Abstract
Media framing of social protests can influence public opinion and governmental response. An extensive line of scholarly work had pointed to the existence of two alternative news frames; public order and debate. We argue that prior work may have been limited by the reliance on deductive strategies using predefined, theoretically-driven frames. Using a data-driven computational method, the Analysis of Topic Model Networks (ANTMN), we examine mainstream news’ framing of two contentious protests that took place during Donald Trump’s presidency; Unite the Right rally in Charlottesville, VA (n = 1231 news articles), and the Black Lives Matter protests (n = 2810). In addition to the frames found in past research, we identify a prominent Politics frame, often focussing on the role of Trump in inciting and reacting to racial tensions. An in-depth analysis of the application of frames to each protest revealed a nuanced use of the Protest Paradigm. We suggest possible revisions to existing theories, and discuss the potential social and political implications of our findings.
Most people learn about social protests through the mass media (Lee, 2014) and as a result are influenced by media portrayals of protests and protesters (Gitlin, 1980). The ways protests are portrayed in the media can influence public opinion (Nelson and Oxley, 1999) and in some cases even influence the approach taken by officials, such as police forces, towards controlling the event (Donson et al., 2004). Thus, media coverage of social movements and protests are crucial for their success or failure (Gamson and Wolfsfeld, 1993). It is unsurprising then that the relationship between protests and the news media received ample academic attention since the social unrest of the late 1960s (McCurdy, 2012). Multiple studies have demonstrated the ability of the media to legitimize or marginalize protesters and their causes (McLeod and Detenber, 1999).
Academics and journalists have argued that framing plays a central role in the journalistic work routine (Nisbet, 2010; van Gorp, 2010). News ‘emphasis frames’ are defined as the selective presentation of events and issues through a focus on arguments, perspectives and facts, that could in turn influence audiences’ thinking and reactions to news items (Entman, 1993; Gitlin, 1980). Prior work on the framing of protests tended to measure media frames deductively, using pre-defined, theory-based, categories, while focussing on individual case studies. For example, a foundational work (Nelson and Oxley, 1999) analysed the effects two frames, freedom of speech and public order on audiences’ support for social protests (Nelson et al., 1997). However, the authors did not establish the prevalence of these frames in real world coverage, instead assuming their salience. The current study advances our understanding of social protests by using a mixed-method analysis of the coverage of two social protests in multiple media outlets. Additionally, we provide a comparative perspective by examining two high-profile protests, organized by both the left and the right, around questions of racial injustice. Specifically, we analyse the coverage of the Unite the right rally in Charlottesville, VA and the BLM demonstrations following the death of George Floyd.
Media framing of social protests
Framing remains one of the most prominent, yet contested, theories in communication research (Matthes, 2009), resulting in continuing debates around how to best conceptualize, identify and measure frames’ prevalence in news corpora (D’Angelo, 2018; van Gorp, 2010). News emphasis framing, as opposed to equivalency framing is the journalistic practice involving the selective presentation of information that could influence attitudes, beliefs, decisions and behaviours (Chong and Druckman, 2007). Framing choices are central to the journalist’s work (van Gorp, 2010), an ‘unavoidable reality of the public communication process’ (Nisbet, 2010: 44), guided by ideology, newsworthiness and other factors (Gitlin, 1980).
Studies on media framing of protests took either issue-specific or general approaches to framing (D’Angelo, 2018). General approaches seek broad coverage patterns used across contexts and topics. For example, a study (Wouters, 2015) on protests in Belgian television news harnessed the general frames of thematic and episodic, suggested by Iyengar (1990, 1994) for the description of other social and political issues. Other studies identified media frames that were specific to the context of social movements and protests. For example, the prominent ‘protest paradigm’ (McLeod, 2007) claimed that the news media often serve as instruments of control, generally taking an oppositional stand against social protests, portraying them as deviants who depart from social norms (McLeod and Shah, 2015). This is done, in part, by using frames that ‘focus on protesters’ appearances rather than their issues, emphasize their violent actions rather than their social criticism, pit them against the police rather than their chosen targets, and downplay their effectiveness’ (McLeod and Detenber, 1999: 3). They identified the prominence of derogatory, marginalizing news frames, emphasizing confrontations between police and protests, riots, using terms like ‘deviants’, ‘criminals’, ‘circus’, ‘freak show’ and ‘extremists’ (Hall et al., 1978; Solomon, 2000). Rarely, they argued, does the media frame protests as a conversation and debate between policy and agenda positions. Similarly, another framework, the ‘public nuisance paradigm’ have argued that American media coverage tends to depict protesters as a bothersome interruption of everyday activities (Di Cicco, 2010).
