The Influence of Internet Access on Regime
Based Selection Bias: International
Newspaper Coverage of the Libyan civil war
University of Groningen
Faculty of Economics and Business
MSc International Economics and Business
June 2017
Laurien Verschuur s2202697
l.c.verschuur.1@student.rug.nl
Supervisor: dr. R.K.J. Maseland Co-Assessor: dr. A.A. Erumban
1
Abstract
2
Table of Contents
1. Introduction ... 3
2. Literature Review ... 5
2.1 Mass Media and Foreign Policy ... 5
2.2 Regime Based Reporting Bias ... 7
2.3 War Journalism Dynamics ... 10
2.4 Technological Development ... 13
2.5 Feedback and Competition ... 14
3. Data and Methods ... 16
3
1. Introduction
Anyone following recent coverage on American president Donald Trump might be aware of the fact that his current relationship with and towards the press is not at its best of terms. If anything, it could be characterized as being hostile with him not attending the White House Correspondents’ Dinner last April and hosting a political rally in Pennsylvania instead. At the annual Conservative Political Action Conference in February he even announced that: “We are fighting the fake news, it’s fake news, fake. The fake news doesn’t tell the truth. It doesn’t represent the people. They have no sources, none” (The Telegraph, 2017). Yet, following the elements of journalism, according to Kovach and Rosenstiel (2014), journalism’s first
obligation is to the truth while its essence is a discipline of verification. Any deviation from this element of truth could be labeled as media bias.
Even more, August 2016 a photograph of a young boy sitting in an ambulance, his face covered in blood and dust from an airstrike in Syria, became a symbol of the civilian suffering caused by Syria’s relentless civil war (The Independent, 2016). The image quickly spread amongst news agencies and social media, and gained a lot of international attention. Yet, recent news stories from almost a year later, June 2017, show a different side to the story, with the father of the little boy claiming that the image was used for propaganda purposes and that his son only suffered minor injuries, with the blood on his face actually having dripped from one of his own wounds (BBC, 2017). This recent coverage spreads a new light on the effects that internet, and the speed with which stories and images spread, can nowadays have on media reporting and more specifically media bias.
Media or reporting bias is ‘the media’s tendency to systematically underreport or overreport certain types of events’ (Baum and Zhukov, 2015). Different reports based on a similar set of underlying facts can each communicate a significant different impression of an event due to selective omission, word choice and variable credibility assigned to primary sources (Gentzkow and Shapiro, 2006). Media bias comprises of at least two distinct
deviations from an ideal standard of objectivity (Snyder and Kelly, 1977). The first deviation references to the selection of news; the coverage across classes of events with different completeness. The second type of bias entails the distortion of content of news; such as inaccurate reporting (Jacobs, 1967) or an imbalanced interpretation of events (Bagdikian, 1972; Sigelman, 1973).
4
establish an either conservative (Alterman, 2008; Franken, 2004) or liberal (Coulter, 2003; Goldberg, 2014) overall media bias in domestic reporting.
But besides influencing domestic reporting, reporting bias is a continuous problem for both participants and observers of armed conflict, and shapes the public understanding of war (Baum and Zhukov, 2008). The content of reported news can significantly affect political attitudes and outcomes (Strömberg, 2004; Gentzkow and Shapiro, 2004; Gentzkow, 2006; Gerber, Karlan and Bergan, 2007; DellaVigna and Kaplan, 2006). Furthermore, the extent to which systematic deviations are pertinent in media reporting on political events could
potentially have considerable implications for public knowledge, policy and scientific inference (Bennet, Lawrence & Livingston, 2008; Hallin and Mancini, 2004; Iyengar and Kinder, 1987; Livingston and Bennet, 2003; McCombs and Shaw, 1972). Moreover, if news providers publish different accounts of domestic and international conflict events, this difference will have important consequences for both theory and practice (Snyder and Kelly, 1977; Schrodt, Simpson and Gerner, 2001; Reeves, Shellman and Stewart, 2006). Therefore it is important to create an understanding of what drives media bias.
Furthermore, nowadays journalism no longer is the first or sole bringer of news, nor the sole explainer or source with authority, power and truth. New technologies have created ample room for more voices to be heard, and more platforms for news to be broadcasted on. Leading newspapers today all own their online versions, and user generated content is becoming more often referred to due to the possible direct connections through smartphones. These technological developments are likely to influence the construction of news content and therefore also likely to affect media bias, as shown by the aforementioned example
concerning the little boy in Syria. The research question of this thesis will therefore be whether or not advancement in communication technologies, represented by the availability of internet, has influenced media bias.
5
The rest of the paper will be structured as followed. The next section will review the relevant literature describing the relevance of media bias in foreign policy formation and international conflict reporting, the developments concerning war journalism and media bias, and the effects of technological developments on international conflict coverage while the hypotheses are established. After that, the data and methodology will be explained upon, followed by a section that discusses the results of the analysis. Lastly, the paper will give a discussion of the results, discuss some limitations and possibilities for future research, and end with a few concluding remarks.
2. Literature Review
The literature review will encompass the following contents. First the relationship between mass media and foreign policy, indicating the importance of media bias in affecting foreign policy, will be elaborated on. This will be followed by literature concerning regime based reporting bias to establish the dependence of reporting bias concerning international conflict on the political environment of the home-country in which a media outlet is based. Hereafter, changes in war journalism dynamics over the last decades will be discussed and linked to media bias and technological development. Consequently, the development of the technical components with regards to international conflict reporting will be highlighted. Lastly, feedback and competition effects on news construction and media bias will be described after which the association with internet will be made in order to establish a link between the potential effects of communication technology developments, in the form of increased internet availability, and regime based reporting bias.
2.1 Mass Media and Foreign Policy
Regarding the relationship between media and the coverage of international events there are two agreed upon beliefs shared amongst policy makers from all political positions. Firstly, the mass media often do not display the message decision makers would like them to contain and secondly, the media not only shapes the public understanding of war but also the public opinion of foreign policy (Baum and Potter, 2008). Existing political science literature however does not focus on the media’s role as having independent influence on public
opinion and policy but points out the media as a ‘conveyer belt’ which passively delivers elite opinions, and more specifically the opinions of the most influential elites, to the public
6
reporting according to political elite messages (Hallin, 1986; Bennet, 1990; Bennet, Lawrence and Livingston, 2006). Hence, in political science the media is seen as merely a linkage mechanism that communicates elite rhetoric and not as an individual strategic actor which shapes the policy-making process (Malek, 1996; Brody, 1991; Bloch and Lehman-Wilzig, 2002).
Political communication literature on the other hand does describe processes of mass media shaping political opinions as an individual influence (Kuypers, 1997; Graber, 2014; Paletz, 2002). Gatekeeper literature for example, ascribes a more proactive role to the media than indexing theories, as it describes that reporters shape news based on newsworthiness of events (White, 1950; Galtung and Ruge, 1965; Patterson, 1993). Nevertheless, within this strand of political communication literature the relationship between media reporting and public political opinion is not consistently linked to actual policy outcomes. Yet, scholars examining the interactions of the public, decision makers and the media have come to a growing recognition of the fact that these three actors are interdependent and that excluding one or two of them could cause distorted theoretical frameworks and empirical results (Entman, 2000; Nacos, Shapiro and Isernia, 2000; Holsti, 2009).
