After Michael Brown, an unarmed Black teenager, was shot by a police officer in Ferguson, Missouri, in August 2014, protests erupted there and soon spread across the nation. These protests occurred under the banner of Black Lives Matter (BLM), a loosely connected social movement focused on a mix of national and local political issues (Foran Reference Foran2015; Lowery Reference Lowery2016). The movement returned to the national headlines after the murder of George Floyd in May 2020, which sparked additional waves of protests that continued through 2021 (Buchanan, Bui, and Patel Reference Buchanan, Bui and Patel2020).
Protesters across the United States encountered widely divergent responses from the police. In December 2014, for instance, San Francisco Magazine highlighted a “tale of two pictures”: in Richmond, California, police chief Richard Magnus joined demonstrators and held a “#BLACKLIVESMATTER” sign, whereas several days later in Oakland, an undercover officer pointed a gun directly at protesters (Lucas Reference Lucas2014). Protests in Buffalo, New York, in June 2020 tell a similar story: Some police officers knelt with demonstrators, while reports showed police pushing an elderly protester to the ground the very next day (Chen Reference Chen2020; Treisman Reference Treisman2020).
A closer look at the data suggests that this variation was not limited to high-profile examples. In the first section of this article, we present a newly collected dataset of police behavior at more than one thousand BLM protests between 2014 and 2017. This dataset provides new descriptive evidence that police responses to these protests varied widely, even when comparing protests with similar messages and tactics.
In the second part of the article, we investigate the implications of this variation in protest policing. Specifically, we draw on research on the subjectiveness of perceptions of violence (Edwards and Arnon Reference Edwards and Arnon2019; Manekin and Mitts Reference Manekin and Mitts2021) and the success of nonviolent resistance (Chenoweth and Cunningham Reference Chenoweth and Cunningham2013; Chenoweth and Stephan Reference Chenoweth and Stephan2014) to examine whether protest policing can create perceptions of violence and depress public support for BLM protests. We test this argument with a survey experiment conducted in April 2022. The experiment presented respondents with a news article about a BLM protest and manipulated the description of the police response. We find that when news of a protest is accompanied by a photo of a large and armed police deployment, readers are more likely to describe the protest as violent and to ascribe violent and trouble-making intentions to protest participants. We also find that exposure to a single police response is insufficient to shift broader support for the protest movement. These findings underscore how protest policing decisions could contribute to and shape perceptions of protest violence. In an exploratory investigation of racial heterogeneity, we also find that white respondents are more responsive to images of a heavy police presence at a protest than nonwhite respondents and that Black respondents appear nonresponsive to the heavy police presence cue. These results build on research that finds racial differences in perceptions of the police (Jefferson, Neuner, and Pasek Reference Jefferson, Neuner and Pasek2021; Tuch and Weitzer Reference Tuch and Weitzer1997) and of public opinion after protests (Carey and Cisneros Reference Carey and Cisneros2023; Davenport, McDermott, and Armstrong Reference Davenport, McDermott, Armstrong, Bonner, Kubal, Kempa and Seri2018; Enos, Kaufman, and Sands Reference Enos, Kaufman and Sands2019).
These descriptive and experimental findings make several contributions to literatures on social movements, policing, and BLM. First, we extend a growing literature on the mutability of perceptions of violence. This literature demonstrates how the mass public’s view of protests depends on the partisan, ethnic, and racial identities of protesters (Edwards and Arnon Reference Edwards and Arnon2019; Hsiao and Radnitz Reference Hsiao and Radnitz2021; Manekin and Mitts Reference Manekin and Mitts2021; Peay and Camarillo Reference Peay and Camarillo2021) and on media coverage and framing (Arora, Phoenix, and Delshad Reference Arora, Phoenix and Delshad2019; Kilgo and Harlow Reference Kilgo and Harlow2019; Phelps and Hamilton Reference Phelps and Hamilton2021). We extend this literature by showing how the police can play a similar role: Protest policing provides cues about the nature of the protest and may thus lead observers to infer information about the violence and intentions of protesters.