It was argued that such coverage supports existing power structures and the status quo (Shoemaker and Reese, 1996) by eroding public support in protest activities (Boykoff, 2006). The argument gained empirical evidence (Nelson et al., 1997) where participants exposed to news stories framing a KKK rally through free speech were found to express more tolerance towards the demonstrators than those exposed to a public order frame. While the Protest paradigm remains prominent in the coverage of cases such as the anti-World Trade Organization protests in Seattle (McFarlane and Hay, 2003) or Occupy Wall Street (Xu, 2013), more recently researchers have shifted away from the assumption that news media is inherently biased against social protests into examining how media coverage is affected by the characteristics of the protest and protesters (Boyle et al., 2012). The change in was due, in part, to two changes in media structure, such as fragmentation and polarization (Stroud, 2011). In addition, emerging online citizen journalism and digital news media were relatively free from the journalistic routines of the protest paradigm (Harlow et al., 2020).
Changes in coverage were also influenced by the normalization of social movements (Cottle, 2008). Dixon et al. (2018) found that movements like BLM and events of police brutality have contributed to television news shifting away from stereotypical and negative depiction of African Americans and the movement (Kilgo, 2017) to a more nuanced and balanced framing (Dixon et al., 2018). The more complex, and not necessarily negative, media coverage of protests was moderated by factors, such as ideological slant (AlMaskati, 2012), the extremity of the tactics used by the protesting group (Boyle et al., 2012), conflict and police involvement (Dixon et al., 2018).
Common to the studies of the media framing of social protest described above is the reliance on (a) manual coding of small samples and (b) deductive reasoning, where the identification of frames followed theoretical assumptions that directed the identification of specific coverage patterns (e.g. the assumption that media frames could either express sympathy towards or marginalize social protesters, see Wouters, 2015). Such deductive approaches for framing analysis, while insightful, are prone to theoretical and subjectivity biases (Matthes and Kohring, 2008; van Gorp, 2010). In the following section we discuss those limitations and the potential contribution of computational inductive alternatives. We then apply the Analysis of Topic Model Networks (ANTMN) approach (Walter and Ophir, 2019) to identify the media frames that the American news media had been using across articles and outlets when covering two highly contentious social protests during the Trump presidency and examine whether such a data-driven approach would reveal frames that were overlooked before.
Identifying news frames – The shift from deductive to inductive strategies
A deductive strategy for framing analysis relies on predefined, theoretically-based and limited sets of frames in corpora (Matthes and Kohring, 2008). A prominent example in the realm of protests framing is McLeod’s (2007) analysis of riot and debate frames. In both of these studies, the authors built the case for the existence of the aforementioned frames based on theoretical grounds, and then tested their effects on audiences, without conducting content analyses to verify their existence or prevalence in real world coverage of protests. Similarly, Wouters (2015) looked for thematic and episodic framing devices in news articles based on the assumption that these competing frames are suitable for the analysis of social protests.
The use of deductive analyses limits the lenses through which coverage could be analysed, in a way that is often biased by the background, knowledge and expectations of the researchers (van Gorp, 2010). To reduce biases, researchers recommended identifying frame elements instead of full frame packages (Gamson and Modigliani, 1989). The elements include word choices, metaphors, catchphrases, exemplars and visual icons (Baden, 2018). A frame package is the reoccurring journalistic use of those linguistic features, such as framing devices and logical chain of reasoning devices, together (van Gorp, 2010). In other words, frame packages are what communication researchers have often referred to simply as ‘frames’, but the term ‘package’ highlights their modular nature. Additionally, researchers have recommended adopting inductive approaches (Walter and Ophir, 2019) that do not assume that a-priori frames are suitable for and exhaustive of all contexts (Matthes and Kohring, 2008), and are less prone to finding frames researchers were consciously or unconsciously looking for (van Gorp, 2010).
Importantly, both inductive and deductive approaches are crucial for framing analysis and could complement each other when used in different stages of analysis. Specifically, while deductive approaches are crucial for the development and application of existing theory, inductive approaches can help us identify emerging or changing frames, as communication about social protests could be dynamic and change over time, place and context, or in light of emerging journalistic practices. Inductive approaches, at least at the frame detection and measurement phase, can help us identify those changes to frame structure, that could then be incorporated back into theories after analysis. The method applied here, ANTMN (Walter and Ophir, 2019), combines deductive and inductive analyses to benefit from both. As detailed in the method section, ANTMN identifies frames inductively, and implement theory-based knowledge at the interpretation level deductively.
