and Kirkpatrick 2019).
Moreover, in China's vibrant, yet heavily monitored and regulated, information and media market local protests do receive wide coverage on national, local, and social media, in more traditional formats and through the Internet.Yet, the Chinese media market is often a platform from which the regime broadcasts its labels of those protest events, attempting to shape public opinion about salient political issues (Huang, 2018;Huang and Yeh 2018;Pan et al., 2021).

Protest and Labeling Patterns in China
We quantify these three phenomena about the case-protest, media coverage, and regime labels-by combining existing data on protest events in China with originally collected data on protest coverage in People's Daily.Protest data come from Zhang and Pan (2019) and contain 135,705 events from January 2011 to June 2017 identified with a deep learning approach from social media reports of offline protests.These data also contain a protest's tactics (i.e., violent and non-violent).The People's Daily data contain articles with one of the keywords "protest (抗议)," "mass incident (群体性事件)," "riot (聚众)" and "violence (暴力)," or "riot" and "damage (破坏)" regarding events in mainland China from January 2010 to December 2021.We filter out the articles about events that did not occur in China and read each of the remaining 76 articles to make sure they describe protest events. 9The following analyses explore patterns in protest labeling and suggest it is a strategic process.
Figure 1 depicts descriptive patterns in these data.The stacked bar chart, corresponding to the left-hand y-axis, depicts the monthly frequency of protest coverage for periods in the protest data: January 2011 to June 2017.Within each month of nonzero protest coverage by People's Daily, the label each article applied to protesters is color-coded.Accusatory labels were articles containing words like "violent (暴力)," "criminal (犯罪)," "terrorist (恐怖)," "vile (恶劣)," and "extreme (极端)."Accommodative labels showed understanding of protester demands and criticized public officials, while neutral labels are a residual category. 10Twenty-five percent of the articles were accusatory, 53% were accommodative, and 22% were neutral.Moreover, this coverage in the stacked bar chart represents only a fraction-though by far the most influential and representative fraction (Wu, 1994)-of state informational response to protests in China.The line graph, corresponding to the right-hand y-axis, depicts the monthly frequency of protests from the Zhang and Pan (2019) data.
We also identify several descriptive patterns in our data that point to regime labeling as a strategic response to protest.First, protest frequency is positively and significantly related with People's Daily coverage of protests in a given month-indicating that the Chinese regime attends to protests when they become more salient.Second, the frequency of violent protests is positively and significantly related to the frequency of People's Daily coverage using accusatory labels, suggesting accusation is deployed more when protests pose greater physical threat.Third, as a placebo test for the above, the frequency of violent protests is not significantly related to the frequency of People's Daily coverage using accommodative labels. 11Our survey experiment seeks to answer how this observed strategic calibration of government response to protests, specifically in the Chinese context, modifies public opinion on contentious events.

A Survey Experiment
We fielded our experiment in China to a sample of 2428 internet respondents from July 25 to August 5, 2021.Respondents were required to be adult citizens of the Chinese mainland recruited through the internet (mobile device or computer) without stratifying on a demographic group, although those who do not use the internet will be excluded by nature of the collection method. 12 We field our survey online not only because it is the best way to collect a large sample which avoids direct censorship and monitoring in China, but also because the internet has become a center of dissident and collective action (King et al., 2017). 13,14 As part of the survey, respondents answer a series of pretreatment questions.In this section, we collect data on potentially predictive covariates which include gender, age, province of residence, rural-versus-urban residence, marital status, education level, occupation, income, party membership, news consumption, exposure to protests, and several indicators of opinion on political and local issues such as the salience of environmental issues.The full list of pretreatment covariates on which data are collected can be found in the appendix.
Conducting experiments within an authoritarian context requires adjudicating considerable ethical concerns.Our treatment arms intentionally pertain to environmental issues because they are salient and directly impact citizens' lives but are not considered politically threatening to the central government.In posing little threat to the government, respondents are more likely to respond truthfully without fear of retaliation.We can reasonably expect, therefore, that survey responses have no impact on real-world outcomes for our respondents.Although the treatment vignettes in our survey are based on a real case and actual government responses, we also inform respondents at the end of the survey that the information they read in our treatment vignettes (described below) was fictional, to ensure they are left with no false impressions. 15