Second, our findings contribute to research that finds that nonviolence is more effective than violence in achieving social and political change (Chenoweth and Stephan Reference Chenoweth and Stephan2011; Nepstad Reference Nepstad2015). Explanations for this finding emphasize that nonviolence lowers barriers to participation and attracts more domestic and international support (Chenoweth and Stephan Reference Chenoweth and Stephan2011; Reference Chenoweth and Stephan2014). Our findings suggest that the police play a role in creating perceptions of violence and thus can directly affect movement success in this way. Although we do not find that exposure to a heavy police response directly decreases support for the broader movement, our findings do suggest that repeated exposure to heavy police responses could reduce movement support if increased perceptions of protester violence produce more general and negative beliefs about the broader protest movement. This finding runs up against the argument that police violence generates public sympathy and support for nonviolent protesters (Wasow Reference Wasow2020) and suggests that armed police deployments can undercut this response if their presence creates the perception of protester violence. More broadly, this finding underscores the point that public perceptions of violence can differ from the actual levels of violence. If the mechanism between nonviolence and movement success runs through public opinion, scholars (and social movement actors) may want to consider that perceptions of “violence” or “nonviolence” can differ from actual violence and, specifically, that perceptions of violence can exist in the absence of violent tactics.
Third, our findings contribute to research on policing by suggesting that police are important strategic actors in the construction of narratives about protests and social movements. Law enforcement organizations frequently enter the realm of politics, even though they often avoid characterizing their actions as political (Huey and Hryniewicz Reference Huey and Hryniewicz2012). These organizations take explicit political action like donating to politicians or lobbying on specific bills, as well as less overt actions, such as press releases or videos with positive portrayals of officers (Cheng Reference Cheng2021; O’Connor Reference O’Connor2022; Page Reference Page2011; Sieg and Wang Reference Sieg and Wang2013). We argue that protest policing should be viewed as another form of political action, one with repercussions for public opinion and social movements. This finding has more than theoretical relevance: Political actors seeking to protect social movements’ political rights may want to incorporate these framing effects into their decisions about protest policing. In recent years, elected officials at the state and city levels have used both legislation and litigation to introduce restrictions on how police can respond to protest actions (see, for example, the recent court settlement restricting “kettling” by the NYPD, requiring a graduated protest response, and imposing an official to oversee protest-policing decisions (Capps Reference Capps2014)). These findings illustrate the importance of such actions not only in protecting individual protesters from mistreatment but also in limiting the potential for strategic police action to shape public opinion of protest movements.
Finally, we contribute substantively and empirically to the study of BLM. Although most recent empirical research has focused on the dynamics and consequences of protests (Drakulich and Denver Reference Drakulich and Denver2022; Reny and Newman Reference Reny and Newman2021; Williamson, Trump, and Einstein Reference Williamson, Trump and Einstein2018), we emphasize instead the enormous variation in protest policing. On a substantive front, this leads us to develop a new police-driven explanation for public perceptions of protests. Empirically, we present a new dataset of BLM protests between 2014 and 2017, which provides details about more than 1,000 protests (timing, size, tactics), alongside measures of police reactions to these protests. These data provide more fine-grained and geographically complete information about BLM protests than is currently available. The dataset is available with the replication materials, and we hope it will serve as a resource for future empirical research on BLM.
Descriptive Data on BLM Protests and Policing
We motivate this project with a descriptive look at BLM protests from mid-2014 to early 2017. Our dataset provides a detailed look at what happened at more than one thousand street protests during the first few years of large-scale BLM protest mobilization by bringing together protest details (timing, size, tactics, etc.) with measures of police reactions to those protests. Collecting these various features required time-consuming examination of many news and social media sources, but it allows for a more comprehensive look at protest policing than a single source could have provided. This rich and multifaceted dataset lets us examine variation in protest policing within one key social movement, noting where otherwise similar protests—affiliated with the same movement, responding to the same events, and often occurring on the same dates—faced drastically different police responses. In addition to enabling us to document the range of police responses to the BLM movement across the United States, we hope this dataset will be useful to researchers seeking to examine other features of BLM protests, perhaps in conjunction with more recent protest datasets collected by the Crowd Counting Consortium (2023) and others.
In constructing the dataset used for this analysis, we started from a database of protests compiled by Alisa Robinson (Reference Robinson2017) via her own research and crowdsourcing. This database includes protests covered in both national and local news outlets across the country, as well as some protests documented only by social media posts by participants; it thus captured events that would be missed by searching a single newspaper or database. We then trimmed this database to focus on public street protests (omitting actions like NFL players taking a knee) and edited it, correcting placenames and dates with typographical errors and expanding brief multi-city entries into multiple rows of the dataset. We next undertook a process of investigating each protest through searches for relevant newspaper articles and manually coding various protest characteristics, including measures of police action (police presence, arrests, crowd-control measures) and other background characteristics of the protest (whether a protest occurred after dark, whether it was organized by clergy, whether it involved a highway shutdown or other traffic disruption, etc.). Section 1 of the Supplemental Information (SI) provides more details on the coding process and our coding instructions, including detailed definitions of all the variables we coded about each protest.
