Media, Terrorism, and Political Ideologies:
Framing of Right-Wing and Islamic Terrorism
by Conservative and Liberal U.S. Media
Dana Poláčková
A thesis submitted for the degree of
Master of Science in Crisis and Security Management Faculty of Governance and Global Affairs
Leiden University
Supervised by Dr. P.G.M. Aarten The Hague, January 9, 2018
ii
Abstract
Media frame terrorism in a variety of ways—they highlight different aspects, ascribe moral judge-ments, and give specific meanings to events (Entman, 1993). This paper examined how political ideologies influence the framing of terrorism by conducting quantitative and qualitative analysis of 421 articles by 10 U.S. conservative, center, and liberal media. Results show that conservative media’s terrorism framing reflects the dominant narrative espoused by the Republican Party, is comparatively more sympathetic to the right-wing terrorist and less sympathetic to the Islamic terrorist, and more broadly paints Islam and Muslims in a more negative light. In contrast, liberal media’s framing of terrorism does not reflect the dominant Republican terror rhetoric, shows higher sympathy towards the Islamic terrorist and apathy towards the right-wing terrorist relative to the conservative portrayal, and is comparatively less apathetic towards Islam, with occasional displays of compassion. Nonetheless, terrorism framing by conservative and liberal media does not considerably differ in a number of regards for which ideological differences could be expected to imply so.
iii
Acknowledgements
I would first like to express my sincere appreciation to my advisor Dr. Pauline Aarten as well as Ms. Jessica Sciarone for their unfailing support, assistance, and guidance throughout the whole process. I am particularly thankful for your patience and insightful feedback that you have given me throughout the whole thesis-writing process. I would also like to express a very special grati-tude to Roel Griep for his immeasurable contribution and support along the way. Finally, a men-tion goes to Jessie Johnson for believing in my project and helping me gain access to valuable resources.
iv
Table 1 Operationalization Framework ... 12 Table 2 Final Sample of U.S. Media with Cluster Categorization, Readership Size, and
Ideological Placement Index Score ... 15 Table 3 Distribution of Number of Articles across Main Topic Codes (Row Percentages
in Parentheses) ... 23 Table 4 Multinomial Logistic Regression Results for Main Topic Code Family with
GENERAL REPORTING as the Reference Category ... 24
Table 5 Number (Percent) of Articles Containing Incident Label, Perpetrator Label,
and Agency Codes ... 26 Table 6 Binomial Logistic Regression Results and Odds Ratio Estimates for Incident
Label, Perpetrator Label, and Agency Codes ... 27 Table 7 Number (Percent) of Articles for Victim Label Code Family; Charlottesville
Coverage ... 39 Table 8 Regression Results and Odds Ratio Estimates for Victim Group; Charlottesville
Coverage ... 40 Table A.1 Occurrences within Main Topic, Incident Label, Perpetrator Label, and Victim
Label Code Families ... 66 Table A.2 Final Codebook for Summative and Conventional Content Analyses ... 68 Table B.1 Two-Model Regression Results for the REPUBLICAN and DEMOCRATIC
v
1 Introduction ... 1
1.1 Research Question ... 2
1.2 Societal and Academic Relevance... 3
1.3 Reading Guide ... 4
2 Body of Knowledge ... 5
2.1 Terrorism & Media ... 5
2.2 Media Frames ... 6
2.3 Ideologies in Media ... 7
2.3.1 Theoretical Approaches ... 7
2.3.2 Conservatism and Liberalism in U.S. Media ... 8
2.4 Previous Empirical Research ... 9
2.4.1 Framing Terrorism in U.S. Media ... 9
2.4.2 Conservative & Liberal Media Frames ... 10
2.5 Summary ... 10
3 Methodology ... 11
3.1 Operationalization ... 11
3.2 Mixed Methods Research ... 13
3.2.1 Sampling of U.S. Media ... 13
3.2.2 Sampling of Terrorist Incidents ... 16
3.3 Methods ... 17 3.3.1 Qualitative Analysis... 17 3.3.2 Quantitative Analysis ... 19 4 Results ... 21 4.1 Main Topic ... 22 4.2 Incident Label ... 25 4.3 Perpetrator ... 30
4.3.1 Labeling the Perpetrator ... 30
4.3.2 Agency Attribution ... 33
4.3.3 Motivation & Character ... 35
4.4 Victims ... 37
4.4.1 Virtuous & Heroes ... 37
4.4.2 Victim Group ... 38 4.5 Broader Narrative ... 40 4.5.1 Future Threat ... 40 4.5.2 Radicalization ... 41 4.5.3 Islam ... 42 4.5.4 Political Agendas ... 43
vi
5.2 Limitations & Future Research ... 50
5.3 Concluding Remarks ... 52
References ... 53
1 “All I know is just what I read in the papers, and that’s an alibi for my ignorance.”
― Will Rogers (1978, p. 200)
1
1
Introduction
Contents
1.1 Research Question ... 2 1.2 Societal and Academic Relevance... 3 1.3 Reading Guide ... 4
When covering a terrorist event, media are confronted with various choices. They give meaning to events, ascribe moral judgements, and propose various countermeasures (Entman, 1993). In short, media choose to frame terror in a certain way. All of these choices have an impact on the audience, which perceives the reality through such frames (Tuchman, 1978). Specifically, different patterns of terrorism framing can alter the readers’ evaluation of the act, the perpetrator, or the government’s response. More troublingly, they can alter the audience’s preconceptions about ter-rorism on a broader level (Powell, 2011).
Political ideologies are one of the factors influencing the choices media make in respect to framing of terrorism. In fact, news coverage is no longer regarded as impartial, but rather implicitly or explicitly driven by various political agendas (Entman, 2007; Shor, Rijt, Ward, Askar, & Skiena, 2014). One might expect to find strong ideological differences especially in the coverage of terror-ism for multiple reasons. On the one hand, terrorterror-ism has long been thought of as a ‘Republican-owned’ issue, meaning that Republicans—and, in extension, conservatives—attach higher im-portance to it (Arbour, 2014). Perhaps most importantly, differences in framing by ideologically diverse media are likely to be shaped by the recent distancing of the mainstream conservative and liberal terrorism rhetoric terrorism. Since the arrival of Donald Trump, the dominant conservative depiction of terrorism would increasingly feature chauvinistic, islamophobic, and anti-immigration
2
sentiments—which are usually eschewed in the mainstream liberal terrorism discourse (Munestri, Adi, & Nurintan, 2017).
The urgency of analyzing terrorism framing by ideologically diverse media thus emerges particu-larly nowadays, when terrorism discourses in the United States are increasingly intertwined with questions of race, religion, or nationality. Since media can either echo the dominant ideology of the elite or serve as a platform for the dissemination of opposing views (Croteau & Hoynes, 2013), outlets face the decision to either counter or perpetuate the dominant depictions of terrorism. This paper addresses this issue by testing the relationship between political ideologies espoused by media and framing of terrorism.
1.1 Research Question
In realizing its research objective, this study poses the following question:
To what extent and how do the political ideologies of conservatism and liberalism influ-ence U.S. media’s framing of right-wing and Islamic terrorism committed on U.S. soil in 2017?
As shown, the research question aims to examine the specific ways in which ideologically diverse media frame terrorism differently. In respect to feasibility, the scope is limited to the framing of right-wing and Islamic terrorism, each represented by one incident.1 The research analyzes
cover-age in depth, looking for thematic elements, nuances, and meaning constructing the discourses of terrorism. To provide some direction to the research, a number of sub-questions are proposed:
SQ1: Do ideologically diverse media cover different aspects of a terrorist incident more
extensively?