Baum and Potter (2008) in their publication also acknowledge this interdependency between the public, leaders and mass media and introduce a market equilibrium framework which explains how the dynamics between an unequal division of information amongst the public and its leaders, in which the public is largely informationally disadvantaged therefore allowing the political decision makers to influence preferred informational frames, shapes the field of foreign policy. Within this line of reasoning the media is a central actor in the foreign policy marketplace and the key market commodity that collects, frames and distributes the information. The mass media is believed to be more accurately displayed as a distinctive strategic actor instead of a mere conveyer, while within this framework the political elites still clearly influence news reporting consistent with earlier indexing theories (Hallin, 1989; Bennet, 1990; Zaller and Chiu, 1996). Baum and Potter (2008) reason a market equilibrium framework is necessary as the conveyer-belt analogy does not entail the nature of the media as primary link between leaders and the public.
It is discussed that in order to unravel the foreign policy making as a dynamic
7
information, which the public usually prefers to be unframed and objective. It is suggested that the media is captured in a continuous process of delivering news in frames consistent with, often opposing, requirements from both decision makers and the public (Baum and Potter, 2008).
Baum and Potter (2008) argue that concerning foreign policy this results in the media reporting ‘rally-round-the-flag’ frames preferred by political leaders in the early stages of political conflict. This market framework is in line with Zaller and Chiu’s (1996) research which indicates that the media usually rallies along because of the dependence on political sources for information. These ‘rally-frames’ are accepted by the public due to the
informational disadvantage of the public at the early stages. However, when a conflicts’ duration lasts longer and more sources of verification become available these frames will disappear and publications containing increasingly diverse opinions will arise. Baum and Potter’s (2008) market framework also relates to Entman’s (2004) arguments concerning culturally congruent frames: when decision makers discuss culturally ambiguous frames with diverse interpretations the media could potentially independently affect foreign policy while independent media influence is not expected when leaders refer to culturally congruent frames shared by most of the public.
To conclude it can be stated that according to Baum and Potter (2008) the interplay between pressure of both the public and political leaders on the media, to obtain and
respectively control information flows concerning international news events, influences and shapes the field of foreign policy. Hence, what the media reports and whether or not this is biased is likely to have important consequences for foreign policy formation and therefore it is important to gain more understanding on the determinants of reporting bias in international news coverage.
2.2 Regime Based Reporting Bias
8
discussed that the reporting of a conflict is dependent on two sources of bias namely newsworthiness and politics. Hence, it depends on the type of event that takes place on the ground and the regime in which a media organization is located.
These two sources of bias are argued to stem from similar forces as the opposing pressures of the public and political leaders on the media described by Baum and Potter (2008). Baum and Zhukov (2015) reason that newsworthiness causes reporting bias due to the commercial incentives of both journalists and editors to maximize readership, hence to fulfill the demand of the public mentioned by Baum and Potter (2008). For politics to result in media bias Baum and Zhukov (2015) refer to the political preferences of incumbent
governments in order to maintain their power. This is similar to the argument made by Baum and Potter (2008) that news organizations are dependent on incumbent governments, and thus corresponding preferences, for the supply of information. Baum and Zhukov (2015) state that, just as with the ambivalent forces by the public and leaders, both sources often oppose each other and that the selection of reported events is dependent on the media freedom that is allowed by a country’s political system.
Baum and Zhukov (2015) illustrate how media organizations seek to increase readership by prioritizing news reports which are remarkable and salient for the intended readers (Galtung and Ruge, 1965). Moreover, recent literature presents evidence of the fact that irregular or counterintuitive events receive more coverage than events without such elements (Snyder and Kelly, 1997; Graber and Dunaway, 2014; Earl et al., 2004). Other perceptions of newsworthiness are determined by large and dramatic events, conflict or ‘bad news’ and physical or cultural proximity amongst others (Wooley, 2000; Davenport and Stam, 2006; Patterson, 1996; Sabato, 2000; Baum and Groeling, 2010; Morton and Warren, 1992; Rosengren, 1974). ‘Coverage fatigue’ on the other hand will cause a decrease in the amount of war reporting due to media consumers losing interest in a conflict after a while (Davenport and Stam, 2006; Baum and Groeling, 2010).
As mentioned before however, news organizations are not the sole actors determining the newsworthiness of events as otherwise only small cross-national variations in the selection of news is to be expected (Galtung and Ruge, 1965). The political system in which a media outlet is embedded may potentially also imply its preferences concerning appropriate
9
al. 2003; Enikolopov, Petrova and Zhuravskaya, 2011); second, political systems can influence privately owned media outlets indirectly through regulations concerning licensing requirements, taxations and laws that restrict freedom of expression (Whitten-Woodring and James, 2012); and last, states could standardize certain norms which force editors and reporters to self-censor and refrain from watchdog reporting in order to maintain a positive working relationship with the government, which holds an important position as supplier of information (Bennett, Lawrence and Livingston, 2008; Schudson, 2003; Djankov et al., 2003).
Thus far, the link between reporting bias due to commercial and political sources has been discussed and connected to the media freedom that is allowed by a country’s political system. Baum and Zhukov (2015) then explain that the extent to which media bias is displayed is dependent on how media outlets deal with the political systems that shape the political environment in which they are embedded. Governments of non-democratic and democratic states could have different preferences on which events receive coverage as the costs of losing power are higher for dictators than elected politicians (Debs & Goemans, 2010). Therefore, dictators are incentive to not have certain events that directly or indirectly threaten their own position covered.
It is stated that when there is limited governmental pressure on media organizations, often in democratic regimes, this results in the commercial bias prevailing. Concerning the coverage of armed conflict this is reasoned to result in disproportionate reporting on uprising prioritizing change of the status quo. This is due to the fact that relatively free media outlets would prefer unexpected, dramatic and large-scale changes from the status quo as determined by journalistic perceptions of newsworthiness. If political constraints on media reporting are more evident, often in non-democratic regimes, this will give space for the rise of the political bias influencing media reports. With regards to war journalism, this will cause increased reporting on the inevitability and rightfulness of the prevailing in order to enhance political self-preservation of autocratic regimes.
10
wounded or killed by rebel violence. The hypotheses described in Baum and Zhukov’s (2015) paper to test this theory provide the basis for this research. Therefore the first two hypotheses to be tested in this thesis will similarly be:
Hypothesis 1: In international newspaper coverage of the Libyan civil war, media organizations in democracies will have a selection bias towards non-violent protests and
events concerning civilian casualties due to government violence.
Hypothesis 2: In international newspaper coverage of the Libyan civil war, media organizations in non-democracies will have a selection bias towards events concerning
civilian casualties due to rebel violence.
2.3 War Journalism Dynamics
War journalism, and more specifically war propaganda, is of huge importance during armed conflicts as the media inform and influence political actors and shape the public debate. An armed conflict is therefore not solely fought by soldiers of the army, but also depends largely on the media battle that is fought over public opinion and the populations’ and troops’
willingness to enter the battlefield (Taylor, 2002; Thussu and Freedman, 2003). Nohrstedt (2009) in his article analyses how war journalism has changed since the end of the Cold War by examining the Swedish reporting of four armed conflicts since the 1990s. It is argued that as a consequence of media technology developments, including satellite TV channels that telecast new 24/7 so quickly that live reports from a combat zone have been made possible, media and journalists have established a rising central position within conflicts and wars. This results in the parties at war even more than ever trying to influence, steer and control coverage by the international media, most specifically concerning visual reports as photographs and videos.