Treatment Arms
Following the pretreatment questions, respondents are exposed to a short message in the style of a news article describing an environmental protest event which occurred in Qidong, Jiangsu in July, 2012.The residents of Qidong protested against the city government for passing a waste-water pipeline project.They took to the streets and ultimately entered a local government building with both peaceful and violent confrontations with the police and the bureaucrats.The local government was forced to admit fault and scrap the project.The protest was widely reported in China and around the world.In China, opinions on the protest are divided: while some insist that the protesters were guilty of inciting violence, others believe the event was effective political participation which protected citizens' rights.Even People's Daily was ambivalent.It criticized some of the actions, but not the protesters themselves, and indicated that the reason for the protest was legitimate. 16 After reading the message, respondents then see a fictionalized statement of a response to the event.We induce random variation in the content and source of this response-making it the experimental treatment.The content of the response is a label: a descriptive, evocative, and short statement about the event by a specific source.We hold constant the details of the event itself outside the source's response.
The survey assigns respondents to each treatment group through simple random assignment.There are two main dimensions along which the treatment is randomized corresponding with our main hypotheses.The first dimension we randomize is the label given to the protest.Recall protest labeling-as a subtype of emphasis framing (Druckman, 2001;Iyengar, 2017)-is a set of descriptive and evocative terms an elite source uses to describe a protest.In the accusatory label treatment condition, the headline informs them of an official stating that "We need to stop the violence now and restore order."In the accompanying vignette respondents read, they are told that criminals were the ones who orchestrated the protest and they "violently stormed" the government building and "attacked government workers." In the accommodative label condition, respondents read a headline noting the official stating "we need to listen to our citizens, if common sense shall prevail."Survey participants are informed that people, a nondescript and generic label unlike "criminals," were the perpetrators of the protest.The actions are described as entering the government building regardless of the restriction and protest impulsively but not violently.It is important to note that, while the accusatory label includes mention of both criminality and violence, these terms only enter the vignette through the statements of the quoted source itself, rather than the background conditions for the event described in the news article.
We also create a control condition in which the protesters similarly "entered the building" and "confronted with" the government workers but without further descriptions.The headline of the control condition makes the official comment about the incident simply: "we need to pay attention." Given the nature of a contentious protest, each of these treatment arms are plausible descriptors of the same event and underlying facts.The difference lies in the labels used, the description of their actions, and the source's attitude corresponding to the descriptions.The control is designed to suggest neither criminality nor sympathy with respect to the protesters.We expect that an accusatory label increases support for repression and that an accommodative label decreases it, consistent with Hypothesis 1a and 1b.
Next is the source.We randomly vary who describes and labels the protest event.In the non-government condition, the source of the protest label is a scholar, whereas in the treatment condition the label comes from "the official media."In China, the term appears commonly in news articles and are widely interpreted as representing the regime.We do not specify the ideological position or the identity of the scholar because we would like to avoid priming survey participants on this information.The goal of the study is not to compare effects from two sources of different partisanship or ideologies.Rather, we would like to investigate how citizens respond to labels from sources based on their actual perceptions of a nongovernment actor who publishes comments on protests as they would in a real-world authoritarian context.In accordance with Hypothesis 2a and 2b, we expect this change in the source to lead to divergent responses in support for repression conditional on whether the government gives the protest an accusatory or accommodative label.
An example treatment vignette with the government source and different labels is shown in

Outcome Measures
Following exposure to the environmental protest vignette with a randomly varied label and source, survey respondents answer a series of questions related to the outcome measures of interest.Of primary importance, is support for repression, which we measure through both retrospective and prospective questions.First, we mention that the government carried out arrests and also concessions in response to the protest-in particular canceling the pipeline project-and gauge respondents' support for each measure retrospectively.Second, we probe respondents' support for protest in various forms.Specifically, we ask respondents if they would support a future protest similar to the one described in the vignette, including supporting it through posting on social media, signing letters, providing material support, joining the protest, or organizing the protest.The protest formats are adapted from the questions on the China Panel of the World Values Survey.
We next benchmark this support for repression against respondents' attitudes toward the protesters and the environmental issue motivating the protest as potential mechanisms explaining their support for repression: asking them their opinion about whether protesters were too violent, whether their actions are improper, and whether environmental pollution is a serious problem.