SQ2: Are right-wing and Islamic terrorist incidents labeled differently between
ideologi-cally diverse media and how?
SQ3: Do ideologically diverse media portray right-wing and Islamic perpetrators
differ-ently and how?
SQ4: What are the differences in victim portrayal between ideologically diverse media?
SQ5: Are different types of terrorist incidents ‘politicized’ by ideologically diverse media?
SQ6: Is the future threat narrative presented differently between ideologically diverse
media and how?
SQ7: In what ways are right-wing and Islamic terrorism framed differently?
In answering the research question and its sub-questions, the thesis used a mixed methods design with a combination of quantitative and qualitative methods. It analyzed the media coverage of a
1 Throughout this paper, the terms ‘incident’ and ‘attack’ are used interchangeably, even though some semantic
3
vehicular terrorist attack in Charlottesville (right-wing perpetrator, August 2017) and in Manhat-tan (Islamic perpetrator, October 2017). Content analysis examined the coverage by a sample of 10 U.S. media published on the first five days since the attacks took place. The usage frequency of a specific phrase or element throughout coverage was analyzed though logistic regression models. Finally, quantitative results were substantiated by more in-depth, qualitative observations.
1.2 Societal and Academic Relevance
Mass media influence politics at an increasing pace. On the one hand, political institutions are being ‘mediatized’ in the sense that media shape matters of politics ever more (Mazzoleni & Schulz, 1999). On the other hand, media strongly influence the way people think about political issues (McCombs & Shaw, 1972). This is true especially in the American context, where an average citizen spends over 10 hours a day by consuming media (Austin, Barnard, & Hutcheon, 2016). This study’s societal relevance thus concerns the ability to elucidate that the failure to consume news from ideologically varied sources can inhibit a person’s ability to form well-rounded opinions on terrorism-related issues. This means that individuals exposed to one-sided coverage may be “severely lacking in a solid understanding of oppositional arguments” (Stroud, 2011, p. 10). In addition, one-sided news exposure further deepens political divisions in society. Political polariza-tion (the ideological distancing of Democrats and Republicans) has seen an increase in recent years in the United States (Hershey, 2017). Strongly polarized electorates are unable to distinguish between more and less substantive arguments, and favor frames endorsed by their party, regardless of the frame’s strength (Druckman, Peterson, & Slothuus, 2013). For instance, extreme counter-terrorism measures, such as Trump’s short-lived travel ban introduced in February 2017, are in situations with highly politically polarized society met with overwhelming support from conserva-tive electorates regardless of the policy’s soundness. Therefore, the elucidation of the specific ways in which ideologies shape the framing of terrorism is the first step towards a better-informed public, with a lesser degree of political polarization and a higher regard for sound policy arguments. Academically, the proposed research aims to bridge an important gap in literature. On the one hand, the mutual relationship between media and terrorism has been explored in a myriad of studies (Kellner, 2015; Nacos, 2016; Norris, Kern, & Just, 2003; Picard, 1993; Schmid & Graaf, 1982; Weimann & Winn, 1994). Empirical research into framing of terrorism by media is equally abundant. Studies have, for instance, established that terrorism coverage by U.S. media over-whelmingly focuses on Islam in general and Muslim attackers in particular (Kearns, Betus, & Lemieux, 2017), presents primarily violent responses to conflict (Hallowell, 2016), and portrays domestic terrorism as a lesser threat than international terrorism (Powell, 2011). On the other hand, despite the richness of literature on the relationship between terrorism and media, studies into the differences between conservative and liberal media’s terrorism framing remain very lim-ited. By explaining the role of political ideologies in portraying terrorism, the research bridges a
4
gap in the field of terrorism studies as well as media studies. Additionally, the study enriches the field of security studies, within which terrorism research has become one of the dominant strands since 9/11.
1.3 Reading Guide
The rest of the thesis proceeds as follows. The ensuing section introduces the theoretical founda-tions of this study. The section discusses the complex relafounda-tionship between media and terrorism; shows how media construct frames in terrorism coverage and points to the process in which such frames influence public opinions; it lays the foundations for the differences between ideologically diverse media by a discussion of the moral foundations of conservatism and liberalism; and finally, discusses previous empirical work. Next, the methodological section introduces and justifies the choice of the research design and methods. It proposes a mixed methods research with a combi-nation of quantitative and qualitative methods, which point to both the extent and the specific ways in which political ideologies influence terrorism framing. The consecutive results section presents the main findings, sectioned by five main themes found in coverage: main topic, incident label, portrayal of the perpetrator, portrayal of the victims, and narrative. The section ends with a summary of the main quantitative findings. The thesis concludes with a discussion of the find-ings, a brief deliberation on its limitations and recommended directions for further research, and finally, it provides a few concluding remarks.
5 “Knowledge is a weapon. I intend to be formi-dably armed.”
― Terry Goodkind (2015, p.141)
2
2
Body of Knowledge
Contents
2.1 Terrorism & Media ... 5
2.2 Media Frames ... 6
2.3 Ideologies in Media ... 7
2.3.1 Theoretical Approaches ... 7
2.3.2 Conservatism and Liberalism in U.S. Media ... 8
2.4 Previous Empirical Research ... 9
2.4.1 Framing Terrorism in U.S. Media ... 9
2.4.2 Conservative & Liberal Media Frames ... 10
2.5 Summary ... 10
The following chapter presents the most relevant literature upon which the study builds. It is sectioned into five main themes. The first one introduces the interaction between terrorism and media in its complexity. The second section zooms into the different ways in which media can portray or frame an issue. Third, ideologically-shaped media are discussed in the context of the United States. The fourth section looks at the previous research and its relation to the literature gap addressed here. Finally, the last section provides a brief summary of the whole chapter and delineates the theoretical foundations of this research.
2.1 Terrorism & Media
“Communication is at the heart of terrorism”—this is how two scholars opened their book (Seib & Janbek, 2010, p. 1). They further claimed that terrorists’ “principal accomplishment” is “not killing … people, but rather terrifying millions more through the reports and images” which are
6
conveyed by the media (p. 1). Seib and Janbek thus called terrorism “a communicative act” (p. 1). Many other scholars have argued for the importance of media in the attainment of terrorists’ objectives (e.g., Castells, 2010; Laqueur, 2004; Nacos, 2016). The importance of the communicative function of terrorism is what Nacos (2016) calls mass-mediated terrorism, defined as “political violence against civilians and non-combatants committed with the intention to publicize the deed, to gain publicity and thereby public and government attention” (p. 10, emphasis in original). A more critical evaluation comes from Picard (1993), who claimed that while media coverage remains an important aspect to the terrorism phenomenon, it is not indispensable.
The media also maintain a mutually beneficial relationship with the terrorists. Acts of terror intrinsically have a high news value due to the great threat to human security and “media’s … appetite for horrific images” (Nacos, 2016, p. 33). Nevertheless, scholars remain divided on the extent to which media coverage provides legitimacy to terrorist actors and encourages ‘copycat’ terrorist attacks elsewhere (Norris, Kern, & Just, 2003).