11
by the ‘clinical war’ image yet when Iraq attacked Israel the horror and suffering were all reported from nearby. Within media research this is referred to as the distinction between worthy and unworthy victims in media reporting (Herman and Chomsky, 1988). To conclude it can be stated that in the 1990-1991 Gulf War the image of a clinical war was predominant yet this image turned out to be completely misleading as there were large number of Iraqi civilian casualties (Nohrstedt, 2009). Even though there was around-the-clock broadcasting and the largest number of correspondents yet, this could not overcome a certain reporting bias covering the war.
The second conflict of which Norhstedt (2009) evaluates the media coverage is the Kosovo Conflict in 1999. Within this conflict the NATO first also tried to spread the image of a clinical war and first only the Kosovo Albanian refugees where stamped as ‘worthy
victims’. However, later on this image became more balanced due to probably both geographic and cultural characteristics. As the conflict was in closer proximity it became possible to establish professional contacts with the media in the Balkan. Because of this the one-sided propaganda image that the NATO initially tried to disseminate was counterbalanced and Serbian civilian victims also received coverage as ‘worthy victims’. To summarize, initially a one-sided propaganda image of a clinical war was delivered, however this became more balanced and contradicted by local reports due to closer proximity of the conflict.
12
The last conflict that Norhstedt (2009) researches is the 2003 Iraq War or Second Gulf War, in which the internet saw its breakthrough concerning war journalism. What first stands out is that over the years, since the 1990-1991 Gulf War, journalistic practices have changed towards an increased emphasis on civilian suffering from being nearly present early in the Afghanistan War to being a significant theme early in the Iraq War (Nord, Shehata and Strömbäck, 2003). Moreover, journalists were embedded by the American army in larger numbers than the First Gulf War (Norhstedt, 1992). The 600 journalists that were supposed to report on site first had to undergo hostile-environment training and learnt how to adjust themselves between the soldiers (Tumber and Palmer, 2004).
Consequentially these journalists turned into elements of the war propaganda as they became to share the military’s point of view on events due to experiencing the soldiers’ daily lives and being dependent on the army for safety. Because of the embeddedness of journalists and America’s superior public relations management the overall Swedish media discourse was biased towards pro American reporting (Nord, Shehata and Strömbäck, 2003). Yet, in multiple Arab countries television reporting became less reliant on CNN due to the rise of Arabic satellite television channels such Al-Jazeera which delivered critical reporting
concentrating on US dominance in the Middle East (El-Bendary, 2005; El Gody, 2005). Also, again visual materials became increasingly important and the conditions of war journalism itself received attention in the media.
13 2.4 Technological Development
Instead of analyzing changes in war journalism by examining coverage of armed conflict over time as Norhstedt (2009), other scholars have focused on the development of the technical components concerning international conflict reporting. For example, with the rise of internet resulting in the creation of online media it has been argued that due to technology development common citizens will be able to challenge and overcome the authority of media organizations and diminish the power that is attributed to them (Armstrong and Zúniga, 2006; Reynolds, 2006). Livingston and Bennet (2003) further examine this matter by investigating if technology is changing the news making process through gatekeeping, indexing and live event news.
For Livingston and Bennet’s (2003) research a distinction between managed and unmanaged news is made by definitions of pseudo-events and event-driven news. Pseudo-events are referred to as being arranged, planted or scheduled with an absence of spontaneity which makes them attractive to media outlets as they can be anticipated, controlled and integrated such as, for instance, press conferences which are tailored to accompany news production routines and deadlines (Boorstin, 1977). Event-driven news focuses on the impact of spontaneous, dramatic news events which are harder for political leaders to control and therefore pushes officials to respond to a news agenda rather than set it for media outlets (Lawrence, 2000).
14
This last argument could be compared to what in the field of international affairs coverage is called the CNN effect. This effect refers to the fact that technological
developments, resulting in the 24-hour news cycle, have radically changed the relationship between foreign policy and public opinion (Giboa, 2005). The CNN effect is argued to force politicians into a more rapid response to conflicts, or let alone even a response, so that responsiveness to public demand and control of the situation is shown to the public (Robinson, 2000; Robinson, 2001).
Yet, there is no consensus on or whether or not it is the 24-hour news cycle that changed the relationship between foreign policy and the media. Other scholars argue this could also be due to other developments in the international system such as the end of the Cold War or the 9/11 terrorist attacks (Entman, 2004; Zelizer and Allan, 2002). Robinson (2005) even discusses that the CNN effect altogether has been undermined by following events, such as the war in Iraq, and increased governmental media management. It is stated that the fundamentally transformative influence of the CNN effect expected by scholars was less significant than proposed (Baum and Kotter, 2007). Nevertheless, the hypothesized CNN effect still accentuates the influences that technological developments in the last decades have had on media. Technological advancement in international conflict coverage has created an environment in which event-driven news flourishes while allowing for less political influence in setting the news agenda, thereby potentially decreasing media bias and thus indirectly affecting foreign policy.
2.5 Feedback and Competition
Gentzkow and Shapiro (2006) in their article formulate a model in which media bias does not stem from consumer preferences for confirmatory information, journalists’ motivations to promote their own ideological views or incumbent politicians’ ability to pressure the media. Rather, media bias is argues to originate as a logical outcome of media organizations’ aim to create and reputation of trustworthiness, even though limiting reporting bias increases the welfare of all the agents in the economy. Within this research three main results are identified; first, media organizations are inclined to bias their reports towards prior beliefs of their
15
As the research of this thesis will focus on the effect that technological developments have on regime based media bias in international conflict reporting, this second and third finding on verification possibilities and competition have interesting implications. It is suggested that the course and intensity of media bias is determined by feedback and
competition (Genztkow and Shapiro, 2006). Concerning feedback more particular, research on sports picking by local newspaper reporters and financial reporting indicates that ex-post feedback can decrease media bias (Gentzkow and Shapiro, 2006; Lim, 2001). According to Gentzkow and Shapiro (2006), direct feedback has a significant impact on reducing media bias because if the probability of ex-post feedback on the true state of the world increases it allows for ex-post verification by consumers. This diminishes media outlets’ motivation to misreport as attestable bias will damage their reputation as supplier of truthful information and therefore deter consumers. Sequentially Gentzkow and Shapiro (2006) argue that
competition also decreases media bias as it enhances the chance of ex-post exposure because a higher level of competition allows for higher levels of ex-post feedback on news reports.
Both feedback and competition could be increased due to technological developments such as access to internet. The more information that can be found online, the higher the competition in news coverage will be and the faster feedback on reporting can be provided. As direct feedback and increased competition are argued to decrease bias, internet availability can therefore potentially influence media bias through increasing direct feedback and
competition.
To summarize the last three sections regarding war journalism dynamics,
technological development and feedback and competition it can be recapitulated that: firstly, it is argued that changes in war journalism can be linked to subjects of communication technology and changing patterns of media bias; secondly, technical developments in international conflict reporting allow event-driven news to prevail over political news preferences possibly decreasing reporting bias; and thirdly, higher levels of direct feedback and competition due to increased internet availability potentially decrease media bias.
16
Hypothesis 3: In international newspaper coverage of the Libyan civil war, media organizations in democracies will have a selection bias towards non-violent protests and
events concerning civilian casualties due to government violence, and this selection bias decreases when internet access is high as opposed to low.
Hypothesis 4: In international newspaper coverage of the Libyan civil war, media organizations in non-democracies will have a selection bias towards events concerning civilian casualties due to rebel violence and this selection bias decreases when internet access
is high as opposed to low.