Estimation of Treatment Effects
The main quantity of interest we estimate in this experiment is the average treatment effect (ATE).This is estimated by taking the difference in means between a treatment condition and different reference categories.We then reestimate these quantities using regression estimates which include additional individual-level covariates.The regression estimation for Hypotheses 1 takes the following form where Label i is assignment status for either the Criminal label for respondent i, X i is a vector of pretreatment covariates, and ϵ i is a robust error term.The coefficient of interest is β, the effect of the label on repression support.The regression estimation for Hypothesis 2 takes the following form where Govt i is assignment status for the government source treatment.The coefficient of interest in this estimation is γ, the effect of the label-criminalon repression support among those respondents who received the randomly assigned government source treatment.Table 1 illustrates the comparison groups.Under Hypothesis 1, the effect of the accusatory label treatment compares the average support for repression among those assigned to the accusatory condition-cells 3 and 4-and those assigned to the control conditions-cells 1 and 2. We later discuss estimates comparing the accusatory label groups to the pooled control and accommodative groups.
The comparison of interest for Hypothesis 2 is the interaction of the accusatory label and government source, estimating the change in the labeling treatment effects among those assigned to a government source-the right column-and those assigned to a scholarly sourcethe left column.In particular, Hypothesis 2a predicts the effect of the government's accusatory label (cell 4 compared to cell 2) is expected to be larger than the effect of the scholar's accusatory label (cell 3 compared to cells 1).Hypothesis 2b predicts the effect of the government's accommodative label (cell 6 compared to cell 2) is expected to be larger than the effect of the scholar's accommodative label (cell 5 compared to cell 1).

Treatment Balance and Manipulation Check
We report results from balance tests in the appendix.Tables SI.5, SI.6, and SI. 7 contain difference in mean estimates across the three treatment conditions for each pretreatment covariates.Of the 54 individual difference in means tested, only two (income for the government source compared to the scholar source and marital status for the accommodative label compared to the control) are significantly different at the p = 0.05 level.This rate is what would be expected if the significant results arose by chance.We include income and marital status in all models with covariates, and this does not change the results.
To ensure that our results are not influenced by respondents' potential inattention, we ask an attention check question before introducing the treatment.Twelve percent of the participants fail the check.Neither conditioning on the failure nor interacting it with the treatments in the models changes our results.We also ask two post-treatment factual questions to check respondents' comprehension of the treatment.The two questions ask about the city where the protest happened and the source of the news report, and the failure rates are three percent and eight percent, respectively.

Labeling Effects
We first present results for Hypotheses 1a and 1b in Table 2, examining the effects of the accusatory and accommodative labels on respondents' expressed attitudes toward repression and willingness to participate in an environmental protest similar to the one described in the vignette.Each table presents estimates of average treatment effects from linear regression with and without individual-level covariates.
The results suggest consistent statistically significant results across all models for the accusatory label.An accusatory label causes an increase of .23-.27 in support for repression-that is, arresting protesters-on a five-point scale.Even-numbered models include covariates, while odd-numbered models withhold controls.Models 1-2 compare the accusatory label and control (or neutral) label, and Models 3-4 compare the accommodative label with the control label.Coefficients for the accusatory label are in the expected direction: support for arresting protesters increases across the different comparison groups (Hypothesis 1a).However, coefficients for the accommodative label (Hypothesis 1b) are not significant, suggesting respondents are no more supportive of protests when reading a label sympathizing with protesters than a neutral label conveying no sympathetic tone.Overall, the difference in outcome variables between the accommodative and neutral labels is minimal-preventing us from rejecting the null for Hypothesis 1b.

Effects of Labeling by Source
We present the results for labeling effects disaggregated by source in Tables 3  and 4.Here we test Hypotheses 2a and 2b, examining whether the source of a label differentially affects respondents' attitudes when interacted with the labeling treatment.Table 3 examines the same outcome as Table 2, and is structured similarly.Models 1-2 compare the accusatory and control conditions, and Models 3-4 compare the accommodative label treatment with the control condition.Odd-numbered models include no controls, and evennumbered models include them.As described in Table 3, we find consistent effects of accusatory labels on support for repression (arresting protesters) but no evidence for a source effect.Whether labels are issued by the regime, with the power of the state behind them, or by a scholar unaffiliated with the regime, the effects remain statistically and substantively similar.In this test, the accommodative label remains similarly ineffective in shaping respondents' support for arrest, regardless of the source deploying the label.Because the accommodative label's effects are indistinguishable from the neutral label, in the tests that follow we focus on probing additional outcomes and mechanisms to substantiate the findings for the accusatory label.
In Table 4, we also examine several additional auxiliary outcome variables related to protest support for the accusatory label which produced significant effects on support for repression in the main analysis.The outcome variables below refer to the question, asked post-treatment: "If there is a similar protest against pollution or other public welfare issues, are you going to support it, and how?"The options are adapted from World Values Survey, including (1) not support in any form, (2) post support on social media, (3) sign name on letters for support, (4) provide material support, (5) participate in the protest actions, and ( 6) organize or lead the protest.We measure responses with each type of support, coding 1 if the corresponding option is selected and 0 otherwise.We create three binary variables indicating respondents' selection of the first three options, no support, social media post, and signing letters, as these options are low cost and thus more feasible for most respondents. 17 Figure SI.2 in the appendix presents the distribution for each option across treatment conditions.An accusatory label is associated with a .1-unitincrease in the probability of not supporting future protests or actions, and in some models the label also decreases willingness to write social media posts and sign letters supporting the protest.As with previous results, we see no source effect, as the source interaction term is statistically and substantively insignificant. 18In the following section, we discuss possible mechanisms for these findings.