2.2 Media Frames
In reporting on terrorism-related stories, media consciously or unconsciously choose a specific angle through which they portray the issue at hand. In that way, media present a lens for their audience through which to perceive reality—in other words, media socially construct reality (Tuchman, 1978). Two approaches prove useful in examining this phenomenon: agenda-setting and framing. Agenda-setting theory maintains that the weight media attribute to a given topic has an effect on the perceptions of the topic’s importance by the audience (McCombs & Shaw, 1972). By reporting on some issues more extensively than others, media set public agendas by ascribing various levels of importance to subjects. This concerns, for instance, the volume and length of articles published on a given issue. On the other hand, framing theory (Entman, 1993; Goffman, 1974), sometimes referred to as the second-level agenda setting theory, suggests that there exist various viewpoints on a given issue and introduces the concept of framing as a “process by which people develop a particular conceptualization of an issue or reorient their thinking about an issue” (Chong & Druck-man, 2007, p. 104). According to Entman (1993), “To frame is to select some aspects of a perceived reality and make them more salient in a communicating text” (p. 52). Framing thus provides meaning to certain events (Gamson & Modigliani, 1987) and endorses specific understandings of political issues over others (Shah et al., 2002). In sum, as Balmas and Sheafer (2010) put it, the first level agenda setting concerns media telling the audience ‘what to think about’ versus the second-level agenda setting prescribing ‘how to think about’ an issue.
The process of framing results in media employing various news frames in their coverage. Such frames can, for instance, range from strategic, game, to conflict frames, or what Iyengar (1991) distinguishes as episodic and thematic frames. The mainstream, conventional frames (of terrorism)
7
“furnish consistent, predictable, simple, and powerful narratives that are embedded in the social construction of reality” (Norris, Kern, & Just, 2003, p. 6). Employing a specific terror frame means that the media can not only portray the events in a certain way, but also, among other things, alter the way blame is assigned and promote preferred interpretations and judgments of the gov-ernment’s policy. Due to the well-documented influence of mass media on public opinion formation (Graber & Dunaway, 2014; Jamieson & Campbell, 2000; McCombs & Shaw, 1972; Parenti, 1987), frames can alter the audience’s perception of the specific attack and attacker as well as the ter-rorist threat in general. When conducting a study of media, the specific frames or framing patterns can be discerned from the use of “key words, metaphors, concepts, symbols, and visual images emphasized in a news narrative” (Entman, 1991, p. 7).
2.3 Ideologies in Media
There are numerous factors which influence the framing process of media. Each outlet has distinct traditions, routines, identity, political leaning, as well as political agendas. All of these shape the way a medium works and how much and what kind of content it publishes about a certain issue (Shor et al., 2014). The influence of political ideologies on news frames has been confirmed in a bulk of previous research (Edelman, 1993; Scheufele, 1999; Shoemaker & Reese, 1996; Tuchman, 1978). In fact, Gitlin (1980) claimed that “the mass media have become core systems for the distribution of ideology” (p. 2). To understand the role ideology plays in news frames creation, it is important to first explore the dynamics within which ideologically-shaped media operate.
2.3.1 Theoretical Approaches
Numerous theoretical frameworks point to the strength of the role that political ideologies play in influencing media coverage. For instance, Chomsky and Herman’s (2011) propaganda model takes account of the media-specific institutional and relational factors and argues that media are prop-aganda tools for the powerful elite that funds them and, consequently, promotes specific societal objectives through such media. At the very fundamental level, the dominant ideology of the elite sustained in the U.S. media favors “capitalism, patriarchy, heterosexism, individualism, consum-erism, and White privilege” (Entman, 2007, p. 170).
Next to the fundamental contemporary dogmas, also more specific, such as political, ideologies are perpetuated in news coverage. In trying to understand the influence of such ideologies, Shoemaker and Reese’s (1996) theory of news content, later revised as the hierarchy of influences model (Shoemaker & Reese, 2014), proves valuable. This framework hierarchically organizes five levels of influences on media coverage from macro to micro-level: social systems, social institutions, media organizations, routine practices, and individuals (p. 9). The social systems level (previously called ‘ideological’ level) corresponds to the influence that (political) ideologies have on news frames, and explains “how the national and cultural context affect media performance” (p. 8).
8
In the original research, Shoemaker and Reese (1996) provide a set of hypotheses on how ideology shapes media content: (i) more deviant stories gain more coverage; (ii) reporting on other coun-tries’ stories intensifies with increased levels of the foreign country’s significance; (iii) deviant behavior is likely to be reported non-objectively; (iv) when ideological tenants are endangered, media will face opposition from the elite; (v) tighter ties of media with elites will translate into more consistent reflection of the elite’s ideologies; (vi) anomalies to the ideology need to be cor-rected; and (vii) TV coverage generally professes more ideological slant than print coverage (pp. 260-261). These hypotheses provide a basic understanding of the ways in which ideologies shape news coverage.
2.3.2 Conservatism and Liberalism in U.S. Media
Partisan media were long a norm in the United States, primarily a result of being partially funded by either political party. In the beginning of the twentieth century, however, partisan reporting declined due to higher media professionalism, commercial factors (such as the changes to the donor structure and the decline of party-funded newspapers), and external influences (for instance, the Cold War diminished party divisions in face of a greater enemy). Only in 1960s and 1970s did partisan reporting resurface in the United States, when the Vietnam War deepened cross-party differences (Baughman, 2011). In addition, the unpopularity of many contemporary liberal pro-jects or their byproducts (such as the failing Vietnam War, the anti-war movement, or the Black Power movement) meant that liberalism was losing the position as a discourse-defining ideology which it held in the 1950s. Thus, conservatism and liberalism slowly established themselves in the American media without any one being much more dominant (Alterman, 2010).
The reason why conservatism and liberalism are particularly interesting to analyze in the context of terrorism framing lies in their adherence to a different set of values, as explained by Haidt and Graham’s moral foundations theory (Graham, Haidt, & Nosek, 2009; Haidt & Graham, 2007). The theory proposes five moral foundations: harm/care, fairness/reciprocity, in-group/loyalty, authority/respect, and purity/sanctity. It stipulates that while conservatives rely on each of the foundations, liberals value primarily the first two foundations.
The moral foundations theory helps show that protecting issues relating to care and justice are much more important for liberals than for conservatives. As a result, liberals value open-minded-ness and tolerance more, whereas conservatives are less tolerant and trusting, and more dogmatic and risk-averse (Greenberg et al., 1992; Stone, 1980). Additionally, the three purely conservative foundations were found to be positively correlated with antagonism between groups, hence con-servatives discriminate against others more (Kugler, Jost, & Noorbaloochi, 2014).
In cases of a terror threat, conservatism’s appeal is connected to its emphasis on “resistance to change and justification of inequality,” and its ability to “manage uncertainty and threat” better (Jost, Glaser, Kruglanski, & Sulloway, 2003, p. 339). A study by van de Vyver, Houston, Abrams,
9
and Vasiljevic (2016) showed that in the aftermath of a powerful terrorist attack, people become more conservative. That is because the threat to life leads people to value in-group loyalty more and fairness less. The change in endorsement for these foundations leads to, for instance, rising islamophobia in after an Islamic terror attack. Conservative terrorism coverage is therefore likely to be more unsympathetic to out-group terrorists (with justified inequality and injustice) than liberal media.
2.4 Previous Empirical Research
Several studies in the past have undertaken research objectives related to those delineated by this paper. A discussion of such studies can help situate the study’s findings within a broader academic context.
2.4.1 Framing Terrorism in U.S. Media
Numerous scholars have analyzed how U.S. media frame terrorist attacks. Reese and Lewis (2009) argue that the War on Terror (WoT) frames employed by U.S. media vary per type of source. While TV channels have adopted the dominant political frame of the WoT, print media provided a more nuanced coverage. Nevertheless, Reese and Lewis describe the WoT frame presented by the incumbent administration as a powerful organizing principle in the news.