For more than 90% of the BLM protests that occurred across the United States between July 2014 and March 2017, we were able to code the approximate size of the protest and details about police presence; for most protests we were also able to record details about protest tactics, such as whether the protest took place at night and whether it involved a highway shutdown. This new dataset provides a valuable window into the BLM protests and the breadth of police responses.
Figure 1 maps the protests in the dataset and gives a sense of how widespread BLM protests were during this period. Protests occurred in 45 states and in Washington, DC, with many metropolitan areas seeing repeated protests. These protests were a nearby occurrence for many people: In the second half of 2014 alone, more than one-third of the population lived in a county that had a protest, and more than 70% had a protest in their own or a neighboring county.
Protest Locations in the Contiguous United States
Note: Point sizes are scaled by approximate protest size.
Variation in BLM Protest Policing
These widespread protests provided ample opportunity for Americans to develop opinions about the Black Lives Matter movement. Importantly, the movement also provided the public with a range of protest images. On August 14, 2014, for example, 88 cities held vigils to observe the “National Moment of Silence for Victims of Police Brutality.” In most cases, people marched or stood quietly with little to no visible police presence. In other cases, however, the police engaged in a more visible and forceful response. After the vigil in Minneapolis, for instance, a woman was arrested and placed in a squad car for a supposed violation of traffic laws, prompting a demonstration at the police station (Collins Reference Collins2014). In New York, protesters used social media to warn each other about police using “kettling” tactics on demonstrators in enclosed areas (Capps Reference Capps2014).
Subsequent waves of protest paint a similar picture: despite reacting to the same events with the same message and tactics, protesters across the country received varied responses from the police. On average, two-thirds of protest events in our dataset had some police presence, with 17% seeing at least one arrest and 14% seeing police deploying some sort of crowd-control measures.Footnote 1 Even within the same type of event, such as the 2014 National Moment of Silence or the nationwide protests following the non-indictment of Darren Wilson for killing Michael Brown, there was broad variation in whether police appeared at protests and whether they took further actions.
What explains this variation in protest policing? A rich literature in sociology, criminology, and political science describes the determinants of police response to protest action, and our analysis of this BLM dataset includes variables intended to capture some of those known predictors. Existing work would predict both heavy policing of system-challenging protests such as BLM demonstrations, with particularly harsh treatment of unruly protests or those using extreme tactics; larger protests and those with more Black participants would attract heavy policing as well (Davenport, Soule, and Armstrong Reference Davenport, Soule and Armstrong2011; Earl and Soule Reference Earl and Soule2006; Reynolds-Stenson Reference Reynolds-Stenson2018; Soule and Davenport Reference Soule and Davenport2009).
Table 1 uses the protest dataset to examine variation in police behavior. Specifically, we explored whether protest characteristics predict police presence at a protest (columns 1–2), arrests at a protest (columns 3–4), and the use of crowd-control tactics (columns 5–6).Footnote 2 Observation counts varied across specifications because some protest characteristics (notably protester race) and outcome variables were missing. Yet, these data show that some protest features were associated with police responses in ways consistent with past research; for instance, when protesters blocked highways or had protests in the evening, protests were more likely to be met with police presence and arrests. Similarly, protests with some “other disruption,” such as protesters blocking local traffic or chaining themselves to objects, were more likely to prompt police presence, arrests, and crowd-control tactics. And larger protests of one thousand or more people were more likely to see police presence than smaller gatherings.Footnote 3
Protest Characteristics and Police Response
Note: *p<0.05.
However, the protest characteristics in Table 1 explain only a fraction of the variance in protest policing. In none of these models does the
$ {r}^2 $
measure exceed 0.3. Even when we used state or county fixed effects to compare police departments in the same area, less than half of this variation in police responses is explained by protest tactics (tables shown in SI Section 2). This pattern leaves room for police departments to exercise discretion in their responses to BLM protests, and such discretion can yield different protest experiences in different places. In addition to shaping the experiences of people who attended these protests who may have faced arrest or worse depending on police decisions, we hypothesize that harsher police responses can shape public perceptions of the protest and of the broader movement it represents.
