1
Media Ownership and News Coverage of International Conflict
Matthew A. Baum
John F. Kennedy School of Government, Harvard University
Yuri M. Zhukov
Department of Political Science, University of Michigan
Abstract: How do differences in ownership of media enterprises shape news coverage of international
conflict? We examine this relationship using a new dataset of 598,981 articles on U.S.-led multinational
military operations in Libya, Iraq, Afghanistan and Kosovo, published by 2,765 newspapers in 110 countries. We find that independent newspapers offer more extensive coverage and place greater emphasis on
military and policy issues (hard news). Media outlets within larger ownership networks display the opposite patterns – less frequent coverage and more emphasis on human interest and personalities (soft news).
Newspapers owned by the same parent company also tend to feature very similar types of news coverage.
Our data suggest that the consolidation of media outlets into large, multinational conglomerates may
leave citizens with less information to monitor or influence their leaders’ foreign policy activities.
1
The authors are grateful to the Weatherhead Center for International Affairs for financial and in-
stitutional support, and to Barbara Halla, Jane Lim, Raul Quintana, Katie Towt, Disha Verma and Gretchen Yuan for research assistance.
Rationalist theories of international conflict rest on the proposition that the efficient flow of information –
between political leaders and their domestic audiences, as well as between states involved in disputes –
can mitigate the prevalence of war (Fearon 1994, 1995, Lake & Rothschild 1996, Smith 1998, Schultz
2001). Yet models of domestic politics have long challenged the possibility of a perfectly informed world
(Downs 1957: 213). Citizens face incentives to transfer the costs of procuring, analyzing and evaluating
information to third parties, like the news media. Because journalists can gather and report only a small
subset of potentially innumerable data points at any given time, the resulting exchange of information
becomes inherently selective and biased. The scope and content of media coverage may reflect corporate
preferences, market incentives, the constraints of the political and economic environment, as well as the
day-by-day unfolding of events. The resulting variation in news coverage mediates the credibility, transparency and availability of information in the public domain (Strömberg 2004, Gentzkow & Shapiro
2004, Gentzkow 2006, DellaVigna & Kaplan 2007, Gerber et al. 2006).
Until recently, however, the onerous data requirements of comparative media research have impeded our ability to uncover sources of variation at competing levels of analysis: within and between
countries, within and between individual media outlets, and over time. We investigate the role of media
ownership in news reporting of international conflict, using new disaggregated data on media coverage of
the 2011 NATO-led intervention in Libya, the 2003 U.S.-led invasion of Iraq, the 2001 U.S.-led operations in Afghanistan, and the 1999 NATO-led intervention in Kosovo. We endeavor to explain two outcomes: (1) daily decisions to publish a news story on the crisis (scope), and (2) the type of coverage given
to the story (focus) – a “soft news” emphasis on human interest and personality-oriented stories (e.g., the
relationship between a chief executive and her cabinet or stories about soldiers’ family members), or a
“hard news” focus on military and policy questions.
We find that ownership structure and newspaper-level attributes have a profound impact on the
volume and content of news coverage. Independent newspapers were significantly more likely to report
2
on each crisis than their media conglomerate counterparts, and were more likely to publish stories on hard
news issues of military operations and policymaking. The likelihood of crisis reporting and hard news
coverage was also greater among high circulation newspapers than among their smaller, regional and local counterparts.
Contrary to previous research on American newspapers (Gentzkow & Shapiro, 2007), our crossnational results also uncover compelling evidence of ownership network effects: ceteris paribus, newspapers owned by the same parent company tend to adopt very similar frames and perspectives. Our data
show that the increasing consolidation of media outlets into large, multinational conglomerates may reduce information diversity, providing citizens with less information to monitor or influence their leaders’
foreign policy activities.
Our study is organized as follows. Section 1 reviews existing research on comparative media systems and political communication, and derives several hypotheses on the determinants of foreign policy
coverage for different media outlets. Section 2 describes our data on newspaper coverage of international
conflict and media ownership. Section 3 examines the empirical relationship between coverage, ownership structure and a range of other covariates at the newspaper-day level. Section 4 evaluates these results
in the context of broader academic and policy debates on media ownership and coverage, summarizes our
findings, and identifies several directions for future research.
Press Ownership and News Coverage
Rational choice theories of international conflict typically hold that the prevalence of war depends on the transparency, reliability, and availability of information to actors involved in disputes
(Fearon 1995, Lake & Rothschild 1996). The literature on domestic sources of foreign policy in general –
and domestic audience costs in particular (Fearon 1994, Schultz 2001, Smith 1998) – emphasizes information credibility as helping to determine which inter-state disputes escalate to violence and which are
3
resolved peacefully. Despite the centrality of information to extant theories of war, international relations
scholars have largely ignored the process by which states disseminate information within and between
themselves. Most research assumes that information – and any credibility or transparency it conveys –
passes efficiently from leaders’ mouths or actions to the intended recipients. If so, the only remaining
uncertainty – which underpins much of the formal conflict literature – concerns what information a leader
transmits or withholds and whether or not the intended recipient views it as reliable. Where such information passes through an intermediary with its own strategic incentives (Baum and Groeling 2009), however, this assumption seems problematic.