A more in-depth discussion of the media’s terrorism portrayal is found in a report for the American Friends Service Committee, in which Hallowell (2016) analyzed over 600 articles relating to ter-rorism. She found that (i) terrorism coverage overwhelmingly couples Islam and terrorism, with Muslims being portrayed as a homogenous violent group; (ii) extremist groups are portrayed as rational and irrational actors with little consistency; and (iii) violent responses to terrorism are cited more frequently that non-violent responses, obscuring the complexity of counter-terrorism measures.
Similarly, Kearns, Betus, and Lemieux (2017) asked why certain terrorist attacks receive higher amounts of coverage than others, and on the basis of their content analysis of 2,413 articles identified several ‘predictors of media coverage’ in relation to terrorism: background of the at-tacker, type of target, number of casualties, and whether the attacker was arrested. Kearns et al. found that the background category proved to have the most influence on media coverage, where attacks committed by Muslims received 449% more coverage than attacks by non-Muslims. Further, Powell (2011) analyzed U.S. news coverage of several terrorist attacks and found that while domestic terrorism is often portrayed as a minor threat, mostly caused by mental illness, international terrorism is seen as a much bigger threat, with perpetrators being portrayed as vengeful, evil individuals. Neither of these four studies, however, included partisanship or political ideologies as a factor influencing coverage.
10
2.4.2 Conservative & Liberal Media Frames
Numerous studies have empirically analyzed differences between conservative and liberal media in the framing of certain issues, including ageism (Marier & Revelli, 2017), climate change (Dotson, Jacobson, Kaid, & Carlton, 2012; Feldman, Maibach, Roser-Renouf, & Leiserowitz, 2012), eco-nomic news (Larcinese, Puglisi, & Snyder, 2011), female subjects (Shor et. al, 2014), gay marriage legitimization (Pan, Meng, & Zhou, 2010), migrants (Greenberg, 2000), or poverty (Kim, Car-valho, & Davis, 2010).
Most notably, in her research on the differences in portraying the 2004 presidential campaign, Stroud (2011) noticed that the weight attributed to the issues of terrorism and the Iraq War by conservative media was consistent with the Republican nominee’s rhetoric and by liberal media with the Democratic nominee’s rhetoric. Stroud further found that subscribers to liberal media were less likely to identify terrorism as a significant national threat than subscribers to conserva-tive media. Beyond Stroud’s research, empirical analyses of terrorism framing by U.S. liberal and conservative media often center on a single case study and are published in non-academic news-papers or blogs (e.g., Friedersdorf, 2013; Rodrigues, 2013). This paper aims to bridge this literature gap by systematically examining terrorism coverage by 10 U.S. conservative and liberal outlets.
2.5 Summary
Media often maintain an important, mutually-beneficial relationship with terrorists. Terrorists need their deeds to be publicized in order to spread fear; media value bombastic news elements which terrorist attacks intrinsically offer. There are various ways in which media can portray, or frame, terrorism. Framing terror in a certain way introduces a conceptual lens through which reality is perceived and meaning is constructed—and, essentially, has an impact on the perceptions of the public.
Political ideologies are a crucial element influencing frame construction. Shoemaker and Reese’s (1996) theory of news content showed that dominant ideologies are perpetuated and defended by the media. This means that the varying tenants of conservatism and liberalism influence terrorism coverage differently. Liberals value open-mindedness, tolerance, and trust; conservatives are more dogmatic, risk-averse, but also antagonistic and discriminative against others. Correspondingly, conservative terrorism coverage is expected to stress in-group loyalty and out-group animosity (especially rising islamophobia) more than terrorism coverage by liberal media.
Nonetheless, existing empirical studies on the relationship between media and terrorism fail to provide a thorough analysis of framing patterns different across ideologically diverse media. This research analyzes the differences in framing Islamic and right-wing terrorism by conservative, center, and liberal media. The different patterns of framing are captured by an examination of themes, labels, connotations, and associations found in news coverage.
11 “Don’t be a novelist; be a statistician. Much more scope for the imagination.”
— Darrell Huff (1954, p. 2)
3
3
Methodology
Contents
3.1 Operationalization ... 11
3.2 Mixed Methods Research ... 13
3.2.1 Sampling of U.S. Media ... 13
3.2.2 Sampling of Terrorist Incidents ... 16
3.3 Methods ... 17
3.3.1 Qualitative Analysis... 17
3.3.2 Quantitative Analysis ... 19
The methodological chapter operationalizes important concepts, introduces the chosen research design, and, most importantly, provides an overview of and justification for the choice of the research methods used in this research.
3.1 Operationalization
In order to examine the extent and the ways in which the political leaning of media influence the framing of terrorism, several concepts need to be operationalized. Table 1 presents the operation-alization framework for the concepts of terrorism framing and the political leaning of media.
12 Table 1
Operationalization Framework
Concept
Definition
Indicator
Data Source
Terrorism framing
The “selection, emphasis, and ex-clusion that furnish a coherent in-terpretation and evaluation of [terrorist] events” (Norris, Kern, & Just, 2003, p. 5)
The use of specific conceptual elements or themes when covering terrorism (see Table A.2)
Content analysis of selected me-dia’s coverage of two terrorist at-tacks
Political leaning of media
The tendency of a medium to re-flect either conservative or liberal viewpoints, or the lack thereof
‘Ideological placement’ index, based on the ideological composition of media’s audience (see Table 2)
Research on polit-ical polarization and the media (Mitchell et al., 2014)
In addition to the clear demarcation of these two concepts, choosing an appropriate definition of terrorism is crucial, as the nature of the definition has a direct impact on the choice of attacks as case studies. The study adopts the official definition of terrorism specified in the Crimes and Criminal Procedure title of the U.S. Federal Criminal Code:
1. the term “international terrorism” means activities that—
a. involve violent acts or acts dangerous to human life that are a violation of the criminal laws of the United States or of any State, or that would be a criminal violation if committed within the jurisdiction of the United States or of any State; b. appear to be intended—
i. to intimidate or coerce a civilian population;
ii. to influence the policy of a government by intimidation or coercion; or iii. to affect the conduct of a government by mass destruction, assassination,
or kidnapping; and
c. occur primarily outside the territorial jurisdiction of the United States, or transcend national boundaries in terms of the means by which they are accomplished, the persons they appear intended to intimidate or coerce, or the locale in which their perpetrators operate or seek asylum;
2. the term “domestic terrorism” means activities that—
a. involve acts dangerous to human life that are a violation of the criminal laws of the United States or of any State;
b. appear to be intended—
i. to intimidate or coerce a civilian population;
ii. to influence the policy of a government by intimidation or coercion; or iii. to affect the conduct of a government by mass destruction, assassination,
13
c. occur primarily within the territorial jurisdiction of the United States. (Definitions, 2011)
The study adopts the definition because this specific wording forms the legal basis upon which terrorism investigations can be commenced in the United States and thus usually guides authori-ties in presenting dominant interpretation of (terror) attacks. It should be noted that the U.S. code definitions do not specify that the direct target of violence is different from the actual target audience of terrorism—an aspect of many academic definitions of terrorism.2
Finally, right-wing terrorism is throughout the paper understood as an act of international or domestic terrorism as specified in the U.S. Code, conducted by a perpetrator espousing far-right beliefs; Islamic terrorism is understood as an act of international terrorism as specified in the U.S. Code, conducted by a perpetrator espousing extremist Islamic ideology. The reason why Islamic terrorism is understood only as international is because Islamic extremism inherently transcends U.S. boundaries.