Mechanisms: What Perceptions Do the Label and Source Change?
To determine why accusatory labels shape support for repression and opposition to protests, we examine which of respondents' perceptions about protesters the labels affect.A story about protest is complex, and the labels may trigger various changes in how respondents perceive the protest facts in the vignette.The most direct mechanism in our theory is that accusatory labels elicit negative views on protesters and their behaviors' legitimacy.Another, related mechanism is that an accusatory label redirects respondents' attention, taking focus away from the underlying cause of the protest (Munger et al., 2019)-an environmentally harmful project-and redirecting it to protester conduct.Through this mechanism, an accusatory label would improve perceptions of government officials who are perceived to be less culpable for the social instability.Moreover, by labeling a protest as committed by "criminals," the accusatory labels challenge the legitimacy of the protesters' demands, which, in this case, are to cancel the project.
Our findings affirm the mechanism of the accusatory label changing respondents' perceptions of protesters.We ask the survey participants to report their agreement with three statements: "the protesters' behaviors were violent," "the protesters' behaviors were improper," and "the protesters' behaviors deserved support."Table 5 shows that the accusatory label affects responses to all three questions regardless of the source.It makes citizens more likely to view the protesters as violent, that their actions were improper, and that they do not deserve support.
However, we do not find evidence that the accusatory label shifted attention away from culpable government officials or the controversial project.
To test this, we create outcome questions based on the factual conclusion of the Qidong Protest-the project was canceled and none of the officials in charge of introducing or allowing the project were sanctioned.We ask the respondents how much they agree that the local officials responsible for the project should pardoned and that the project should be canceled, each on a five-point scale.Table 6 shows no clear treatment effect on respondents' support for pardoning officials or canceling the pipeline project, either as a baseline effect or interaction with source.The results suggest that the accusatory labels inflame negative perceptions of the protesters but do not change the blame assigned to the local government for environmental damage.