We should note, here, that there are several aspects of the interaction between protests and police responses that we cannot analyze with these data. For instance, given the nature of the data, we treated protest policing as an outcome that may be affected by observable characteristics of the protest but is otherwise independent of how the protest unfolds. However, protesters and police may influence each other within a single protest event: for example, the presence of police may spark a disturbance among protesters, which may lead to an arrest, which may spark more disturbance, which may lead to crowd-control measures.Footnote 4 Although we cannot account for these kinds of dynamic interactions with this dataset, we focus here on what the data do reveal: that there exists meaningful variation in the observable characteristics of protest policing responses.Footnote 5 We conducted a survey experiment in an attempt to isolate the implications of this variation for observers.
Protests, Policing, and Public Opinion
In the previous section, our analysis of BLM protests from 2014 to 2017 suggested that there was room for discretion in police departments’ responses to protest. Data from more recent waves of protests paint a similar picture: Despite the fact that BLM demonstrations were overwhelmingly peaceful, more than 9% were met with government intervention in 2020 (ACLED 2020). Moreover, these data show that the police used force-like tear gas, rubber bullets, pepper spray, and batons in more than half the protests in which they intervened. Overall, then, these data indicate that police departments used their discretion to take a disproportionately heavy-handed approach to the protests.
What are the implications of this use of discretion? Beyond their effect on protesters and protest dynamics (Davenport Reference Davenport2007; Siegel Reference Siegel2011; Steinert-Threlkeld, Chan, and Joo Reference Steinert-Threlkeld, Chan and Joo2022; Young Reference Young2019), we suggest, in what follows, that police responses can shape mass public opinion on the protests and the broader BLM movement. Specifically, we propose that large deployments or heavily armed police can create the perception of protest violence. Importantly, this test coheres with a key feature of the polling data on BLM; namely, the fact that BLM protesters were disproportionately viewed as violent, despite the overwhelmingly peaceful nature of the demonstrations (Chenoweth and Pressman Reference Chenoweth and Pressman2020; Skelley Reference Skelley2020).
Theoretically, this argument builds on recent work on the mutability of perceptions of protests. One strand of this literature highlights how public perceptions of protests depend on the partisan, ethnic, and racial identities of protesters. Manekin and Mitts (Reference Manekin and Mitts2021), for instance, show that nonviolent resistance by ethnic minorities is perceived as more violent than identical resistance by ethnic majorities. Peay and Camarillo (Reference Peay and Camarillo2021) find that protests with all Black participants are perceived to be more likely to end in violence than more diverse demonstrations. Similarly, Hsiao and Radnitz (Reference Hsiao and Radnitz2021) show that people perceive higher levels of violence in protests by partisan outgroups.
A second strand of this literature demonstrates the role of media framing in shaping elite and mass attitudes about protests (Arora, Phoenix, and Delshad Reference Arora, Phoenix and Delshad2019). In a study of BLM protests, for example, Kilgo and Mourão (Reference Kilgo and Mourão2021) find that media frames that emphasize disruptive or violent protest incidents can have damaging effects on perceptions of the movement. Edwards and Arnon (Reference Edwards and Arnon2019) show that public perceptions of whether a protest is violent shift based on the framing of the types of protest action and the identities of participants. Wasow (Reference Wasow2020) similarly finds that positive mainstream media coverage of civil rights protesters in the 1960s tilted public opinion and policy making in their favor.
We argue that the presence and behavior of police can play a similar role by providing cues about the nature of the protest. Specifically, we propose that news coverage of protests with heavy police deployments and tactics (riot gear, tanks, tear gas, etc.) may lead observers to infer that the protest is violent. This argument generates the following hypothesis:
H1: Protests met with a heavy police presence are more likely to be perceived as violent than identical protests without a heavy police presence.
We also expect protest policing to affect public support for the broader protest movement. This argument builds on an extensive literature suggesting that nonviolent campaigns are more likely to achieve their goals than violent campaigns (Chenoweth and Stephan Reference Chenoweth and Stephan2011). The proposed explanation for the effectiveness of nonviolent resistance centers on the appeal of nonviolence: Nonviolent campaigns are thought to reduce barriers to participation and attract more domestic and international support than their violent counterparts (Chenoweth and Cunningham Reference Chenoweth and Cunningham2013; Chenoweth and Stephan Reference Chenoweth and Stephan2011; Nepstad Reference Nepstad2015).