3.2 Mixed Methods Research
The relationships between political ideologies and coverage of terrorism is in this thesis primarily studied quantitatively, but the statistical findings are either complemented or substantiated by qualitative analysis. In situations when “a second method is needed to enhance a primary method,” mixed methods research (MMR) is the most appropriate design to employ (Creswell and Clark, 2011, p. 8). By combining quantitative and qualitative methods, MMR can help identify general patterns in reporting by conservative, center, and liberal media, as well as point to the nuances and other qualitative indicators found in coverage. In this sense, the qualitative research “pro-vides[s] an enhanced understanding” (Creswell & Clark, 2011, p. 10) of the quantitative research. In realizing its research objectives, the thesis analyzes a selection of articles published by 10 U.S. media outlets covering one right-wing and one Islamic attack committed on U.S. soil in 2017. The coverage is analyzed by content analysis and interpreted by regression analysis. The following sections discuss sampling methods for both the media and the attacks.
3.2.1 Sampling of U.S. Media
The sample of U.S. media is purposively selected and is based on Pew Research Center’s study on political leanings of (predominantly U.S.) media (Mitchell et al., 2014; Pew Research Center, 2014). Although there exist various measures of the political inclination of U.S. media—Gentzkow
2 Schmid and Jongman (1988) analyzed 109 academic conceptualizations of terrorism for common themes and proposed
an academic-consensus definition. The definition included a clause specifying that terrorism is a method “whereby—in contrast to assassination—the direct targets of violence are not the main targets” (p. 28). This understanding is still being recognized by academics as a useful tool in interpreting terrorism (Martin, 2012; Schmid & Jongman, 2017; Williamson, 2016).
14
and Shapiro’s (2010) slant index; DBKP’s top 100 conservative and liberal media (Skaggs, 2010a, 2010b), or New Revere Daily Press’ (2017) ranking of top 100 conservative outlets—the Pew study is selected for it is the most up-to-date ranking of conservative, center, and liberal U.S. media. It analyzes ideological composition of 32 media’s audiences, and provides a 10-point scale of catego-rization of media along political ideologies (Figure 1). Pew’s list of media was in respect to feasi-bility filtered in this thesis, and the final sample excluded foreign newspapers, TV channels, radio stations, TV/radio shows, news aggregators, news agencies, and purely print outlets. It should be noted that for the small number of conservative outlets in the filtered sample, Fox News was added back as the most popular and trusted medium among conservatives (Mitchell et al., 2014).
Figure 1. Pew Research Center’s Ideological Placement of Media’s Audiences (Mitchell et al., 2014, p. 9)
The final sample consists of 10 U.S. newspapers, magazines, and multiplatform news networks, presented in Table 2. The table shows each outlet’s respective readership size as determined by Pew, its ‘ideological placement’ index, assigning higher score to conservative audiences. Further, the table show that the media outlets were categorized into a conservative, center, and liberal cluster. Note that instead of three equal intervals on the scale of the index, the categorization refers to a ranked distribution, arranging the media into intervals of 40%, 30%, and 30% according to their absolute rank. Although the categorization separates media positioned very closely to each other on the index scale into two different clusters—while HuffPost (index of -3.3) is categorized as liberal, Washington Post (-3.3) and Politico (-3.2) fall within the center cluster—the grouping
15
is particularly effective in differentiating these media based on their character. HuffPost is con-sidered a liberal medium for it is a novel, digital outlet; Washington Post and Politico are consid-ered center media for they are more traditional media with a long history of print journalism, and thus with less ideological bias expected. Hereafter, the three clusters are simply referred to as conservative, center, and liberal media.
Quantitative analysis is conducted on the level of the categorized media clusters. There are mul-tiple reasons why this study does not measure the relationship between the ideological placement index and media coverage directly. On the one hand, there are reasons not to regard the index as an absolute, objective measure of the leaning of each source. First, the index is a proxy measure, assessing the leaning of audiences and not media themselves. Second, the Pew study is already four years old, and shifts in media’s ideological leaning are expected to have occurred. And third, the index scores are only extracted values based on data in Figure 1, since the full data was not publicly available.3 Nonetheless, the index is a useful approximation of the media’s ideological
leaning and serves as a guideline for the formation of media clusters. A validation of the approach to consider media clusters as the unit of analysis is further discussed in Appendix B. On the other hand, cluster categorization provides three statistical advantages. First, the categorization allows for non-linearity in the relationship between coverage and the ideological placement index. This
3 Despite personal communication with the staff of the Pew Research Center over the course of over a month, Pew was
not able to deliver the requested data before the deadline.
Table 2
Final Sample of U.S. Media with Cluster Categorization, Readership Size, and Ideological Placement Index Score
Media Cluster Medium Readership (%) Ideological Placement Index
Liberal Slate 3 -5.7
New York Times 13 -4.3
BuzzFeed 4 -3.8
HuffPost 13 -3.3
Center Washington Post 8 -3.3
Politico 4 -3.2
Wall Street Journal 10 -0.6
Conservative Fox News 39 1.9
Breitbart 3 5.4
The Blaze 5 5.7
Notes. Readership corresponds to the percentage of respondents consuming their news about politics or government from that source. Ideological placement index ranges from -10 = consistently liberal viewpoints to +10 = consistently conservative viewpoints
16
is particularly important as the research question does not presuppose a linear relationship be-tween ideological leaning and framing of terrorism, but merely asks whether are there differences between ideologically dissimilar media in framing terrorism.4 Second, inferences on the direct
re-lationship between the ideological placement index and media coverage are easily affected by only a single outlet not adhering to the general pattern. Third, by studying groups of media instead of outlets separately, the number of observations per unit of analysis considerably improves, increas-ing the robustness of the study’s findincreas-ings.
3.2.2 Sampling of Terrorist Incidents
For feasibility issues, only one Islamic and one right-wing attack was analyzed: a vehicular attack in Charlottesville (right-wing perpetrator, August 12, 2017) and a vehicular attack in Manhattan (Islamic perpetrator, October 31, 2017). The attack in Charlottesville refers to the incident when James Alex Fields Jr. crashed his car into a crowd of counter-protesters at a white supremacist rally in Charlottesville, Virginia. Fields killed one woman and injured over a dozen. Two more fatalities were attributed to the attack when a helicopter crashed, killing two police officers called to help quiet the tensions around the crime scene. The Manhattan attack (also called the New York attack, NYC attack, or Halloween attack) occurred when Sayfullo Saipov crashed a rented truck into cyclists and pedestrians in Manhattan, killing eight and injuring dozens.
The two cases were selected purposively for three reasons. First, they share similar characteristics: they are both car-ramming attacks, they occurred within less than three months from each other, and incurred a similar casualty count. Analyzing the coverage of two similar terrorist attacks is necessary to minimize the possible impact of incident-specific attributes on coverage. Second, analyzing car-ramming attacks is interesting from an academic point of view. This terror tactic has recently been gaining prominence world-wide and forms the pillar of ‘low-tech, high-impact’ terror attacks, which analysts sometimes see as the new wave of terrorism (Bergen, 2016; Briskey, 2016; Clarke & Klarevas, 2017; Crabtree, 2017; Hoffman, 2014; Zoli, 2017). And third, the selection of Charlottesville is further justified by it being the only right-wing terrorist attack on U.S. soil with at least one fatality in 2017 at the time of writing. Such attacks are generally more important for national security, and are thus expected to warrant not only broader, but also deeper and richer media coverage.