The rapid growth in digital data (DiMaggio et al., 2013) led communication scholars to develop automated approaches for deductive identification of news frames, often based on a systematic identification of frame elements that appear together (co-occur) across texts (Matthes and Kohring, 2008). These approaches are consistent with the definition of news frames as repeating use of ‘objects and traits, using identical or synonymous words and symbols in a series of similar communications that are concentrated in time’ (Entman et al., 2009: 177). Several approaches were taken towards conducting inductive framing analysis, often using topic modelling (DiMaggio et al., 2013) or semantic networks (Baden, 2018). A recent approach, ANTMN, combined both approaches into a multi-step method (Walter and Ophir, 2019).
ANTMN consists of three steps. First, frame elements are identified using topic modelling, without assuming, as was done in past studies (DiMaggio et al., 2013), that topics could be interpreted as frames. Instead, topics are operationalized as frame elements (Gamson and Modigliani, 1989). Second, the relationships between topics are explored in the form of a network, in which topics serve as nodes, or objects in the network, and the relationships between them (commonly referred to as edges) are calculated by their co-occurrence over documents. Third, community detection techniques (Walter and Ophir, 2019) are used to cluster together topics co-occurring in documents into coherent frame packages.
Here, we employ ANTMN to the analysis of major and highly contentious social protests that took place during the Trump presidency, that was accompanied by myriad protests, including the March for science, March for our lives, the Women’s march and the two studied here, the Unite the Right rally and the BLM demonstrations. We examine the media frames employed by four mainstream media outlets – the Washington Post, the New York Times, the Wall Street Journal and USA Today. To allow for a nuanced comparative analysis, we looked at one social protest that was largely condemned by the public and political establishment, the Unite the Right rally in Charlottesville, VA, and one that, while being strongly shaped by partisan identity, has received overwhelming support from the American population on average, resulting from the murder of George Floyd and other Black Americans by law enforcement (Voytko, 2020). We complement the automated analysis with an in-depth analysis of the application of each frame to each protest in light of Entman’s (1993) typology, claiming that ‘to frame is to select some aspects of a perceived reality and make them more salient in a communicating text, in such a way as to promote a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation’ (p. 52). We ask the following research questions:
RQ1: What media frames were employed by journalists covering the Unite the Right and BLM protests?
RQ2: Did the media use different frames for the coverage of the two protests?
RQ3: Did the use of frames change over the timespan of the protest?
Case studies
On the evening of August 11th, about 250 white nationalists marched through the University of Virginia campus, carrying torches and chanting slogans (such as ‘You will not replace us’, ‘blood and soil’ and ‘White lives matter’). A fight erupted between the marchers and counter-protesters, leading to minor injuries. Rallies continued on the next day, and clashes between the hundreds of white nationalists, some of whom were armed with weapons, and thousands of counter-protesters ended on a tragic note, after a car driven by a white nationalist accelerated into crowds of counter-protesters, resulting in dozens of injuries and the death of 32-year old Heather D. Heyer. A public outrage followed Trump’s statement on August 12th, that ‘we condemn in the strongest possible terms this egregious display of hatred, bigotry and violence on many sides, on many sides’. Two days later, Trump released a second statement, blaming ‘K.K.K., neo-Nazis, white supremacists and other hate groups that are repugnant to everything we hold dear as Americans’. After members of the President’s American Manufacturing Councils moved to disband the councils in protest of his response, Trump dissolved both councils (Heim, 2017).
The 2020 BLM protests
The second social protest studied is a series of protests which began in Minneapolis, MN, on May 26th 2020 in response to the killing of a 46 year old African American George Floyd by police officer Derek Chauvin, who reportedly knelt on Floyd’s neck for 8 minutes and 46 seconds (a number that became a rallying cry by activists in future protests). in the presence of three other officers, who were later arrested (Taylor, 2020). The protests soon expanded to marches and events taking place in over 2000 cities and towns in over 60 countries around the globe. While the majority of protests have been peaceful, some have escalated into riots, looting, vandalism and violent clashes between protesters, the police and counter-protesters. The weeks-lasting protests led to thousands of arrests and to Trump’s deployment of National Guard forces. The protests refuelled discussions of removal of Confederacy symbols (Taylor, 2020).