Recent experimental evidence supports this mechanism by showing that violence reduces public support for protesters and social movements (Feinberg, Willer, and Kovacheff Reference Feinberg, Willer and Kovacheff2020; Huff and Kruszewska Reference Huff and Kruszewska2016; Muñoz and Anduiza Reference Muñoz and Anduiza2019; Orazani and Leidner Reference Orazani and Leidner2019). Dahlum, Pinckney, and Wig (Reference Dahlum, Pinckney and Wig2022), for instance, conducted a survey experiment in 33 countries and find that nonviolent tactics strongly increase movement support relative to violent tactics. Arves, Cunningham, and McCulloch (Reference Arves, Cunningham and McCulloch2019) examine the impact of rebel behaviors on American public opinion and find that the use of terrorism decreases public support. Simpson, Willer, and Feinberg (Reference Simpson, Willer and Feinberg2018) find that the use of violence in antiracist protests against white nationalists reduces public support for protesters. This experimental literature, then, provides compelling support for the mobilizing effects and success of nonviolent movements.
Building on this literature and our first hypothesis, we argue that protest policing may have far-reaching effects on the movement’s ability to mobilize support. Specifically, we propose that if police presence can create the perception of violence, so too may it undermine public support for the movement. We specify this second hypothesis as follows:
H2: Protests met with a heavy police presence are more likely to reduce public support for the broader protest movement than identical protests without a heavy police presence.
This hypothesis qualifies the argument that repression generates sympathy and support for nonviolent protesters (Wasow Reference Wasow2020) and suggests that heavily armed or large police deployments may undercut this sympathetic response from the public if they create the perception of protest violence. In line with our proposed mechanism, polling data show that public support for BLM peaked when the 2020 protests were first reported and declined sharply after (Civiqs 2023).
Survey Experiment
Study Design
To test our hypotheses, we conducted an online survey experiment in April 2022. We recruited a diverse sample of 2,710 US-based respondents through Prolific and presented them with a realistic news story about a BLM protest. In the news story, we randomly varied the police response to a BLM protest while holding constant all other information about the demonstration (figure 2). Some respondents were presented with a news story describing a march and an image of nonviolent protesters (figure 3), whereas other respondents received the news story, the same image of nonviolent protesters, and an additional sentence about and image of armed police standing behind barricades at the protest (figure 4). The italicized text in figure 2 indicates the heavy police presence condition. Respondents were randomly assigned to one of these conditions with equal probability.
News Article Vignette
Control Condition Image
Treatment Condition Images
We did not explicitly describe the protest as peaceful because news stories do not often report that a protest was unequivocally peaceful.Footnote 6 Instead, we provided a vignette without images or descriptions of specific protest violence. This approach has some limitations: by not affirmatively stating that a protest was nonviolent, we leave room for different observers to update their beliefs in different ways about the nature of the protest.Footnote 7 But the treatment resembles a common kind of protest news story in the real world: accounts of non-extreme interactions between protesters and police that leave some room for interpretation by observers. This approach also resembles recent research in political science that uses newspaper articles about protests to experimentally examine public reactions to protests (Manekin and Mitts Reference Manekin and Mitts2021; Naunov Reference Naunov2024).
We then asked outcome questions relating to our two hypotheses, perceptions of protester violence (H1) and support for BLM protests (H2). We measured perceptions of violence by asking respondents whether they agreed or disagreed with three statements on a five-point Likert scale: “The event in question was violent,” “The protesters had violent intentions,” and “These protesters were out to cause trouble.” We measured support for BLM protests (as portrayed in the vignette) with four measures. First, we asked respondents whether they agreed or disagreed with the statements (1) “I would consider getting involved with a group who supported causes similar to those of the protesters” and (2) “I support these protesters” on a five-point Likert scale. Then, we asked respondents to indicate their willingness to (3) “Go to a protest like this one” and (4) “Post something positive about a protest like this on social media” on a scale from 1–100, where 0 means that a respondent would “absolutely not take that action” and 100 means that a respondent would “definitely take that action.”