Accounting for No Source Effects
We also explore three explanations for the lack of source effects in our main results: the perceived credibility of sources according to survey respondents, the distinction between government intimidation and persuasion, and respondents' understanding of scholars as non-governmental sources in China.First, if source effects do not change respondents' perceptions and judgments of protests, respondents should perceive the two sources' credibility similarly.Studies on fake news and rumors have found that the content, the source, or the interaction of the two may significantly influence people's trust in information they consume (Bai et al., 2015;Berinsky, 2017;Huang 2015aHuang , 2015b;;Nyhan et al., 2020;Zhu et al., 2013).The null findings for the distinction between the government and scholar sources on support for repression could occur because people find the government's accusatory labels less credible than the scholar's for their more obvious bias, but are also intimidated by government accusatory labels-creating offsetting effects.Source credibility also suggests an alternative explanation for our labeling findings: that respondents may perceive an accusatory label as more credible because in the authoritarian context protesters are frequently accused.Thus respondents reading the accusatory labels may not be more likely to support repression, but they are simply more responsive to the message because they tend to trust the content more.
To address the possibility of varying source credibility, we first maximize credibility across treatment arms by drawing on real news reports from official media and mainstream Chinese news platforms.Adopting similar expressions as those in real news about a protest event is more credible than using fictitious wording.At the end of the survey, we ask respondents to rate the credibility of the news content on a five-point scale, and find their average rating of the government (3.38) and scholar (3.34) sources is in the top half of the scale.Second, Figure 2 shows that government and scholarly sources have similar effects on credibility across both accusatory and accommodative labels.These results suggest respondents did not perceive the sources differently, weakening the possibility the government source is seen as more or less biased.See Table SI.8 and Table SI.9 in the appendix for full results. 19 Part of our expectation for government source effects is that an accusatory official comment on a protest event carries the connotation of state authority or coercion, intimidating respondents and increasing conformity with the official position (Trinh & Truong, 2020). 20This is an added effect over nongovernment scholars, who can only inform and persuade.We evaluate the possibility of government source intimidation effects through two analyses.First, as mentioned above, an intimidating accusatory label by the government would signal coercion and generate compliance above and beyond a scholar's accusatory label.This is not the case, as Table 3 shows no interaction effect for a government source and accusatory label on support for arrest.Second, if the government source works through intimidation, we would expect respondents to default to the self-evident official position on protests rather than a persuasion effect, which would cause them to weigh and evaluate new information before arriving at an opinion.Intimidation would therefore reduce survey response times compared to persuasion. 21Tables SI.11 and SI.12 show no change in survey duration for either government sources or accusatory labels, lending support to the notion that both government and scholarly sources' labels persuade rather than intimidate respondents.Another possibility is that a scholar's use of labels could be seen as indistinguishable from the government's in an authoritarian context like China due to censorship of both media and scholars.We acknowledge that this is entirely possible, but argue that scholars and intellectuals are not necessarily seen as pro-government in the Chinese context.Their criticisms of policies and local governments are to some extent allowed by the regime.Of course, pro-government scholars who endorse repression might be more likely to appear on media than those who support protesters, but criticisms from scholars are not rare.In fact, in the mainstream news media, some scholars published comments on the Qidong protest (on which we base our survey vignettes) that supported citizens' fight for their rights while calling for government responsiveness. 22 In the eyes of Chinese citizens, while some scholars actively support the regime, many others are seen as liberal, dissident, and even pro-Western.In expressing these views, scholars who deviate from a government position are frequently criticized on Chinese social media.It is exactly in this highly censored media environment that citizens observe non-government sources that are labeled as anti-regime. 23Again, our study examines whether labeling protesters has differing effects across government and non-government sources.A non-government source, pro-government or not, does not alter the significant effects of the accusatory labels on citizens' support for protests or repression.

Discussion: Source Effects in Democracies versus Single-Party Regimes
The robust null source effects that we have demonstrated between the scholar and government source leave us with a puzzle: why is a regime source-with its additional ability to signal coercion or permissiveness toward protest (Carter & Carter, 2021;Huang, 2018;Pan et al., 2021)-no different from a scholar in shaping attitudes with protest labels?Research on political communication, particularly in the United States, has shown that the same message can have divergent effects depending on whether a particular political actor or an independent, nonpartisan expert delivers it.We argue that our findings deviate from this literature because the institutions and constraints on expression in an authoritarian context change citizens' perceptions of sources.
In liberal democracies, citizens are exposed to a wide range of contradictory messages on a given issue.Groups and institutions such as social movements (Benford & Snow, 2000) and political parties (Iyengar and Valentino 2000) lay claim to issues and deploy certain frames to characterize those issues and related political events.This exposure to competing messages by different sources places conditions on the credibility and therefore effectiveness of these sources.Parties' messages lose effectiveness if the content diverges from voters' "ingrained expectations" about the issues and frames those parties own (Iyengar and Valentino 2000).Partisan messages also lose effectiveness when juxtaposed with competing frames: voters tend toward greater reflection and anchoring in their own prior beliefs rather than adopting a source's view in these situations (Sniderman and Theriault 2004).
Another characteristic of political communication in democracies is frames' origination in the media.Cohen (2011) describes this process with the Mods and Rockers panic in the United Kingdom: organic media coverage increased public support for state coercion against perceived deviants.Iyengar (2017) makes a broader point: independent media coverage shapes public opinion in democracies, which then gives election-motivated officials an incentive to lay claim to issues and frames that resonate with voters.The combination of voter exposure to competing partisan frames and the origination of frames in independent, politically unconnected media means that partisan figures are less effective sources than nonpartisan officials and academic experts (McGuire, 1985).Druckman (2001), for example, finds the most credible and persuasive source in an US study to be a career military officer.
As we outlined above, the communication environment in authoritarian regimes with limited press freedom differs significantly from that in democracies.In China, the protest coverage the state permits is curtailed.State media sees its role as managing public opinion with carefully calibrated and consistent responses to specific events such as protests (Pan et al., 2021;Steinhardt, 2017).The state does this with greater agenda-setting power than in democracies, working in concert with journalists to stake out the boundaries of coverage on an issue (Repnikova, 2017).The public, in turn, sees the state as "[affirming] the social order and society's shared moral values" (Tsai, 2021, 39) when it issues a response to protests against local officials.State sources therefore should not suffer from the same credibility deficit as discordant partisan sources in democracies relative to nonpartisan and expert sources.Moreover, the limited range of permitted viewpoints in China creates selection effects in the independent sources citizens see.The public may draw inferences about these sources' characteristics given their being uncensored-in particular, the source may be more credible because the regime does not regard it as potentially destabilizing (Repnikova, 2017).As a result, independent sources such as scholars may get a boost in credibility compared to the counterfactual of no limits on the range of permitted viewpoints.