Method
Data
We collected all the articles published by the New York Times, Washington Post, Wall Street Journal and USA Today from Factiva (an online database owned by Dow Jones & Company containing the full archive of the sources listed above). For the Unite the right rally we searched the terms ‘Charlottesville’ and (‘rally’ or ‘protest’ or ‘demonstration’ or ‘riot’), yielding 1620 articles between 07/11/17 and 09/11/17. For the BLM protests we used the terms (George Floyd or BLM or blacklivesmatter) and (‘rally’ or ‘protest’ or ‘demonstration’ or ‘riot’), resulting in 2502 articles between the day of Floyd’s killing, 05/27/20 and 07/27/20. The Charlottesville rally was announced months before it took place and coverage began before the actual protest. We collected data for Charlottesville from 1 month before and until 1 month after the protests. The BLM protests erupted surprisingly as response to the murder of George Floyd, and coverage was more prolonged. We therefore collected data for 2 months after the initiating event.
Procedures and measures
Topic modelling
Topic modelling is an unsupervised method for text analysis (Blei et al., 2003), based on a Bayesian generative approach, in which the model attempts to ‘mimic’ the writing process of a given corpus of documents. ‘Topics’ are sets of frequency distributions of words, based on the linguistic assumption that words that are being used more frequently in the same documents also associate thematically (co-occurrence). Every word in the corpus has a probability of appearing in each topic, and every document is composed as a mixture of all topics. Topic modelling is a ‘bag-of-words’ approach, which means that narrative, location in the text and syntax are not taken into consideration. We used Latent Dirichlet Allocation (LDA) and Gibbs Sampling (LDA, see Blei et al., 2003), using the LDA and LDAtuning packages for R.
Texts for both the George Floyed and Unite the Right corpora were modelled together 1 and pre-processed by removing stop-words, converting to lowercase, removing punctuation, numbers and words appearing in more than 99% of documents or in less than 0.5% (in this order). We refrained from lemmatization or stemming due to its possible negative influence on topic stability (Walter and Ophir, 2019). Next, we used statistical indicators of fit to choose the optimal number of topics. A range of models containing 10–100 topics was first estimated by ‘skips’ of 2. Based on four fit indicators (Walter and Ophir, 2019) we chose a model of 54 topics (see Appendix). To interpret the topics, we examined three types of information; the words with the highest loading over each topic, the words that are both prevalent and exclusive to each topic (FREX), and the full documents that are most representative of each topic. Two researchers qualitatively analysed and labelled each topic based on a close reading of the word lists and the full documents.
Network parameters
To create the network, we calculated pairwise cosine similarity between topics based on their co-occurrence in documents (using the theta matrix, in which columns represent the various topics and the rows represent the documents. The result is a fully connected, undirected, weighted network, where edges provide information on the extent to which topics are related (Walter and Ophir, 2019). In simpler terms, in this network, the nodes or objects, are the topics identified and labelled, and the edges, or connections between the topics, are defined by how likely they are to appear together. Finally, a Louvain community detection algorithm (Walter and Ophir, 2019) was employed to identify frames (or clusters of topics that tend to co-occur).
In-depth qualitative discourse analysis
Discourse analysis is an interpretive method that evaluates texts against the context(s) in which they are created. In this case, the texts were considered against the larger context of both protests. For the purposes of this project discussing news media framing, the analysis was firmly guided by the four frame functions theorized by Entman (1993): ‘problem definition, causal interpretation, moral evaluation, and/or treatment recommendation’. The qualitative analysis of the application of each frame to each protest was done by an expert in discourse analysis of social protests and media, closely reading a total of 100 articles per frame per protest (N = 600 articles). The analysis followed the conventions of discourse analysis and the study of language in use (Gee and Handford, 2013). Accordingly, the researcher read the articles to understand how the frames were constructed at the vocabulary level, to examine the ideological associations of the words and phrases in the articles, and at the structure level, to understand what narratives and themes were being proposed (e.g. who are the actors, who or what is identified as problematic or as the origin of the issue being discussed). Finally, for the labelling of topics, the researchers also examined the list of top-words, that is, the words that were most associated with each topic (Walter and Ophir, 2019).
Results
RQ1 asked what media frames were employed by American newspapers covering the two social protests. Our ANTMN (Walter and Ophir, 2019) model of 54 topics (The most representative words for each topic can be seen in the Appendix). Topics were clustered by the Louvain community detection algorithm into three frames. Four topics referring to the non-political circumstances of the protests (COVID-19, Juneteenth and 4th of July) were not associated with any frame (accounting for only 7.41% of the corpus) and, although clustered together, did not constitute a news frame, but rather served as additional linguistic cues that existed along the main frames (McLeod and Shah, 2015).