Experimental Results
We present the estimated effects of our “heavy police response” treatment condition on two sets of outcome measures in turn, analyzing the experiment as described in a pre-analysis plan filed with the Evidence in Governance and Politics (EGAP) registry before the experiment was fielded.Footnote 8
Figure 5 reports the estimated effects of a heavy police response on perceptions of protest violence. Consistent with hypothesis 1, the figure shows that the presence of armed police increases the perception substantially that protesters are “out to cause trouble.” The mean score on this item for those shown the news article with an image of a large police deployment is 0.15 points higher (on a 1–5 scale) than the mean for those shown the control article. For scale, this treatment effect is similar in size to the baseline difference between seniors (over age 60) and other respondents on this question (0.17 points) and to the differences between white and nonwhite respondents (0.22 points) in their perceptions of protesters’ troublemaking; both age and race are often thought of as key moderators of policing and protest opinion (Carey and Cisneros Reference Carey and Cisneros2023; Robinson Reference Robinson1970).
Effects of Heavy Police Presence on Violence Perception
We see similar effects for the violent intentions measure, which shows that the presence of armed police increases the likelihood that protesters are perceived to have “violent intentions” by 0.21 points on a 1–5 scale; it also significantly increases the perception of violence by 0.15 points on a 1–5 scale. These effects are relatively small but notable given the limited nature of the treatment (a sentence and photo in one news article).
Figure 6 displays the difference-in-means tests for the outcomes on support for BLM protests like the one shown in the vignette. The left panel displays outcomes measured on a 1–5 scale, whereas the right panel plots outcomes with a 0–100 scale. Here, we do not see evidence to support hypothesis 2: The figure shows null results for each outcome measure of support for BLM protests. We should note that these point estimates do show small but nonstatistically significant decreases as predicted by the hypothesis; that is, the presence of armed police substantively reduces the (1) willingness to get involved with a similar group, (2) willingness to attend a similar protest, (3) willingness to post something positive about a similar protest on social media, and (4) support for these protesters. However, these point estimates are substantively small and should be interpreted as null results.
Effects of Police Presence on Support for BLM Protesters
Section 7 of the Supplemental Information presents further robustness tests. Combining the survey measures into a single index measure for hypothesis 1 (perceived protest violence) and hypothesis 2 (support for the protesters) yields similar estimates: A heavy police response significantly increases respondents’ perceptions of violence but does not significantly reduce their support for the movement (though the point estimates are again in the expected direction).
Several aspects of these findings are noteworthy. First, and as predicted in hypothesis 1, these findings suggest that protests with a heavy police response are more likely to be perceived as violent. Second, they suggest that a single police response is not sufficient to shift overall support for protest movement actions. This latter result could indicate that people have strong preexisting attitudes about protests that are hard to shift with a single event. Taken together, however, these findings also suggest that repeated exposure to coverage of heavy police responses could reduce support for the protest movement if increased perceptions of protester violence produce negative beliefs about the nature of the broader protest movement.
Regarding the generalizability of these findings, we tried to maximize our external validity by using photos and details from real news coverage of protests in Baltimore after the police killing of Freddie Gray (Burton Reference Burton2015). The experiment was thus designed to resemble actual news coverage of BLM, such as news consumers might see after protests. Further, we expect our findings to travel beyond the immediate time period of this survey experiment. That we found similar results in a pilot study during an earlier (fall 2019) period supports this point.Footnote 9 In addition, BLM news coverage was at a low ebb during the 2019 pilot, which addresses the related concern that our findings depend on the presence of salient news about violence between police and protesters (as was the case in 2022). Finally, although our study is based on BLM protests, we do not think our results are unique to this movement. Variation in protest policing is not unique to BLM or the United States (Della Porta and Reiter Reference Della Porta and Reiter1998), and we expect the presence of armed police to contribute to perceptions of protest violence beyond this particular movement and context.
Racial Heterogeneity: Who Responds to Police Cues?
Our main preregistered hypotheses are about the average effects of protest-policing imagery on people’s reactions to protest events. But it is possible that some people are particularly likely to react to police cues; for example, people who hold the police in high regard may trust that police responses are proportionate to the threat posed by protesters and thus be especially likely to infer protester violence when they see heavy police deployments and tactics. People who distrust the police may not reach the same conclusions.
In this section, we look for heterogeneity in experimental treatment effects. Because we did not preregister this test or design the study with it in mind, it should be considered exploratory. We focus here on racial differences in response to the experimental treatment condition, given well-established differences in how people experience policing and view the police along racial lines (Jefferson, Neuner, and Pasek Reference Jefferson, Neuner and Pasek2021; Tuch and Weitzer Reference Tuch and Weitzer1997). We do not have direct measures of respondents’ preexisting attitudes toward police, but we expect that Black survey respondents will on average have lower trust in police than white respondents. We thus expect that Black respondents may be less likely to follow police cues and infer from images of armed police that a protest was violent.