This literature review starts with stipulating the importance of media bias by discussing how the interplay between the public, political leaders and mass media, and thereby media bias, determines foreign policy outcomes. Sequentially, it is mentioned that regime based reporting bias in international newspaper coverage, due to variating
governmental constraints on press freedom, results in a ‘change bias’ in democracies while resulting in a ‘status-quo bias’ in non-democracies. Hereafter changes in war journalism, limiting some reporting bias, are described and connected to increased journalistic
globalization due to technological advancement. This is followed by discussing the potential decreasing influence of internet access – which represents technological developments, direct feedback and competition effects – on media bias in international newspaper coverage of conflicts. Or in this case more specifically, the influence of internet access on regime based reporting bias in international newspaper coverage of the Libyan civil war.
3. Data and Methods
The data and methods section will elaborate upon the construction of dataset employed, discuss the research sample and describe how each dependent, independent and control variable has been defined and measured. This will be followed by the methodology used for analysis after which the models used for hypothesis testing will be highlighted. Lastly, the use of a robustness check is explained.
3.1 Dataset
17
the scope of this project, Baum and Zhukov’s (2015) dataset available for replication purposes will be gratefully employed in combination with variables added. Baum and Zhukov (2015) give three reasons why, in light of different forms of electronic and print media available for news reporting, a sample based on solely newspapers was constructed. Firstly, newspapers are still seen as the main source of intelligence on political, social and economic events with three times more daily newspaper readers than broadband internet consumers 2011 (World
Association of Newspapers and News Publishers, 2014). Secondly, it allows for a
comparatively stable and representative data sample since internationally newspapers adhere to three standard formats with similar editorial constraints and newspaper articles are
generally archived better than broadcast transcripts. Lastly, it allows for continuity and comparability with previous findings because newspaper based event data have traditionally been employed in the study of social movements and violence and protest data are often nor available from governments or NGOs while offered by newspapers articles.
3.2 Sample
The dataset provided by Baum and Zhukov (2015) contains 213.406 selected and coded articles on 1.510 events by 2.252 newspapers in 113 countries for the time period between December 18th 2010 and October 23th 2011. It was constructed by administering the count of all electronic and weekly newspapers per country archived in electronic databases LexisNexis and ISI Emerging Markets. For the 2.225 newspapers identified, each article containing the term ‘Libya’, in English or the newspaper’s language, was then selected for the afore
mentioned time period. This specific timeframe was applied as December 18th marks the first day of Tunisian protests, which led to the Arabic Spring, and October 23th marks the third day following the capture and death of Muammar Gadaffi. The 1.510 events during this time window were, to refrain from an overlap with the selection of newspapers, identified by a mutually exclusive group of electronic sources and newswires such as CNN, BBC and Al-Jazeera and recorded by timing, participants, type of violence and casualties amongst others. For this examination two levels of analysis will be investigated: the country level and the country-day level. This means that at the country level the sample consists of 113 countries while at the country-day level it consists of 311 days per country. For a full list of countries included in the sample see Appendix A.
3.3 Dependent Variable
Concerning the country level, the dependent variable 𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒𝑗is the proportion of
18
events: nonviolent protests, rebel caused civilian casualties and government caused civilian casualties. Hence, each country has three 𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒𝑗coverage variables.These three types of events were selected as it is argued by Baum and Zhukov (2015) that these types of events in combination with the regime in which a newspaper is based influence whether or not an article on Libya is published. This variable can be decomposed as follows:
𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒𝑗 = 1
𝑡(𝑥) ∑ 𝑡(𝑥)( 1
𝑖∑ 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑖𝑛𝑔𝑖 𝑖𝑗𝑡𝑡(𝑥))
𝑡 (1)
where 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑖𝑛𝑔𝑖𝑗𝑡 refers to a published article on respectively day t in country j by
newspaper i and 𝑡𝑥 refers to the set of days following and event of type x. At the country-day level, the dependent variable is a binary variable, 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑖𝑛𝑔𝑗𝑡, indicating per country j whether or not at least one newspaper published an article on Libya at day t.
3.4 Independent Variables
For this research the effect and moderating effects of three variables – regime type, conflict type and internet access – on publishing will be estimated. The first independent variable, type of regime, is also a binary variable, 𝐷𝑒𝑚𝑜𝑐𝑟𝑎𝑐𝑦𝑗 indicating whether or not country j in the sample was a democracy in 2010. For robustness purposes Baum and Zhukov (2015) provide three ways of measurement for democracy, namely a country’s Polity2 score (Jagger and Gurr, 1995), Freedom House’s (2011) Press Freedom score and Cingranelli and Richards (2010) Freedom of Speech score. Because Baum and Zhukov (2015) established that all three measurements are strongly correlated and the Freedom of Press and Freedom of Speech variables take on three possible values whereas the Polity2 score ranges from -10 (strongly autocratic) to +10 (strongly democratic), the last measure will be employed for this research. A country is to be considered a democracy, and hence denoted by a value “1”, by a cutoff of six and higher at the Polity2 scores. Countries with Polity2 scores below six will be denoted by a value of “0” to indicate a non-democracy. This variable is used in both levels of analysis. According to this measure the research sample exists of 58 democracies and 54
non-democracies.
19
Lastly, internet access will be measured in two different ways. The percentage of a country’s population with internet access, derived from the International Telecommunication Union (2014), will be employed for both the country and country-day levels of analysis. It is defined by the ITU as: “The estimated number of Internet users out of total population. This includes those using the Internet from any device (including mobile phones) in the last 12 months. A growing number of countries are measuring this through household surveys. In countries where household surveys are available, this estimate should correspond to the estimated number derived from the percentage of Internet users collected. (If the survey covers percentage of the population for a certain age group (e.g., 15-74 years old, the estimated number of Internet users should be derived using this percentage, and note indicating the scope and coverage of the survey should be provided). In situations where surveys are not available, an estimate can be derived based on the number of Internet subscriptions”.
The internet availability within Libya during the civil war will solely be used for examination at the country-day level as daily observations are not possible at the country level. Internet availability within Libya is determined by ‘Google transparency report’ which shows an estimation of near time traffic for Google products and services and tries to specify systematic interruptions in such traffic. For Libya this shows a significant drop of the query for the time period between March 3th 2011 and August 22 2011. Comparing this data with several news reports suggests that this coincides with the government cutting off Libya’s internets access in March and the rebels taking over Tripoli in August, after which Google search traffic and internet availability were on the rise again. Internet availability will be analyzed by means of a binary indicator as otherwise it cannot be employed for statistical analysis. The periods from 18 December 2010 through 3 March 2011 and 22 August 2011 through 24 October 2011 will be indicated with a value of “1”, as internet in Libya was available during that period. Days in the period from 4 March 2011 through 21 August 2011 therefore take on the value of “0”.
3.5 Control Variables
Besides regime, type of event and internet access, other independent variables could influence coverage of the Libyan civil war and the effects of regime, type of event and internet access on coverage of the Libyan civil war as well. They are employed to control for an
20
country-day level two more control variables, accounting for daily characteristics, were added. The selection of control variables comprises of geographical distance, proportion of publicly owned newspapers, internal conflict years since WWII, number of newspapers, participation of the NATO intervention coalition and at the country-day level also the number of violent war events and a one-day-lagged variable of 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑖𝑛𝑔𝑗𝑡. These control variables are available within the dataset of Baum and Zhukov (2015) and will therefore be extracted from this dataset.