Two frames were congruent with prior conceptualizations of media frames of social protests (e.g. in the ‘protest paradigm’, see McLeod, 2007). The qualitative reading of the topics (top words and top documents) included in each frame (cluster) led us to conclude that the first frame (red in Figure 1, used in 31.48% of the corpus) fit the definition of the public order frame. These topics depicted social unrest, violence and clashes between protesters and enforcement officers. It included broad topics like ‘police violence against protesters’ and ‘looting and destruction’, and more specific ones, like the incident where protesters were dispersed using violence and tear gas to allow Donald Trump to pose with a bible next to a DC church. Other topics included discussion of curfews, damage to businesses, the size of protests (‘crowds’), calls to defund or reform the police and topics focussing on protests in specific cities. Interestingly, the topics of ‘George Floyd’ and ‘Breonna Taylor’ who were not protesters but rather the victims of police violence that incited the second protest, were included in the frame of public order as well. We discuss this classification further below.

Topic Network using ANTMN with Louvain community detection algorithm. Each node represents a frame element (topic), with size corresponding to relative salience. Edges between nodes represent co-occurrence in documents. Colour represents frame packages.
The qualitative reading indicated that the second frame (22.22%) was also congruent with the ‘protest paradigm’. The debate frame (in orange in Figure 1) consisted of topics dealing with the substance of protests (McLeod and Shah, 2015), mainly issues related to racial tensions and injustice, that were at the heart of both the Unite the Right rally and the BLM protests (arranged by white supremacists, and BLM activists, respectively). The frame included topics about the BLM movement, Civil Rights activists (‘John Lewis’), racial discrimination, values, education and narratives (e.g. of Rev. William M. Aitcheson, a former KKK member turned Catholic priest). As explained earlier, the topics ‘George Floyd’ and ‘Breonna Taylor’ were associated with the public order frame, that has more to do with clashes between police forces and protesters and not with the debate frame and its focus on race relations. We believe this was the result of the unique nature of these protests, where police officers served both as the target of the protest and as the official agents responsible for containing the demonstrations. Our model suggests that journalists writing about their deaths by police officers discussed them in tandem with police brutality and violence in the protests. Our empirical data suggest that while journalists could discuss George Floyd and Breonna Taylor as part of their discussion of racial tensions, they more often discussed them as part of the public order frame package.
Lastly, the most prominent frame in our corpus, which we qualitatively labelled the political frame (38.39%), absent in prior research, was dedicated to the political impact of the protests. While some topics examined broad political processes (e.g. ‘economy’, ‘polls’, ‘social media’, ‘legal issues’, ‘Congress & Senate’), others were episodic, often focussing on the individual at the top of the American political system – President Trump. Notably, in the context of our study and the two protests covered, this frame could have been labelled ‘The political ramifications of Trump’s mishandling of the protests’. However, as we attempt to develop the theory of social protests framing in general, and wish not to limit our findings to the two events studied here, we prefer to label it the political frame. We believe that the frame is likely to reappear in the coverage of other protests, but the details and the valanced criticism of Trump would be replaced with the specificities of other events, and hence prefer to label it the political frame.
In the context of the Unite the Right and the George Floyd protests, the political implications surrounded the aftermath of Trump’s perceived-failure to control the event. These included a topic about COEs resigning from Trump’s American Manufacturing Council in response to his failure to condemn the white supremacist protesters (‘CEOs vs Trump’), or a topic on Trump’s threats to contain protests via a deployment of military forces. Even topics that were not exclusively about Trump, like ‘Congress & Senate’ often referred to him intensively, as Congressmen and Senators in this case were often responding to his handling of the protests. Other episodic topics focussed on the murder of counter-protester Heather Heyer and the controversial op-ed by Republican Senator Tom Cotton that was published in the New York Times, titled ‘Send in the military’.