Figure 7 presents experimental effects by race for our three measures of protest perceptions.Footnote 10 Indeed, white survey takers appear quite responsive to images of a heavy police response, showing clear and precise increases in beliefs that protesters had violent intentions, that the event was violent, and that protesters were out to cause trouble. Black respondents appear nonresponsive to the heavy policing cue, with essentially no difference between treatment and control in their perceptions of protester violence. Across all three outcome measures, there appears to be a large difference between Black respondents’ treatment effects and those for the rest of the sample, although only for the “violent intentions” measure does that group difference reach conventional levels of statistical significance. We report this pattern with caution because the study was not designed to be powered for subgroup analyses and interactions of this type, but we do consider it suggestive evidence of heterogeneity.
Racial Heterogeneity in Effects on Perceptions of Violence
We also look for heterogeneity in effects on our “protest-support” outcome measures, although we have less clear predictions for these measures. One prediction builds on our interpretation in figure 7: namely, that white respondents are more likely to react strongly to images of armed police and thus are more likely than nonwhite respondents to reduce their support for the movement. Another prediction is that treated Black respondents are more likely to reduce their support for the movement: If Black respondents are more likely to be afraid of the police than non-Black respondents, they may be less willing to get involved with, attend, and express support for protests with a heavy police response because of fears of police violence. Or, as we see in the full dataset, protest support may be sticky and unlikely to move in response to a single protest incident. These stories point in different directions and yield less clear predictions for treatment effect heterogeneity on our measures of support.
Figure 8 displays treatment effects by race on our four measures of protest support. In line with the cross-cutting stories described earlier, we do not see clear patterns of racial differences in response to the heavy police presence cue. If anything, it appears that Black respondents may become slightly less willing to attend a protest after receiving the heavy policing cue (consistent with fears of police violence), but these differences are small and noisily estimated.
Racial Heterogeneity in Effects on Protest Support
Conclusion
In early 2023, Memphis police fatally beat Tyre Nichols during a traffic stop. In the wake of his death, news stories trickled out about the details of the stop and about the imminent release of police body-camera footage of the beating. As Memphis officials prepared to release the footage on a Friday evening in late January, police departments across the country called up large groups of officers and prepared riot gear and other militarized equipment (Pegues Reference Pegues2023; Zraick Reference Zraick2023). With this police mobilization, media coverage of the situation rapidly became about the anticipated violence of the protesters who would turn out after the video’s release. News stories and social media images included photos of officers lining up and preparing for a long night of protest policing. After an evening of largely peaceful protests, news coverage reported on those peaceful events but continued to stoke concerns of potential violence (Bennett and Cuevas Reference Bennett and Cuevas2023; Heyward Reference Heyward2023; Sadowski, Lee, and Hind Reference Sadowski, Lee and Hind2023). Why, after an event of extreme violence by police officers, was the public conversation so thoroughly focused on hypothetical protester violence?
We believe, at least in part, that this pattern emerged because police mobilizations can help frame media coverage and public perceptions of protests. Large deployments of police imply that there is something to defend against and can thus lead members of the public to believe that protesters are violent and dangerous, even in the absence of violent protest tactics. In other words, we think the chain of events after Tyre Nichols’s death represents a broader phenomenon in which heavy protest policing can create the perception of protest violence.
Our original dataset of more than one thousand BLM protests between 2014 and 2017 demonstrates that the police pursued strikingly varied responses to these protests and that observed protest characteristics explain only a fraction of this variation. These findings suggest, first, that police departments exercised discretion in their response to the demonstrations. When coupled with the largely peaceful nature of the protests, these findings also suggest that the police pursued a disproportionately militarized and active response to BLM, an observation that finds additional support in more recent data on protest interventions (ACLED 2020).