Geographical distance refers to the news value of proximity and is measured by the physical distance to Libya in kilometers. Proximity is described as a determinant of
newsworthiness by Galtung and Ruge (1965) in a way that events concerning people with similar cultural characteristics as the audience tend to receive more coverage. Hence, a country geographically closer located to Libya potentially publishes more articles on its civil war.
The second control variable concerns the proportion of publicly owned newspapers per country to account for direct government influence on the news agenda and is displayed as a value between zero and one. As public ownership tends to be higher in non-democracies, this could influence the regime-based effect on coverage of the Libyan civil war (Djankov et al., 2003). Data on ownership were collected from another research by Baum and Zhukov (2013), for which they subtracted ownership characteristics from industry organization listings, international media guides, financial databases, annual company reports and the websites of individual news organizations and their parent companies.
The war years experienced since WWII are measured as the amount of internal conflict years since 1945. This potentially influences coverage of the Libyan civil war as experiencing internal conflict years could affect the impact a Libyan civil war event story has. Stories with a bigger impact have better chances of being published according to Galtung and Ruge’s (1965) news values. The number of conflict years a country experienced since 1945 are listed according to the Peace Research Institute Oslo’s (PRIO) Armed Conflict Dataset (2009) and supplemented by Themnér and Wallensteen’s (2011) paper on 1946-2010 armed conflict.
21
census of newspapers is determined by all weekly and daily newspapers per country, excluding weekend supplements, inserts evening editions and similar associated materials, listed in LexisNexis and ISI Emerging markets electronic databases.
Countries actually intervening in the civil war by joining the NATO coalition could face wartime restrictions on their coverage such as ‘rally-round-the-flag’ effects (Groeling and Baum, 2008). This implies that a county’s media will initially rally along with its political leaders and their ‘rally-frames’, thereby potentially increasing coverage. Moreover, the
audience of NATO coalition country will identify better with the topic due to both impact and proximity which, as mentioned before, could increase coverage as well (Galtung and Ruge, 1965). NATO coalition participation is determined by countries participating in the
Operations: Harmattan, Ellamy, Mobile, Odyssey Dawn and Unified Protector. A complete list of participating coalition countries can be found in Appendix B.
Above mentioned control variables will be used at both levels of analysis, as they are country specific. The next two control variables will only be employed at the country-day level of analysis. For the independent type of event variable merely non-violent protests and civilian casualties are taken into account. As other events are also likely to influence civil war coverage the number of violent war events will be controlled for. Violent war events are determined as the number of events that include rebel violence against the government and government violence against rebels at day t from the constructed events database (Baum and Zhukov, 2015). Because unexpected and bad news, also according to Galtung and Ruge’s (1965) news values, is more likely to be reported, violent war events are expected to positively influence coverage on Libya during the civil war.
The last control variable concerns the time effects of what has been published a day earlier in a country. Galtung and Ruge’s (1965) news value of continuity entails that events with a continuing impact, such as a civil war, will be able to establish an ongoing narrative or running story as it develops. This is due to the already present reporters at site and previous reporting making a story more understandable to the audience. Therefore a one-day-lagged dependent variable 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑖𝑛𝑔𝑗𝑡−1is calculated and used to control for these continuity effects.
3.6. Methodology
22
variable is defined as a continuous variable which measures the coverage proportion per type of event by values between zero and one. At the country-day level a random effects logistic regression will be employed, as the sample consists of panel data with a binary dependent variable. A random effects analysis is used to account for potential unobserved heterogeneity in countries’ individual attributes. This is chosen over a fixed effects analysis, as fixed effects analysis does not allow for coefficient estimation of time-invariant variables, which concerns two of the main independent variables, democracy and internet access, and all five control variables used at both levels of analysis.
Yet, before any of the variables can be examined it is important to assess whether there is an issue of multicollinearity. Multicollinearity refers to a situation in which two of the independent variables appear to be highly correlated. This can influence a regression analysis by giving an overstated estimation of the effects of the independent variables or cause high standard errors for the individual coefficients (Brooks, 2014), hence making the analysis less reliable.
Table 1: Correlation matrix country-level
1 2 3 4 5 6 7 8 9 10 1 protests x 2 cas. by gov. 0.965 x 3 cas. by reb. 0.677 0.761 x 4 democracy 0.414 0.374 -0.041 x 5 internet 0.377 0.335 0.203 0.492 x 6 distance -0.026 0.015 -0.042 0.112 0.026 x 7 public -0.280 -0.262 -0.105 -0.393 -0.262 -0.155 x 8 war years -0.069 -0.032 -0.102 -0.051 -0.308 0.076 0.027 x 9 newspapers -0.009 -0.024 -0.074 0.202 0.341 0.172 -0.106 0.094 x 10 coalition 0.373 0.344 0.258 0.227 0.402 -0.140 -0.115 -0.075 0.401 x
There is no general agreement upon which correlation values indicate the existence of multicollinearity. Some researchers argue that for correlation values above 0.5
23
no corrective measures need to be taken. Regarding the country-day level dataset all independent variables also have correlation values below 0.493 as shown by Table 2, once againindicating there is only small to little correlation. Furthermore, both correlation tables give a first idea about the correlation between the dependent and independent variables. However, only 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑖𝑛𝑔𝑗 and 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑖𝑛𝑔𝑗𝑡−1seem to be highly correlated.
Table 2: Correlation matrix country-day level
1 2 3 4 5 6 7
1 publish x
2 publish t-1 0.605 x
3 democracy 0.119 0.120 x
4 protests t-1 0.023 0.011 0.000 x
5 cas .by gov. t-1 0.040 0.024 0.000 0.365 x
6 cas. by reb. t-1 0.016 0.015 0.000 -0.019 -0.014 x 7 violence 0.134 0.121 0.000 0.170 0.202 -0.019 x 8 int. access 0.292 0.292 0.492 0.000 0.000 0.000 0.000 9 int. cut-off -0.127 -0.132 0.000 0.144 0.082 -0.109 -0.151 10 distance -0.019 -0.019 0.112 0.000 0.000 0.000 0.000 11 public -0.077 -0.077 -0.393 0.000 0.000 0.000 0.000 12 war years -0.018 -0.017 -0.051 0.000 0.000 0.000 0.000 13 newspapers 0.283 0.283 0.202 0.000 0.000 0.000 0.000 14 coalition 0.266 0.267 0.227 0.000 0.000 0.000 0.000 8 9 10 11 12 13 14 8 int. access x 9 int. cut-off 0.000 x 10 distance 0.026 0.000 x 11 public -0.262 0.000 -0.155 x 12 war years -0.308 0.000 0.076 0.027 x 13 newspapers 0.341 0.000 0.172 -0.106 0.094 x 14 coalition 0.402 0.000 -0.140 -0.115 -0.075 0.401 x 3.7 Models
As mentioned at the methodology section, the country level analysis will be done by means of a multiple linear regression. The relationship between a county’s proportion of newspapers covering Libya after three types of civil war events, democracy and internet access, and its dynamics by means of interaction, will be tested according to the following model:
24
where 𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒𝑗 denotes the proportion of newspapers covering Libya after three types of civil war events for country j, 𝐷𝑒𝑚𝑗 is a dummy variable indicating regime type per country and 𝐼𝑛𝑡𝑗 refers to the internet access per country. 𝐷𝑒𝑚𝑗× 𝐼𝑛𝑡𝑗presents the interaction term between regime type and internet access. The second model is established by adding the control variables to the first model and can therefore be specified as follows:
𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒𝑗= 𝛽0+ 𝛽1𝐷𝑒𝑚𝑗+ 𝛽2𝐼𝑛𝑡𝑗+ 𝛽3𝐷𝑒𝑚𝑗× 𝐼𝑛𝑡𝑗+ 𝛽4𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑗+ 𝜀𝑡 (3)
where 𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒𝑗 denotes the proportions of newspapers covering Libya after three types of civil war events for country j, 𝐷𝑒𝑚𝑗 is a dummy variable indicating regime type per country and 𝐼𝑛𝑡𝑗 refers to the internet access per country. 𝐷𝑒𝑚𝑗× 𝐼𝑛𝑡𝑗 presents the interaction term between regime type and internet access. 𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑗 is a vector of the control variables that can influence the proportion of newspaper coverage.