RQ2 examined differences in the coverage of the two protests. A series of t-tests found significant differences between conditions (see Figure 2). Public order (t(4039) = 29.6, p < 0.001), was used significantly more in the BLM protests (M = 37.3%, SE = 0.003) than in the Unite the Right rally (M = 18.1%, SE = 0.005). The politics frame (t(4039) = −46.6, p < 0.001) was used more in the Unite the Right rally (M = 59.4%, SE = 0.005), though it was relatively prominent in the BLM protests as well (M = 30.8%, SE = 0.003). The debate frame (t(4039) = 8.94, p < 0.001) was used prominently in both, but more in the coverage of the BLM protest (M = 24.1%, SE = 0.003) than in that of the Unite the Right rally (M = 19.3%, SE = 0.004). Finally, the topics related to specific events (e.g. Juneteenth or COVID) were small in both protests, but relatively more prominent (t(4039) = 12.8, p < 0.001) in the BLM protests (M = 7.8%, SE = 0.001) than Unite the Right (M = 3.28%, SE = 0.003).

Relative use of media frames by protest.
To better understand the argumentative nature of each frame in our corpus, the top representative texts for each frame, used for each protest (N = 600) were closely read with attention to Entman’s (1993) typology of problem definition, causal interpretation, moral evaluation and treatment recommendation.
The public order frame
The t-tests showed the public order frame was significantly more prominent in the coverage of the BLM protests that of Unite the Right rally. This might be explained by the greater complexity of the BLM protests. Our analysis of the coverage of the Charlottesville events identified only three main actors: members of the alt-right, counter-protesters and the police. Each group was covered differently: alt-right members were depicted as violating pre-established agreements for the rally; peaceful counter-protesters were carefully distinguished from that of a minority of more militant Antifa members; the police’s response was characterized as inadequate.
As opposed to the more limited nature of Charlottesville, journalists covering the BLM protests had to report and contextualize unrest in multiple cities (e.g. Portland, Minneapolis, Louisville and Washington D.C.) over a longer period of time (months), in response to a myriad events (the deaths of George Floyd, Rayshard Brooks, Breonna Taylor and David McAtee). Furthermore, the BLM protests involved many law enforcement actors, with some confusion regarding the specific role of each (e.g. police, military, CBP and DHS agents). Additional confusion arose from the lack of clarity regarding the responsibilities and mandate of cities, states and the federal government. The frame consisted of description of violence against and by protesters (similar to the definition of the frame at the Protest paradigm), but due to the unique circumstances, also violence against Black citizens, which were the source of the protests. To a lesser degree, the media did cover violence by the protesters and civilians (in topics such as ‘looters and destruction’). Taken together, these explain the media’s heavier reliance on the public order frame for the BLM protests.
Examined through the lens of Entman’s typology, both protests were covered with a focus on violence by law enforcement agents, and not the protesters themselves, thus the onus of responsibility rested on these actors. The journalistic definition of problems and solutions were markedly different for each protest. In Charlottesville, the blame rested on the alt-right’s deviant, aggressive behaviour. The proposed solution, therefore, was for law enforcements to better respond to alt-right violent acts, in ways that could prevent the death of Hether Heyer. The blame in the application of the public order frame on BLM protests rested in police brutality and violent acts by officials before and during the protests, in light of lingering discrimination of African Americans. The solution suggested had to do with broader law enforcement reforms. As law enforcements’ systematic racism was the cause for the unrest, the protest were generally evaluated as morally necessary by the journalists. Therefore, this frame of public order was more suitable and prominent for the coverage of BLM.
The politics frame
Although the politics frame was prominent in the coverage of both protests, the t-tests found a higher reliance on it for the Unite the Right rally. Both events were influenced by the actions and words of Trump and the politics frame was mostly deployed to contextualize the president’s rhetoric and its consequences. Our analysis revealed the frame promoted an interpretation of the racially-charged events through the lens of an economic-political perspective. After the Charlottesville events, Trump’s statement that there were ‘very fine people on both sides’ was followed by a swift backlash from the public, journalists and the business community, resulting in the resignation of CEOs from the Manufacturing Jobs Council. Potential repercussions for the political campaigns for the following year mid-term elections were also discussed. Journalists focussed their blame on Donald Trump’s inability to condemn neo-Nazis and the alt-right, thus suggesting a clearer condemnation as a solution to the problem. Pressures by CEOs on Trump were aimed towards that solution as well.
When applying the politics frame on the BLM protests, journalists focussed primarily on the response of social media companies (e.g. Facebook and Twitter) to Trump’s inflammatory and misleading discourse online, promoted also by right-wing media. The application of the frame put the blame for the protests on Trump again, but sought a solution in regulation of and action by social and mainstream media companies, that provide Trump with a platform to spread his messages. Here again, the proposed solutions were broader, and go beyond the specific protests into larger questions of freedom of speech and media regulation. This analysis further demonstrates that politics was salient in coverage of both protests, but the president himself was more central to the coverage of Unite the Right. When using more episodic framing, by positioning Trump’s vitriolic rhetoric as problematic, the coverage of the two protests shifted the locus of individualistic attribution of responsibility, typical of the protest paradigm, away from the protesters themselves, and towards the president’s conduct (Iyengar, 1990).