In the second part of this article, we build on these insights to examine whether heavy protest policing can contribute to perceptions of violence and depress support for BLM protests. Our survey experiment shows that when a photo of armed police accompanies a report of a protest, respondents are more likely to describe the protest as violent and to ascribe violent and troublemaking intentions to participants. This finding suggests that a heavy police presence can create the perception of violence, regardless of the behavior of protesters, and that these results are the strongest among white respondents. We also find that exposure to a single police response is insufficient to shift support for the broader protest movement. Taken together, these findings suggest that (1) exposure to single episodes of police presence may not shift attitudes and intentions but that (2) repeated exposure to heavy police responses could reduce support for the protest movement if perceptions of protester violence affect perceptions of the broader protest movement over time.
This article makes several contributions to research on BLM, social movements, nonviolence, and protest policing. First, we make a substantive contribution to the study of BLM by presenting a police-centered explanation for public perceptions of BLM protests. Empirically, we present an original dataset that researchers can use to explore various aspects of the movement. The dynamic nature of digital news and social media means that it can be difficult to reconstruct what happened at protests years after they took place. Researchers working with current datasets about recent BLM protests may thus find it useful to refer to this dataset on earlier protests to learn about continuity or change within the movement.
Second, our results join recent work on the effects of media framing and protester identity on perceptions of violence in social movements (Arora, Phoenix, and Delshad Reference Arora, Phoenix and Delshad2019; Hsiao and Radnitz Reference Hsiao and Radnitz2021; Manekin and Mitts Reference Manekin and Mitts2021). Our findings show that the police can play a similar role in the construction of perceptions of violence, underscoring the broader point that these perceptions are subjectively shaped. Future work might bring these experimental findings into conversation and ask, for instance, how variation in policing and the racial or partisan identities of protesters can affect perceptions of protest violence. Alternatively, researchers might investigate the conditions under which a heavy police presence creates the perception of violent and troublemaking protesters versus public sympathy for the movement (Wasow Reference Wasow2020).
Third, and building on this point, our findings highlight that perceptions of nonviolence depend on more than the protesters’ use of violent tactics. This point is significant for the study of nonviolent resistance (Chenoweth and Cunningham Reference Chenoweth and Cunningham2013; Chenoweth and Stephan Reference Chenoweth and Stephan2011; Nepstad Reference Nepstad2015) and suggests, in particular, that scholars should be attuned to the potential for differences between actual and perceived violence and to the role of the police in shaping these perceptions. Social movement actors, too, should be attuned to this possibility: If nonviolence is a more effective means of social and political change, then interactions with the police take on additional strategic importance.
Finally, our findings contribute to theoretical and practical literatures on protest policing. On a theoretical front, we highlight that the police should be viewed as important strategic actors in protests and social movements. Although much of this literature focuses on the interaction between policing and protesters (Earl and Soule Reference Earl and Soule2006; Earl, Soule, and McCarthy Reference Earl, Soule and McCarthy2003; McAdam Reference McAdam1983; Williamson, Trump, and Einstein Reference Williamson, Trump and Einstein2018), we emphasize instead the impact of policing on public opinion about protests. Specifically, our results indicate that the police are strategic and influential actors in the construction of broader narratives about social movements.
From a policy perspective, our findings suggest that political actors may wish to incorporate public opinion effects into their decisions about protest policing. In the wake of the 2020 protests following George Floyd’s death, many city governments considered or passed rules about the police’s use of force or of specific weapons at protests. Boston restricted the use of tear gas and rubber bullets, whereas Columbus passed a law stating that officers cannot inflict pain to punish or deter nonviolent protesters and must have their names clearly visible on their riot gear (Bush Reference Bush2022; WBUR 2021). These policies are clear responses to cases of immediate physical harm done to protesters in these cities. But our findings suggest that cities may also want to constrain the political impact of protest policing on broader movements, because police arriving in riot gear (even with their names visible) at a nonviolent protest can dramatically shift the public perception of that event. As such, police departments may have the opportunity to strategically use their discretion to deploy large and militarized forces in ways that discredit protesters with whom they disagree. Just as top-down directives can limit harsh street-level policing tactics (Mummolo Reference Mummolo2018), policies requiring clear justification and documentation of heavy police deployments and tactics at protests could limit their discretionary use and their political ramifications.
Supplementary material
To view supplementary material for this article, please visit http://doi.org/10.1017/S1537592724002731.
Acknowledgments
For their research assistance, we thank Diana Camilla Valerio, Kelechi Alfredi-Igbokwe, Mackenzie Biven, Andrew Kruse, Avery Nguyen, Rob Pressel, Ashley Ann Thompson, Kathryn Treder, Sarah Vu, and Anna Weissman.
Open access