4 Results presented in Figure 4 confirm these non-linearities, showing that the center cluster used a variety of labels
17
3.3 Methods
The study employed a combination of quantitative and qualitative methods, corresponding to three main methodological approaches: qualitative content analysis, quantitative content analysis, and regression analysis, discussed in detail in the following subsections.
The study analyzed online coverage of 10 U.S. media outlets. Articles were accessed primarily through Factiva; articles not accessible there (Fox News, Slate, Breitbart) were accessed through the media’s respective archives. To ensure that every relevant article was found, keywords per-taining to the location of the attack (‘Charlottesville’; ‘Manhattan’ and ‘New York’) were used in the search, rather than using keywords like ‘terrorism,’ ‘terrorist,’ ‘attack,’ or ‘incident.’ This method is best at eliminating the risk of overlooking articles due to inappropriate or incomplete keyword search.
3.3.1 Qualitative Analysis
The qualitative part of the thesis consisted of content analysis, which is used to “interpret meaning from the content of text data” (Hsieh & Shannon, 2005, p. 1277). It has been the preferred method for explorations into media-related issues (Boykoff & Boykoff, 2007), as evidenced by its use in a number of studies on the framing of terrorism by media (Cho et al., 2003; Hallowell, 2016; Kearns, Betus, & Lemieux, 2017; Powell, 2011).
Specifically, the qualitative part of the content analysis followed a combination of conventional and directed approach to content analysis in Hsieh and Shannon’s (2005) conceptualization. The combination of the two approaches translates into the reliance on predetermined as well as emer-gent codes. Such codes are the central organizing unit of the qualitative coding procedure, which point to important themes and conceptual elements in the coverage. Predetermined codes are created a priori and provide a direction to the analysis; emergent codes evolve from the text and are thus created ad hoc.
Pilot Study
Prior to commencing the analysis, a pilot study was conducted to determine the exact time span of coverage to be analyzed and to validate a set of predetermined codes. In the first step, the pilot randomly selected an outlet (Politico) and examined the volume of its Charlottesville attack cov-erage over the first month after the attack. Based on a headline analysis, the pilot study excluded articles unrelated to the attack. The results show that the biggest volume of articles was published in the first two weeks immediately after the attack, whereas the first bigger decrease in coverage occurred already after five days (Figure 2).
The pilot study confirms that news stories naturally decay. This finding is consistent with previous research; in general, stories in media live for 12-120 days (Watt, Mazza, & Snyder, 1993). In fact, many studies analyzing a specific event limit their time period of analyzed media articles to just
18
several days, for instance, five (Chadwick, 2011) or six (Fahmy, Kelly, & Kim, 2007) days for episodic events, and thirty for more thematic stories (An & Gower, 2009). For this reason, the study limited its time span to only the first five days after an attack. In addition to increasing the study’s feasibility, this approach is expected to capture the biggest differences in coverage between conservative and liberal media. This stems from the hypothesis that in the initial stages of an attack, the lack of information grants media greater space for framing the attack asymmetrically than in later stages of the attack.
Figure 2. Volume of Charlottesville Coverage by Politico over a One-Month Period
The second step of the pilot study was to validate a set of preconceived codes created deductively. For that purpose, the first 10 articles published about the Charlottesville attack by a random sample of outlets (Slate, Politico, New York Times) were coded. The themes identified by the pilot were compared to the main codes identified by two studies on terrorism coverage in the United States (Hallowell, 2016; Powell, 2011). Ultimately, the deductive process of code creation directed the content analysis to focus on the framing of four main themes: perpetrator, future threat, extremist groups, and victims.
Coding Procedure
During the coding process, the volume of articles published on the first five days since the attack took place were further filtered based on their content. All articles where the attack was mentioned only in relation to a more important storyline were eliminated; the only instances considered in the final sample were those where the attack was the immediate impetus for some broader societal or political discussions (such as gun regulations), essentially pointing to the ‘politicization’ of the attack. Further, transcripts, interviews, daily newsletters, briefings, blogs, opinion pieces, and personal commentaries from entities other than official authorities were removed from the sample.
0 2 4 6 8 10 12 14 16 18 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 1 2 3 4 5 6 7 8 9 10 11 August September A rti cl e Co un t
19
This decision places the focus on news reporting rather than opinion pieces in order to examine the framing of terrorism employed by media in what is considered as ‘factual’ news reporting. Finally, all emergent codes identified by the conventional content analysis were manually coded using Atlas.ti. The whole coding process was performed by a single coder, eliminating the necessity to account for inter-coder reliability. The codes were further organized into code families and are presented in the final codebook table in Appendices (Table A.2).
3.3.2 Quantitative Analysis
The quantitative part of this research is composed of quantitative content analysis and logistic regression analysis. The former follows summative approach to content analysis in Hsieh and Shannon’s (2005) conceptualization, which aims at “identifying and quantifying certain words or content in text … not to infer meaning but, rather, to explore usage” (p. 1283). The summative content analysis here studied the usage frequency of specific codes by selected media, and thus enabled the study to gather quantifiable data. The codes gathered by summative content analysis (such as labels attached to the attack or the perpetrator) were equally tracked in Atlas.ti, either manually or semi-automatically. To ensure that only those elements pertinent to the analysis were coded, fully automatic coding was not conducted. Finally, the summative content analysis facili-tated the creation of a dataset subsequently analyzed by other quantitative research methods. The unit of observation in this study’s dataset is a U.S. media article, mainly studied in a dichot-omous way: essentially, the analysis asked whether the article used a specific code, not how many times the article did so. This statistical design was primarily chosen to eliminate the effect of article length and the author’s bias towards repetitive writing on this study’s statistical inferences. Additionally, dichotomous classification of the unit of observation enabled the researcher to adopt widely accepted statistical techniques such as contingency tables and logistic regressions. The use of conventional techniques strongly improves replicability and ensures validity of conclusions drawn from this research while avoiding the need for external model validation. The main unit of analysis in this dataset is a U.S. medium (cluster); a secondary unit of analysis is a terrorist incident.
This organization of data was then studied through contingency tables and regression analysis, both performed in SPSS. The appeal of contingency tables is that they provide a first, easily interpretable outcome on the main effect studied in this research. Regressions, however, have the advantage of controlling for the effect of other variables (or covariates) which, if omitted from the statistical analysis, would bias the inference about the relation between the studied variables. For this reason, regression analysis has also become “an integral component” of efforts to analyze a relationship between two or more variables (Hosmer, Lemeshow, & Sturdivant, 2013, p. 1). Since the outcome variable here relates to whether a certain code is or is not mentioned in an article, it is dichotomous. Over the last decades, logistic regression has become the preferred and standard
20
model among researchers to study the effect between one or more explanatory variables and a dichotomous outcome variable (Hosmer et al., 2013). As a result, logistic regression best fits the dataset in this study and was thus used to provide the main statistical results in this study. More specifically, the study relied on two types of logistic regressions: binomial logistic regression and multinomial logistic regression. Binomial logistic regressions were used for the majority of the statistical analyses: when the outcome variables stemmed from codes whose presence did not imply the presence or absence of another code in the same code family, and whose presence thus could be studied independently from the presence of other codes. Examples are codes that refer to various labels (e.g., the TERRORISM and VIOLENCE codes of the ‘incident label’ code family), as
multiple codes in the same family could be used in the same article. To the contrary, multinomial logistic regression was used to study a set of codes whose occurrences in a single article were perfectly correlated. Specifically, the presence of one of these codes in an article excluded the presence of the other codes in the same family, and where one of the codes was always present. The only such instance in this study was the code family ‘main topic,’ where each article was categorized into exactly one of the four codes.