Conclusion
Our study shows that information-based responses to protest affect public attitudes about repression and mass mobilization in an authoritarian context.
Using a survey experiment in China, we find that accusatory labels generate support for repression and decrease support for protest even when the underlying factual basis of the protest's claims and participants is held constant.Through testing possible mechanisms, we find evidence that respondents change their attitudes toward the protesters but not toward the government or the issue around which the protest occurred.
We also find that it is the message, rather than the messenger which shapes respondents' attitudes toward repression and protest: a scholarly source produces the same effects when deploying an accusatory or accommodative label as does a government source.This finding contrasts with past work that authoritarian regimes use "hard propaganda" to attempt to signal their power or preferences and prevent citizens from dissenting (Huang 2015a(Huang , 2015b;;Wedeen, 1998).Our tests support the argument that, in response to protest, the main pathway through which pro-regime messages work is likely persuasion-public support for repression and protest only changes with the labels' content and aligns with respondents' perception of source credibility.Research on fake news, rumors, and fact-checking across regimes suggest that source credibility matters for belief change (Berinsky, 2017;Huang 2015b;Nyhan et al., 2020;Zhu et al., 2013), consistent with our findings that (1) respondents rate government and scholar credibility both highly and at similar levels, and (2) that labels from both sources affect beliefs.However, we do not rule out pathways besides persuasion from credible sources as possible explanations for belief change.Whether labeling works through persuasion, intimidation, distraction, or raising salience could depend on whether the issue at stake triggers differing perceptions of credibility between government and non-government sources.
An important scope condition is the difference in effects when there are visual anchors accompanying protest-related labels.In early 2020, we administered a similar survey experiment using visual anchors that did not find a consistent labeling effect.We included in all labels respondents received the same, real picture of the Qidong protest incident in which a group of protesters overturned and surrounded a police car.Therefore, it is likely that giving respondents a powerful visual anchor of protesters committing property damage washes out the effect of varying text-based labels.Given the prevalence of visual messages on social media today, the difference between these forms can be an interesting path for future research on propaganda in response to protests.In particular, future research could vary the images of protests to which respondents are exposed while holding text-based information constant to determine how this might affect attitudes toward repression and protest.
Finally, to minimize the potential harm to respondents and the fear of expressing truthful opinions regarding protest and repression, we set our experiment on environmental issues, which are less sensitive than more politicized issues such as demands for democratization.It is possible that the government source matters more in that context because citizens may be more likely to believe that the government will sanction its opponents.Nevertheless, our study finds that citizens are susceptible to negative labels of protesters, which is not due to the intentional support of the government's position.This implies that at least the message itself can be a useful strategy for the government to cultivate support for repression regardless there is a signal of coercion or not.20.A similar dynamic signaling regime permissiveness affects accommodative official comments.21.This comports with findings that the threat of repression reduces citizens' skepticism and use of more cognitively demanding System II thinking (Horz, 2018).22.For example, see a column at Sohu.com: http://star.news.sohu.com/s2012/mjzl/23.For instance, people are easily accused as "hostile forces" and even bullied by Chinese netizens even if they post negative opinions about the government or the country on not very sensitive topics (China Media Project, 2021).
Figure SI.1.All respondents are shown the introductory paragraph.Treatment manipulations are in the second paragraph and the headline.A full list of vignettes is included in the appendix.

Figure 2 .
Figure 2. Treatment effect on perceived content credibility by source and label.

Table 1 .
Full Enumeration of Treatment Arms.

Table 3 .
Main Results: Label Effects by Official Source.

Table 4 .
Results: Forms of Protest Support.

Table 5 .
Accusatory Labels and the Perception of Protesters.

Table 6 .
Accusatory Label on the Perception of Government and Project.