With regards to the county-day level a random effects logistic regression will be performed in order to analyze the effects and interaction effects of regime type, event type and internet access on the publication of an article concerning Libya leading to the subsequent model:
𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑗𝑡= 1 𝑙𝑜𝑔𝑖𝑡⁄ [𝛽0+ 𝛽1𝐷𝑒𝑚𝑗+ 𝛽2𝑇𝑦𝑝𝑒𝑡−1 + 𝛽3𝐼𝑛𝑡𝑗+ 𝛽4𝐷𝑒𝑚𝑗× 𝑇𝑦𝑝𝑒𝑡−1+ 𝛽5𝐷𝑒𝑚𝑗× 𝐼𝑛𝑡𝑗+ 𝛽6𝑇𝑦𝑝𝑒𝑡−1× 𝐼𝑛𝑡𝑗+ 𝛽7𝐷𝑒𝑚𝑗× 𝑇𝑦𝑝𝑒𝑡−1× 𝐼𝑛𝑡𝑗+ 𝜇𝑗+ 𝜀𝑗𝑡] (4)
where 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑗𝑡 refers to the publication of an article concerning Libya for country j, 𝐷𝑒𝑚𝑗 is a dummy variable indicating regime type per country, 𝑇𝑦𝑝𝑒𝑡−1 represents the type of event at day t-1, and 𝐼𝑛𝑡𝑗 concerns the internet access per country. 𝐷𝑒𝑚𝑗× 𝑇𝑦𝑝𝑒𝑡−1, 𝐷𝑒𝑚𝑗× 𝐼𝑛𝑡𝑗,
𝑇𝑦𝑝𝑒𝑡−1× 𝐼𝑛𝑡𝑗 and 𝐷𝑒𝑚𝑗× 𝑇𝑦𝑝𝑒𝑡−1× 𝐼𝑛𝑡𝑗denote the interaction terms between all
independent variables to account for their moderating effects. Lastly, 𝜇𝑗takes into account the country-level random effects. The next model that will be employed incorporates the control variables at the country-day level and has the following structure:
𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑗𝑡= 1 𝑙𝑜𝑔𝑖𝑡⁄ [𝛽0+ 𝛽1𝐷𝑒𝑚𝑗+ 𝛽2𝑇𝑦𝑝𝑒𝑡−1 + 𝛽3𝐼𝑛𝑡𝑗+ 𝛽4𝐷𝑒𝑚𝑗× 𝑇𝑦𝑝𝑒𝑡−1+ 𝛽5𝐷𝑒𝑚𝑗× 𝐼𝑛𝑡𝑗+ 𝛽6𝑇𝑦𝑝𝑒𝑡−1× 𝐼𝑛𝑡𝑗+ 𝛽7𝐷𝑒𝑚𝑗× 𝑇𝑦𝑝𝑒𝑡−1× 𝐼𝑛𝑡𝑗+ 𝛽8𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑗𝑡+ 𝜇𝑗+ 𝜀𝑗𝑡] (5)
where 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑗𝑡 refers to the publication of an article concerning Libya for country j, 𝐷𝑒𝑚𝑗 is a dummy variable indicating regime type per country, 𝑇𝑦𝑝𝑒𝑡−1 represents the type of event at day t-1, and 𝐼𝑛𝑡𝑗 concerns the internet access per country. 𝐷𝑒𝑚𝑗× 𝑇𝑦𝑝𝑒𝑡−1, 𝐷𝑒𝑚𝑗× 𝐼𝑛𝑡𝑗,
𝑇𝑦𝑝𝑒𝑡−1× 𝐼𝑛𝑡𝑗 and 𝐷𝑒𝑚𝑗× 𝑇𝑦𝑝𝑒𝑡−1× 𝐼𝑛𝑡𝑗denote the interaction terms between all
25
control variables that can influence the chances of publication. Lastly, 𝜇𝑗takes into account the country-level random effects. For the last two estimations at the country-day level, the influence of internet will not be estimated by internet access per country, but by the availability of internet within Libya. Therefore 𝐼𝑛𝑡𝑗 will be replaced by 𝐼𝑛𝑡𝑡, where 𝐼𝑛𝑡𝑡 entails internet access in Libya at day t.
3.8 Robustness
For the robustness check of this research, an alternative measure of internet access per country will be implemented to verify if two different measures will yield similar results. A robustness check makes it possible to compare results and check the validity of internet access as an independent variable. This implies that no new hypotheses or regression equation models will be formulated. Instead, all four models will be performed with internet access measured by fixed broadband subscriptions per hundred people instead of a population’s percentage with access. These data are withdrawn from the ‘World Development Indicators’ provided by the World Bank and defined as: “Fixed subscriptions to high-speed access to the public Internet (a TCP/IP connection), at downstream speeds equal to, or greater than, 256 kbit/s. This includes cable modem, DSL, fiber-to-the-home/building, other fixed (wired)-broadband subscriptions, satellite broadband and terrestrial fixed wireless broadband. This total is measured irrespective of the method of payment. It excludes subscriptions that have access to data communications (including the Internet) via mobile-cellular networks. It should include fixed WiMAX and any other fixed wireless technologies. It includes both residential
subscriptions and subscriptions for organizations”.
4. Empirical Results
This section first provides some preliminary overview of differences between non-democratic and democratic countries. This will be followed by the descriptive statistics, which show some of the main characteristics of each variable. The section ends with the regression results and some remarks on the robustness check.
4.1 Dataset
26
and 1.4, it can be seen that the government wounded and killed significantly more civilians than the rebels. Moreover, the spike in the number of casualties by the government coincides with the spike of non-violent protests whereas the spike in the number of casualties by the rebels coincides with the spike of violent war events. This first observation could be due to a governmental response to the non-violent protests being violence, resulting in a sizeable amount of civilian casualties. The second observation of coinciding event-spikes could be caused by significant high levels of rebel violence at the time, leading to both an increase in violent war events in general and rebel inflicted civilian casualties.