The debate frame
The debate frame in the coverage of the BLM protests was characterized by the voices of pop culture icons (e.g. Ice-T and Internet memes employing fictional characters like Herman Munster) and Black producers (e.g. Ava DuVernay), activists (Curtis Hayes), young female organizers (e.g. Zee Thomas and Tiana Day) and concerned parents, all of which were cited and mentioned in the news articles analysed. The frame problematized the racial division in American society by foregrounding the role of education as essential to the understanding of racism and privilege, the necessity of allyship and the key role of media in increasing representation. The frame was very consistent in the identification of ignorance as the primary problem in the struggle against racism, and offered education as the solution, whether direct, in the case of family conversations, or mediated by cultural artifacts.
Popular culture is also present in the debate frame of the Unite the Rally coverage, although here it oscillates between being a means of conversation and consciousness raising (e.g. in the context of the MTV video Music Awards) and the cause of such a divisive and racially charged public (e.g. in programs like The Bachelorette). Interestingly, the debate framing of the Charlottesville riot also highlighted the role of the Church, as the White faith community was portrayed as a problematic space that, despite carrying a lot of weight, is still resistant to engage in the necessary sociopolitical transformation, if not blamed for being downright in favour of white supremacy at times.
The debate frame was the most homogenous between the two corpora. In both cases, the problem and solution concentrated on the issue of racism, and offered education via interpersonal and mediated communication as a remedy. These revealed similarities in the interpretation of the problem and the possible solutions (allyship, representation) might explain the relatively similar applicability of the frame to each protest. Once again, the protests are presented as the moral solution to a lingering racism problem (see Figure 2).
Finally, RQ3 examined how framing changed over the timespan of protests. There were substantial differences between the coverage of the BLM (Figure 3) and Unite the Right (Figure 4) protests. Before the Charlottesville events, journalists focussed on racial and political tensions. However, the media quickly abandoned the topic of race in exchange for more political tone as one of the rally organizers, Jason Kessler, sued the City of Charlottesville for requiring him to move it from Lee Park (the location of the Robert E. Lee statue) to McIntrie Park. As protesters and counter-protesters flooded the streets of Charlottesville on August 11 and 12, the media kept focussing on politics, but increased the public order framing. After the marches and the death of Heather Heyer, the media’s attention was strongly directed to politics, specifically to Trump’s infamous reaction (‘very fine people on both sides’), a statement perceived by critics as implying a moral equivalence between white supremacist and those who oppose them. The focus on the political impact and fallout remained prominent throughout the rest of the time studied.

Relative use of media frames over the events timeline of the George Floyd protests.

Relative use of media frames over the events timeline of the Unite the Right rally.
The BLM protests show a very different pattern. Early on, after the killing of George Floyd on May 26th, the media focussed on public order (which consisted of the discussion of his murder). As the initial protests in Minneapolis spread to more than 550 places across the US at the beginning of June, there was an increase in the focus on the racial debate, as well as on political aspects, following Trump’s announcement on June 3rd that if cities and states refuse to take what he considered proper actions to contain the protests, he will deploy the US military and ‘quickly solve the problem’.
The balance between the frames remained relatively unchanged (with minor decreases in public order and increases in politics, once again followed by Trump’s executive order permitting federal agencies to provide personnel to assist the protection of Federal monuments and property) until clashes between protesters and Trump’s deployed federal forces escalated in Seattle and in Portland, where unidentified federal officers wearing camouflage used unmarked vans to detain protesters. These violent encounters accounted for temporary increases in public order framing.
Discussion
Our analyses using comparative mixed-method approach supported the prominence of two frames identified in the protest paradigm; public order and debate. We identified a third, prominent, frame, similar to what was found in framing studies in other contexts (Ophir, 2018), dedicated to the political impact and implications of protests. Our close reading examined the argumentative nature of frames applied to each protest, in light of Entman’s (1993) typology. The case studies presented an interesting example for recent shifts in racial stereotypes and protest paradigm in traditional media: not only had the frequency of association of crime and unrest with Blacks and protesters been reduced, we found evidence that it has been inverted. In other words, the overtly racist nature of the events served to intensify the erosion of the ‘ethnic blame perspective’ identified by Dixon et al. (2018) and the emergence of new frames for the coverage of BLM.