Furthermore, the regression analysis examined the effect of the media clusters as the main explan-atory variable, but also controlled for the effect of the incident, the interaction between media clusters and incident, and the number of days since the event. The incident control variable controlled for the effect that the occurrence of specific codes was correlated with the incident. The interaction variable (or moderator) controlled for the difference in relation between media clusters and the occurrence of a specific code between incidents. The day of reporting variable controlled whether the time that had passed since the event took place influenced the differences in media coverage. As the goal of the analysis was to find the model that best explains the relationship between the code occurrences and the media clusters, variable selection was performed on the explanatory variables other than the media cluster. The method for variable selection was back-ward elimination, where variables were excluded for p-values above .10; the reason for this non-standard p-value is based in the small number of observations in the study. For the same reason, any occurrences of p-values between .05 and .10 are throughout the results section reported as weakly significant.
21 “If you torture the data long enough, … [it] will always confess.” ― Ronald Coase (1988, p. 74)
4
4
Results
Contents
4.1 Main Topic ... 22 4.2 Incident Label ... 25 4.3 Perpetrator ... 30 4.3.1 Labeling the Perpetrator ... 30 4.3.2 Agency Attribution ... 33 4.3.3 Motivation & Character ... 35 4.4 Victims ... 37 4.4.1 Virtuous & Heroes ... 37 4.4.2 Victim Group ... 38 4.5 Broader Narrative ... 40 4.5.1 Future Threat ... 40 4.5.2 Radicalization ... 41 4.5.3 Islam ... 42 4.5.4 Political Agendas ... 43 4.6 Summary of the Main Quantitative Results ... 46The final sample consists of 421 online news articles published in the first five days since the attack took place. The majority of articles covered the Charlottesville attack (n = 240, 57%); 181 articles covered the Manhattan attack (43%). The sample includes a rather equal distribution of coverage by media clusters, with 157 conservative articles (37%), 135 center articles (32%), and 129 liberal articles (31%). On average, media published approximately 21.1 articles per attack (SD = 11.5, range: 6–47). Figure 3 shows that the majority of the coverage was published on the first day after the attack (n = 137, 33%), closely followed by the second day after the attack (n = 135,
22
32%). Conservative media published most articles on the first day after the Manhattan attack (n = 37); center media on the second day after Charlottesville (n = 35); and liberal media on the first day after Charlottesville (n = 22).
The results section is structured along five main themes found in the coverage: main topic, incident label, portrayal of the perpetrator, portrayal of the victims, and broader narrative. Throughout the chapter, results are primarily reported quantitatively, whereas qualitative results provide con-text to the statistical research outcomes. In addition, several themes are reported purely qualita-tively. Note that throughout the section, mentions of any codes analyzed statistically were capi-talized. A summary of the main quantitative results is reported in Section 4.6.
4.1 Main Topic
Each of the 421 articles was categorized into one of four main topic categories (codes): BIGGER DEBATE, GOVERNMENT RESPONSE, GENERAL REPORTING, and ANALYSIS (see Appendix A for the
rationale for this categorization). The distribution of the main topics shows which topics media presented as more salient to their readers; one might expect that media espousing different political ideologies would highlight different issues as more important.
Figure 3. Final Sample’s Coverage Time Span, per Incident and Media Cluster
0 5 10 15 20 25 30 35 40 d d + 1 d + 2 d + 3 d + 4 A rticle Co un t Day of Reporting Charlottesville; Conservative Charlottesville; Center Charlottesville; Liberal Manhattan; Conservative Manhattan; Center Manhattan; Liberal
23 Table 3
Distribution of Number of Articles across Main Topic Codes (Row Percentages in Parentheses)
Media Cluster Incident Code Liberal n (%) Center n (%) Conservative
n (%) Row Total n Statistical Tests
Charlottesville BIGGER DEBATE 8 (32) 7 (28) 10 (40) 25 26.281 ***
GOV’T RESPONSE 27 (24) 50 (44) 37 (32) 114
GENERAL REPORTING 34 (40) 26 (31) 24 (29) 84
ANALYSIS 14 (82) 2 (12) 1 (6) 17
Column Total 83 (35) 85 (35) 72 (30) 240
Manhattan BIGGER DEBATE 8 (30) 6 (22) 13 (48) 27 9.004
GOV’T RESPONSE 15 (20) 21 (28) 38 (51) 74
GENERAL REPORTING 19 (26) 20 (27) 34 (47) 73
ANALYSIS 4 (57) 3 (43) 0 (0) 7
Column Total 46 (25) 50 (28) 85 (47) 181
Both BIGGER DEBATE 16 (31) 13 (25) 23 (44) 52 32.251 ***
GOV’T RESPONSE 42 (22) 71 (38) 75 (40) 188
GENERAL REPORTING 53 (34) 46 (29) 58 (37) 157
ANALYSIS 18 (75) 5 (21) 1 (4) 24
Column Total 129 (31) 135 (32) 157 (37) 421
Notes. Significance indicated by ∗ p < .05, ∗∗ p < .01, ∗∗∗ p < .001. Default statistical tests are Chi Squared Tests, and refer to the asymptotic significance. Italicized coefficients refer to the Fisher’s Exact Test, which was performed instead of the Chi Squared Test when either more than 20% of cells had expected count less than 5 or the minimum expected count was below 1.
Table 3 shows that GOVERNMENT RESPONSE (n = 188, 45%) and GENERAL REPORTING (n = 157,
37%) codes occurred most frequently, encompassing 82% of the articles. The remainder of the articles wrote about a BIGGER DEBATE (n = 52, 12%) or provided an ANALYSIS (n = 24, 6%). The
distribution of main topic codes differed statistically significantly between media clusters in the case of Charlottesville (p < .001); for the Manhattan coverage, no statistically significant difference in the main topic distribution was observed between media clusters. The statistically significant differences for the Charlottesville coverage prevailed when combined with Manhattan articles into one overall sample (p < .001). Most notably, when covering Charlottesville, liberal media domi-nated the ANALYSIS category with 82% of the articles (only three ANALYSIS articles written by
center or conservative outlets); center media wrote the bulk of the GOVERNMENT RESPONSE
arti-cles (44%); and the conservative media wrote the highest number of BIGGER DEBATE articles
(40%).
Regression results (Table 4) further show that there is a significant relationship between each explanatory variable (media cluster, incident, media*incident, and day of reporting) and the main topic code family (see also Section 3.3.2). First, relative to the proportion of GENERAL REPORTING
24
media was statistically significantly higher than the proportion of such articles by conservative outlets. There was no significant relationship between media clusters and the odds ratio of num-bers of BIGGER DEBATE or GOVERNMENT RESPONSE to GENERAL REPORTING articles. Second, the
incident variable showed that the ratio of number of ANALYSIS to number of GENERAL REPORTING
articles was higher for the Charlottesville attack than for Manhattan (p < .001). Third, the inter-action variable was also statistically significant for the ANALYSIS code: the ratio of number of ANALYSIS to GENERAL REPORTING articles increased significantly more for conservative media
than for center (for which the ratio decreases) and liberal media (both p < .001) when comparing Charlottesville coverage with Manhattan coverage. The statistical result should, however, be in-terpreted prudently, considering that conservative media wrote zero ANALYSIS pieces about the
Manhattan incident, making comparisons of ratios between Charlottesville and Manhattan prob-lematic. And finally, the day of reporting control variable showed weakly statistically significant coefficients: relative to the GENERAL REPORTING code, the proportion of articles under the BIGGER DEBATE (p = .081) and ANALYSIS (p = .082) codes increased in the later stages of coverage (on
the second, third, and fourth day after the attack).