Figure 1: Timeline events
Source: Baum and Zhukov (2015)
When examining differences between regime type, as presented by figure 2.1 and 2.2, it can be stated that for most variables a democracy has higher values than a non-democracy on average. Hence, according to figure 2.1 on average democracies tend to have more
publications concerning Libya and have a higher proportion of newspapers covering Libya after violent protests and government inflicted civilian casualties, whereas
non-democracies have a slightly higher proportion of newspapers covering Libya after rebel 0 1 2 3 4 5 6 7 8 Dec -1 0 Jan -1 1 Feb -1 1 Ma r-1 1 Ap r-1 1 Ma y -1 1 Ju n -1 1 Ju l-1 1 Au g -1 1 Sep -1 1 Oct-1 1
Figure 1.1: Non-violent protests
0 5 10 15 20 25 30 Dec -1 0 Jan -1 1 Feb -1 1 Ma r-1 1 Ap r-1 1 Ma y -1 1 Ju n -1 1 Ju l-1 1 Au g -1 1 Sep -1 1 Oct-1 1
Figure 1.2: Violent war events
0 100 200 300 400 500 600 700 Dec -1 0 Jan -1 1 Feb -1 1 Ma r-1 1 Ap r-1 1 Ma y -1 1 Ju n -1 1 Ju l-1 1 Au g -1 1 Sep -1 1 Oct-1 1
Figure 1.3: Casualties by government
0 5 10 15 20 25 30 35 40 Dec -1 0 Jan -1 1 Feb -1 1 Ma r-1 1 Ap r-1 1 Ma y -1 1 Ju n -1 1 Ju l-1 1 Au g -1 1 Sep -1 1 Oct-1 1
27
inflicted civilian casualties. This is in line with the first two theorized hypotheses concerning regime based media bias.
Also, on average democracies have a higher proportion of newspapers covering Libya after violent events in general. But in combination with a notable higher average number of newspapers, as shown by figure 2.2, and a lower average percentage of publicly owned newspapers, as shown by figure 2.1, this could also be due to better developed media markets in democracies. Furthermore, when reviewing the remaining variables regarding differences in regime type in figure 2.1 and 2.2, it can be said that on average the number of conflict years since 1945 are higher in non-democracies. On the other hand, on average the population’s percentage with internet access and number of NATO intervention coalition participants are considerably higher in democracies. It seems logical for democracies to have experienced less internal war years, have a better developed infrastructure to support internet access and to supply more coalition partners to the NATO military intervention as being a democracy is a key requirement for NATO membership (Qatar and the United Arab Emirates being the exceptions in this coalition).
Figure 2: Regime type characteristics
Source: Baum and Zhukov (2015)
4.2 Descriptive Statistics
As two different datasets will be employed for this research there are two tables containing descriptive statistics of the data, one at the country level and one at the country-day level. The research sample consists of observations for 113 countries over 311 days. Yet, as in the country-level dataset one country, North Korea, is missing in the replication files, the first
0 0,1 0,2 0,3 0,4 0,5 0,6 Figure 2.1 Non-democracy Democracy 0 10 20 30 40 50 60
Internet Waryears Newspapers Figure 2.2
28
dataset contains 112 observations for most variables. As for Kosovo the values of distance and population’s percentage with internet access are missing these variables contain 111 observations. So without missing values, there are 111 complete which observations will be employed for analysis at the country-level.
Table 3: Country-level descriptive statistics
Obs Mean Std. Dev. Min Max
Protest coverage 112 0.151 0.172 0 0.882 Casualties by rebels coverage 112 0.169 0.179 0 0.889 Casualties by government coverage 112 0.182 0.186 0 0.933
Democracy 112 0.518 0.502 0 1 Internet access 111 41.066 28.523 0.9 92.3 Public ownership 112 0.118 0.229 0 1 Distance 112 5727.083 4089.558 0 18614.51 Conflict years 112 9.027 13.973 0 63 Number of papers 112 13.661 27.779 2 193 Civil war violence coverage 112 0.162 0.165 0 0.907 Participation coalition 112 0.134 0.342 0 1
The country-day level dataset contains 35.143 observations for most variables. For internet access and distance there are again missing values for Kosovo. For North Korea internet access also has missing values, as well as the number of newspapers. For the lagged
dependent variable and the type of conflict variables there are 113 missing values, all across different countries. For the analyses without control variables this means there are 35.030 and 34.410 complete observations while for the analyses with control variables there are 34.410 complete observations.
Table 4: Country-day level descriptive statistics
Obs Mean. Std. Dev Min Max
Publish 35143 0.421 0.494 0 1 Protest t-1 35030 0.113 0.599 0 7 Casualties by rebels t-1 35030 0.432 3.093 0 36 Casualties by government t-1 35030 5.623 40.423 0 600 Democracy 35143 0.513 0.5 0 1 Internet access 34521 41.066 28.394 0.9 92.3 Internet Libya 35143 0.45 0.498 0 1 Public ownership 35143 0.117 0.227 0 1 Distance 34832 5783.172 4095.793 0 18614.51 Conflict years 35143 8.947 13.875 0 63 Number of papers 34832 13.661 27.655 2 193 Civil war violence t-1 35030 2.252 2.955 0 24 Participation coalition 35143 0.133 0.339 0 1
29 4.3 Regression Results
Table 5 shows the regression results for the different models that were analyzed at the country level with multiple linear regression. Overall it can be seen that no real significant
relationships were identified. Just two of the control variables, namely the number of newspapers and coalition participation, show significant coefficients after all three types of civil war events in models (2), (4) and (6). And only after rebel inflicted civilian casualties, shown by model (5) and (6), internet access shows a slight positive significant effect on the proportion of newspapers covering Libya. After protests and with control variables, in model (2), internet has a slight significant positive effect on proportion of newspapers in
democracies covering Libya after non-violent protests. The significant F-statistic of all models implies that the coefficients of the regression estimation are jointly significantly different from zero. Yet, the low R-squared value implies that only a small amount of the change in the proportion of newspapers covering Libya after one of the three types of events, is explained by the independent variables and therefore no real conclusions can be drawn from these results.