Similarly, the novel political frame emphasized Trump’s moral compass, and his failure to contain social tensions around both protests. However, we believe that the application of the frame to other protests may take a somewhat different focus and valence within the boundaries of political implications. Importantly, while traditional news frames have managed to link social problems to specific groups, the political frame could prime public evaluations of the president’s own racism, suggesting a decisive move towards a breakage of the Black outgroup/White ingroup journalistic distinction, in favour of a heightened evaluation of the larger socio-political context.
Prior work may have failed to identify the political frame due to reliance on pre-defined categories (van Gorp, 2010). Alternatively, journalistic practices may have changed over time and in response to the Trump administration. We found that the political frame prioritized the personal over the abstract (Chong and Druckman, 2007; Iyengar, 1994), with a special attention given to Trump. This could be a result of the perceived-‘Trumpization’ of the Republican party (Rubin, 2020). Personification could intensify information processing and enhance the attention audiences pay to the role played by individuals in the escalation of violence (McLeod and Shah, 2015). Personifying frames can lead to the attribution of responsibility both for the creation and resolution of social problems to the people featured in articles (Iyengar, 1994). In light of that perspective, the news media focus on Trump may, on average, increase public criticism of his handling of the protests. However, while Iyengar’s (1994) work on poverty or McLeod and Shah’s (2015) work on activists examined the media prominence and effects of episodic media framing of unfamiliar individual subjects, Donald Trump is a highly contentious political figure that often induces strong emotions from both supporters and opponents (Crabtree et al., 2018). Future empirical work could examine potential interactions of that frame with factors such as ideology, partisan affiliation, or political attitudes, and whether or not exposure to the political frame reinforces prior predispositions due to partisan-oriented motivated reasoning (Gaines et al., 2007). Future studies may also enhance our understanding of journalistic practices through ethnographic work.
Our qualitative analysis of framing revealed a general support for BLM protesters and their cause. No equivalent support was expressed towards the Unite the Right rally. In both protests, the blame was put on the government, both in terms of systematic discrimination (as in the deaths of George Floyd and other Black citizens) and episodic events (as in Trump’s failure to condemn the alt-right and neo-Nazis). Both quantitative and qualitative analyses pointed to differences frames salience, where BLM protesters were covered more through the public order frame, due to the role played by police brutality as both a cause for the protests and their violent escalation. The increased prominence of the Political frame in the coverage of the Unite the Right rally can be explained by the pivotal role Donald Trump played in intensifying racial tensions, due to his failure to condemn the alt-right. Finally, the debate frame was used about equally in both protests, emphasizing the problem of racism towards African Americans and other racial minorities, while rejecting the white-supremacist claim for anti-white racism. Taken together, in both protests the journalists took a moral stand for racial equality and against systematic racism, supporting recent work on changes in the journalistic application of the Protest paradigm (Boyle et al., 2012; Dixon et al., 2018).
A caveat should be noted regarding the unique nature of the protests. Prior research on the ‘protest paradigm’ argued that the media prioritize stories that depict protests via the lens of ‘protesters versus police’ rather than ‘protesters versus the intended target of the protest’ (McLeod and Shah, 2015). In the BLM protests, however, such distinction is impossible. This resulted, for example, in the fact that the topics ‘George Floyd’ and ‘Breonna Taylor’ were associated with the public order frame, that has more to do with clashes between protesters and police forces, and not with the debate frame that emphasized racial tensions and inequalities.
To sum, using a mixed-method approach, where a data-driven computational approach was used for identifying media frames (and their elements) and a qualitative reading investigated the application of the frames to each protest, we found the that ‘protest paradigm’ was present, yet modified and extended by journalists covering protests at the age of Trump’s presidency. Future studies could shed light on the effects of the increased focus on politics and the president on audiences’ motivated reasoning based on political ideology and partisan identity and its social implications.
Supplemental Material
sj-docx-1-jou-10.1177_14648849211036622 – Supplemental material for News media framing of social protests around racial tensions during the Donald Trump presidency
Supplemental material, sj-docx-1-jou-10.1177_14648849211036622 for News media framing of social protests around racial tensions during the Donald Trump presidency by Yotam Ophir, Devin K Forde, Madison Neurohr, Dror Walter and Virginia Massignan in Journalism
Footnotes
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Supplemental material
Supplemental material for this article is available online.
Notes
Author biographies
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