The comparatively high proportion of ANALYSIS articles published by liberal media can be best
explained by the fact that the majority of such articles critically analyzed Trump’s response to the attacks. With the current overwhelmingly Republican government, center, and especially con-servative, media are less likely to cover the government’s response to a terrorist attack with an Table 4
Multinomial Logistic Regression Results for Main Topic Code Family with GENERAL REPORT-ING as the Reference Category
Main Topic
BIGGER DEBATE GOVERNMENT RESPONSE ANALYSIS
Intercept -2.046 * 0.139 -21.230 *** Media Cluster Center -0.231 -0.052 18.020 *** Liberal 0.181 -0.336 18.456 *** Incident Charlottesville 0.072 0.289 16.593 *** Media*Incident Center*Charlottesville -0.326 0.250 -17.590 *** Liberal*Charlottesville -0.683 -0.318 -16.059 *** Day of reporting d + 1 0.839 -0.140 0.552 d + 2/3/4 1.358 † 0.072 1.870 † R-squared .136 .136 .136
Notes. Significance indicated by ∗ p < .05, ∗∗∗ p < .001, † p < .10. R-squared statistics refer to Nagelkerke’s
25
analytical (critical) stance. An example of such an article is Slate’s “There Was Never Doubt over What Trump Thought of Charlottesville,” which strongly criticized the President’s unwillingness to condemn white supremacists, and argued it was an issue ridden deep within the Republican Party (Bouie, 2017). The only analytical conservative article was published by The Blaze, which openly criticized Trump for failing to “provide leadership … on race [and] hate groups” (Enloe, 2017b). Moreover, the higher proportion of analytical pieces written about the Charlottesville attack might stem from the great controversy around this refrainment of Trump to condemn white supremacists, which media often addressed in separate analytical articles.
Further observations can be drawn by analyzing subcategories of the main topic codes (Table A.1). The proportion of articles that mainly focused on the perpetrator (subcategory within GEN-ERAL REPORTING) for Manhattan coverage was highest for conservative (n = 12, 14% of all
con-servative Manhattan articles), followed by center (n = 4, 8%) and liberal media (n = 3, 7%). Similarly, victims (subcategory within GENERAL REPORTING) were the main topic of a slightly higher proportion of Manhattan articles by conservative (n = 9, 11%) and center (n = 5, 10%) than liberal outlets (n = 3, 7%). No differences in the proportions of perpetrator- and victims-centered articles across media were found in the Charlottesville coverage. The analysis further showed that while conservative media published a slightly higher percentage of perpetrator-cen-tered articles when covering Manhattan, liberal media did when covering Charlottesville. This can suggest a higher interest of conservative media in the Islamic perpetrator and a higher interest of liberal media in the right-wing perpetrator. Finally, the topics of the rise in terrorism, Islam, and the possibility of future terror threat were reported in similar proportions by media clusters.
4.2 Incident Label
Under the incident label code family, each word or phrase used to characterize an attack was tracked. Four main categories (codes) were formed and analyzed: ATTACK, NORMATIVE, TERROR-ISM, and VIOLENCE (see Appendix A for the logic behind this grouping). Examining incident labels
can point to the connotations ascribed to the attacks by outlets. For instance, labeling an incident as ‘crime’ and labeling it as ‘domestic terrorism’ present the reader with a different interpretation of events.
A total of 363 articles (86%) contained at least one label categorized under the four main incident label codes. Table 5 shows that the most prevalent codes were ATTACK (n = 238, 57%) and VIOLENCE (n = 203, 48%). The table reports statistically significant differences across media clusters in the proportion of articles containing the TERRORISM code in the Charlottesville
cover-age (p = .016), where it was used in 6% of liberal articles, no center articles, and 8% of conservative articles. Differences between media clusters for the proportion of articles containing the VIOLENCE
26
combined (p < .001), where it was used in higher percentage of liberal and center than conservative articles. The distribution of the ATTACK and NORMATIVE codes did not differ statistically
signifi-cantly across media clusters.
As reported in Table 5, considerable differences occurred in the use of the incident label codes across incidents: the ATTACK, TERRORISM,and VIOLENCE codes showed in 35%, 5%, and 68% of all Charlottesville articles, compared to 85%, 60%, and 22% of Manhattan articles, respectively. The high number of VIOLENCE code occurrences in the Charlottesville coverage is largely because
Table 5
Number (Percent) of Articles Containing Incident Label, Perpetrator Label, and Agency Codes
Media Cluster Code Incident Liberal n (%) Center n (%) Conservative
n (%) Row Total n (%) StatisticTest
Incident Label ATTACK Charlottesville 27 (33) 29 (34) 29 (40) 85 (35) 1.109 Manhattan 41 (89) 44 (88) 68 (80) 153 (85) 2.539 Both 68 (53) 73 (54) 97 (62) 238 (57) 2.859 NORMATIVE Charlottesville 5 (6) 9 (11) 3 (4) 17 (7) 2.659 Manhattan 2 (4) 5 (10) 5 (6) 12 (7) 1.319 Both 7 (5) 14 (10) 8 (5) 29 (7) 3.769 TERRORISM Charlottesville 5 (6) 0 (0) 6 (8) 11 (5) 7.929 * Manhattan 33 (72) 30 (60) 45 (53) 108 (60) 4.386 Both 38 (29) 30 (22) 51 (32) 119 (28) 3.900 VIOLENCE Charlottesville 61 (73) 64 (75) 38 (53) 163 (68) 10.881 ** Manhattan 7 (15) 13 (26) 20 (24) 40 (22) 1.808 Both 68 (53) 77 (57) 58 (37) 203 (48) 13.243 *** Perpetrator Label ATTACKER Charlottesville 0 (0) 1 (1) 0 (0) 1 (0) 1.712 Manhattan 19 (41) 21 (42) 13 (15) 53 (29) 15.148 *** Both 19 (15) 22 (16) 13 (8) 54 (13) 4.773 † GENERAL Charlottesville 23 (28) 35 (41) 29 (40) 87 (36) 4.017 Manhattan 37 (80) 37 (74) 52 (61) 126 (70) 5.862 † Both 60 (47) 72 (53) 81 (52) 213 (51) 1.328 RADICAL Charlottesville 12 (14) 10 (12) 4 (6) 26 (11) 3.281 Manhattan 9 (20) 12 (24) 45 (53) 66 (36) 18.984 *** Both 21 (16) 22 (16) 49 (31) 92 (22) 12.837 ** Agency AGENT Charlottesville 32 (39) 34 (40) 19 (26) 85 (35) 3.703 Manhattan 29 (63) 41 (82) 56 (66) 126 (70) 5.125 † Both 61 (47) 75 (56) 75 (48) 211 (50) 2.356 PATIENT Charlottesville 30 (36) 29 (34) 29 (40) 88 (37) 0.652 Manhattan 5 (11) 14 (28) 20 (24) 39 (22) 4.532 Both 35 (27) 43 (32) 49 (31) 127 (30) 0.827
Notes. Significance indicated by ∗ p < .05, ∗∗ p < .01, ∗∗∗ p < .001, † p < .10. Italicized coefficients refer to Fisher’s
Exact Test,which was performed instead of Chi Squared Tests when either more than 20% of cells had expected count less than 5 or the minimum expected count was below 1. Chi Squared Tests refer to the asymptotic significance.