Table 5: Linear regression results
Dependent variable: 𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒𝑗 proportion of newspapers publishing an article
Model (1) (2) (3) (4) (5) (6) after protests after civilian casualties by government after civilian casualties by rebels constant 0.068 (2.12)** 0.083 (2.02)** 0.099 (2.80)*** 0.098 (2.15)** 0.117 (3.23)*** 0.139 (2.88)*** Democracy 0.039 (0.68) 0.014 (0.26) 0.049 (0.77) 0.018 (0.30) -0.035 (-0.54) -0.058 (-0.91) Internet access 4.27e-04 (0.47) 1.82e-04 (0.21) 0.001 (0.52) 3.43e-04 (0.35) 0.002 (2.25)** 0.002 (1.87)* Democracy * Internet 0.002 (1.40) 0.002 (1.68)* 0.001 (1.04) 0.002 (1.31) -0.001 (-0.59) -0.001 (-0.49) Distance to Libya 1.07e-06 (0.30) 3.15e-06 (0.79) 1.84e-06 (0.44) Public ownership -0.076 (-1.16) -0.078 (-1.07) -0.094 (-1.20) War years since 1945
30
Table 6: Logistic regression results
Dependent variable: 𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑗𝑡 (publication of an article on Libya)
Model (7) (8) (9) (10)
Percentage internet access Internet access in Libya constant -1.889 (-5.15)*** -2.047 (-6.71)*** -0.548 (-1.91)* -1.314 (-4.19)*** Democracy 0.124 (0.19) 0.232 (0.48) 0.630 (1.57) -0.059 (-0.20) Internet 0.040 (3.82)*** 0.027 (3.44)*** -0.902 (-20.67)*** -0.583 (-12.52)*** Democracy * Internet -0.013 (-0.93) -0.018 (-1.83)* -0.151 (-2.43)** -0.120 (-1.84)* Type: Protests t-1 -0.135 (-2.12)** -0.132 (-2.00)** -0.395 (-2.69)*** -0.240 (-1.54) Dem * Protests t-1 0.178 (1.61) 0.138 (1.17) -0.037 (-0.18) -0.310 (-1.39) Internet * Protests t-1 3.12e-04 (0.20) -1.63e-04 (-0.10) 0.383 (2.52)** 0.165 (1.03) Dem * Internet * Protests t-1
0.003 (1.53) 0.003 (1.06) 0.469 (2.17)** 0.649 (2.81)*** Type: Casualties by government t-1
0.001 (1.68)* 1.92e-04 (0.21) 6.98e-05 (0.04) -0.005 (-2.54)** Dem * Casualties (G) t-1 1.14e-04 (0.07) -6.77e-04 (-0.39) 0.007 (2.56)** 0.006 (1.95)* Internet * Casualties (G) t-1 3.71e-05 (1.39) 2.61e-05 (0.93) 0.003 (1.35) 0.007 (3.28)*** Dem * Internet * Casualties (G) t-1
3.62e-05 (0.95) 3.67e-05 (0.94) -0.006 (-1.85)* -0.004 (-1.53) Type: Casualties by rebels t-1
0.016 (1.64) 0.015 (1.42) 7.38e-05 (0.01) 0.002 (0.27) Dem * Casualties (R) t-1 -0.015 (-0.79) -0.10 (-0.51) -0.001 (0.12) 0.005 (0.55) Internet * Casualties (R) t-1 2.94e-05 (0.11) -8.84e-05 (-0.30) 0.208 (3.34)*** 0.104 (1.57) Dem * Internet * Casualties (R) t-1
3.19e-04 (0.82) 4.05e-04 (0.97) 0.078 (0.85) 0.037 (0.39) Distance to Libya -3.16e-05 (-1.01) 2.94e-05 (-0.86) Public ownership -0.721 (-1.23) -0.752 (-1.19) War years since 1945
0.007 (0.75) -0.003 (-0.31) Number of newspapers 0.021 (3.97)*** 0.024 (4.37)*** NATO coalition participation
0.804 (1.95)*
1.154 (2.64)*** Civil war violence
0.123 (22.35)*** 0.107 (19.04)*** Publish t-1 1.606 (50.15)*** 1.514 (46.55)*** Var(µj) 1.764 1.240 2.113 1.361 N 34410 34410 35030 34410
Wald chi2 statistic 196.14*** 3482.91*** 1193.36*** 3690.62*** Chibar2 statistic 1.2e+04*** 3529.26*** 1.6e+04*** 4089.06***
31
Table 6 shows the regression results for the different models that were
analyzed at the country-day level with random effects logistic regression. Because the original logistic coefficients are listed, it is only possible to interpret directionality and not the
magnitude of the relationship between the dependent and independent variables.
First of all it can be noted that only internet access shows significant coefficients in all four models, yet in models (7) and (8) the coefficient is positive, whereas in models (9) and (10) it is negative. Hence internet access in the home country positively effects article
publications on Libya while internet availability in Libya does not. A positive effect on article publications is to be expected due to internet access allowing higher levels of event driven news, increased direct feedback and more competition. The negative direction seems contradicting, but perhaps the negative effect of Libyan internet availability could be explained by the fact that the internet cut-off in Libya generated more coverage, because of the actual cut-off, than when internet was still available. For the same reason it could be that model (9) and (10) also demonstrate significant negative coefficients for the interaction between democracy and internet availability. This is also the case for model (8) in which home-country internet access in democracies has as negative effect on publications, perhaps because people rather search for news information online than in newspapers.
Non-violent protests in general show significant negative coefficients in models (7), (8) and (9), which could be due to the fact that the lack of violence makes such an event less newsworthy as it does not contain unexpected or large and dramatic events. Furthermore, for the models regarding home-country internet access just one other significant positive
coefficient is identified, namely government inflicted civilian casualties. Yet, as model (7) does not include control variables these estimations could be overstated. And contradictory this variable actually exhibits a positive significant coefficient in model (10). Therefore no real conclusion on the influence of government civilian casualties can be drawn.
32
enhances rebel communication on non-violent protests towards the outside world, thereby positively influencing publications containing stories on Libya.
Other variables that exhibit significant coefficients are the interaction of Libyan internet access with non-violent protests and its interaction with rebel caused civilian casualties in model (9), and its interaction with government caused civilian casualties in model (10). All show positive coefficients, indicating that internet access in combination with events positively influences newspaper publications concerning Libya. This is in line with the theory indicating that technological developments, such as internet, establish a favorable environment for event-driven news. Moreover, model (9) contains one more variable with a significant negative coefficient namely the effect of government caused civilian casualties and internet access in Libya on publications in democracies. This result is also in line with the theory, as it argued that in democracies internet access will cause the selection bias towards government violence to decrease as internet enhances the chances of other news stories also reaching audiences.
Lastly, the coefficients of the control variables show that again the two same country-level control variables, the numbers of newspapers and NATO coalition participation, have significant effects on the dependent variable. Where the number of newspapers had a negative coefficient in the multiple linear regression models, here it exhibits a positive coefficient which is more in line with the competition effect mentioned at the control variables section. NATO coalition participation has a positive coefficient in both types of regression which follows from newsworthiness determined by values concerning proximity and impact, and ‘rally-round-the-flag’ effects. The two extra, time-variant, control variables added in the logistic regression regarding violent civil war events and the one-day lagged dependent variable both show significant positive coefficients. Hence, the number of violent war events and the continuity effects both positively affect the publication of articles on Libya. Direct governmental control of a news agency through public ownership, proximity measured by geographical distance and impact due to the number of conflict years of the home country do not affect the publication of newspaper articles concerning Libya.
4.4 Robustness
33
identified; the same control variables seem to have significant coefficients in the same directions as does internet access after rebel inflicted civilian casualties. The only small differences are that in model (12) the interaction effect is insignificant while in model (13) the effect of democracy on the proportion of newspapers covering Libya after government
civilian casualties demonstrates a small significant positive value. Yet, as R-squared also shows similar low values again no real conclusions can be drawn from these results.
For the random effects logistic regression at the country-day level the robustness check yields some different results from the initial regressions shown by Table 9 in Appendix C. Most variables that show significant coefficients in models (16) and (17), also showed similar significant coefficients’ signs in models (7), (8), (9) and (10). These variables include broadband internet access, democracy interacted with broadband internet access, non-violent protests in both models (16) and (17), and government inflicted civilian casualties in model (17), which is the model including control variables. Moreover, the control variables also exhibit similar results in this robustness check. The robustness checks differs from the initial regressions with regards to the effect of rebel inflicted civilian casualties on publication in general and the effect of non-violent protests on publication in democracies in both model (16) and (17). The former effect exhibits a positive significant coefficient indicating that violent events in general can influence publication on Libya’s civil war. The latter effect has significant positive coefficients suggesting that democracies tend to have a selection bias towards non-violent protests.
5. Discussion
When examining the regression analyses the results contain multiple theoretical implications. When discussing the results, it is referred to the logistic regression results, as the multiple linear regression results did not indicate any real significant relationships. Firstly, when evaluating the first two hypotheses concerning regime based selection bias only slightly supportive evidence is found concerning the event type of government caused civilian casualties positively influencing publication in democracies. Hence, Baum and Zhukov’s (2015) results regarding regime based reporting biased are not confirmed by this research, even though a similar dataset was employed for analysis. However, besides utilizing regression analyses, Baum and Zhukov (2015) draw most of their conclusions based on the construction of counterfactual scenarios.