Citizens primarily learn about their governments’ activities via the mass media. The acquisition
of information is an individually costly enterprise, and citizens face incentives to transfer these costs to
someone else (Downs 1957). This “outsourcing” raises the questions of whether and how the media – as
the primary intermediary for information transmission between citizens and leaders – influence states’
behavior in international conflicts. Existing research on this question (Van Belle 2000, Slantchev 2006,
Choi & James 2007) mostly emphasizes the possibility that a free press might facilitate peaceful conflict
resolution, by raising the domestic political costs to leaders of fighting wars. Yet in order to draw such
conclusions, it is necessary to consider whether and how different media institutions vary in their coverage of foreign policy. Do all media report on the same events, and write roughly the same things about
them? Or do they diverge in systematic ways, with potential consequences for public support of leaders’
foreign policy actions? What are the sources of such divergence?
Previous research (Iyengar 1991) has shown that different types of news can engender quite distinct public responses. For instance, hard news-oriented, thematically framed reporting, with an emphasis
on public policy themes and the broader political, or military context – tends to engender a sense of collective responsibility for a given policy problem. This, in turn, raises the likelihood that consumers will
look to the government for a solution.
4
By contrast, soft news-oriented, episodically framed reporting, with an emphasis on the experiences, characteristics, trials, and tribulations of specific individuals or small groups, tends to push consumers to attribute a problem’s cause and solution to the individual or group, rather than state.2 Such coverage may pull on peoples’ heart strings, but is less effective at generating demand for public action, or
for a government or societal response (Iyengar 1991). Hence, the framing of a problem – including a foreign policy problem – can influence whether or not the public supports government intervention to address it, and whether it looks favorably on the government’s design and implementation of its foreign policy more broadly (Strömberg 2004; Gentzkow & Shapiro 2004; Gentzkow 2006; DellaVigna & Kaplan
2007; Gerber et al. 2006).
A growing body of research indicates that this process can, at least under some circumstances, influence foreign policy decisions, typically by constraining leaders’ perceived freedom of action (Baum
2013; Sobel 2001 and 2012). By shaping public and elite opinion, the media play a key intervening role
between leaders seeking to build or sustain support for their preferred policy initiatives and the citizens
they need to persuade.
Our goal is to pull back a few steps in this causal story, and explain why and when certain media
institutions choose to (a) cover a foreign policy crisis in the first place, (b) present the story from a soft or
hard news perspective. In answering these questions, we hope to gain insight into the circumstances under
which the media can influence public opinion on foreign policy, as well as the likely nature of such influence and its implications regarding public support for leaders’ foreign policy initiatives.
Cross-national political science research on such questions is a rapidly developing field (Hallin &
2
We define “soft news” as a set of story characteristics, including the absence of a public policy
component, sensationalized presentation, human-interest themes and emphasis on dramatic subject matter
(Patterson 2000)
5
Mancini 2004, Iyengar et al. 2010), but has until recently been limited by data constraints. Within this
literature, most explanations of the sources of media coverage reside at the level of national political and
economic attributes, like party systems, wealth and education. While these aggregate characteristics are
surely important, an emphasis on cross-national differences leaves much subnational variation unexplained. Aggregate explanations describe the environment in which media organizations operate, and the
broader systemic constraints they face. Yet two media organizations may not navigate the same environment in a uniform fashion, and similar environmental conditions could produce quite different outcomes
in the nature and extent of media coverage. Family-owned newspapers, for instance, face very different
resource constraints and audiences than those owned by governments or multinational media conglomerates, and these differences may shape editorial choices about whether and how to cover particular stories.
In addition to country-level political and economic characteristics, many more immediate factors
shape the incentives media outlets face and strategies they adopt. Their ownership structure, the size of
their audience, and the daily editorial choices made by co-owned media organizations are three potentially important sub-national factors that we investigate in the present study.
Ownership structure
Communication scholars (Bagdikian 2000, McChesney 2000, Herman & Chomsky 2002) have
long worried that concentrated private ownership of media organizations stifles public debate. When
newspapers are independently owned, they are relatively free to follow the whims of individual owners,
which may be driven by profit, ideology, a sense of civic duty, or any combination thereof. They are typically the public’s primary sources for news about local or regional issues, like crime, the arts, local government, taxes and social services (Rosenstiel et al. 2011). While it is certainly possible that such outlets
might be hesitant to criticize a particular national government, perhaps for ideological reasons, their profitability is primarily locally determined and hence less directly affected by national public policy. When,
6
however, they are part of larger networks of media outlets, they are increasingly subject to the sorts of
bottom-line pressures facing other corporate conglomerates. Herman thus argues:
“Private owners, especially those of major media, are likely to favor markets, of which
they are a part and of which they are major beneficiaries. There is some dispute about the
extent to which owners influence media behavior and performance. At a minimum…the
controlling management will face strong pressure to focus on the bottom line. This in itself has policy implications, as such a focus implies catering to advertisers, cultivating relationships with dominant information sources, and avoiding conflict with other powerful
constituencies” [emphasis added]. (Herman, 2002: 64)
Both additional research (DiMaggio 2009, Herman and Chomsky 2002, Kellner 2005) and anecdotal evidence tend to support Herman’s argument. In the latter case, large media conglomerates, like Rupert Murdoch’s Sky TV Network, have consented to modify their news content – indeed, Sky TV created a
unique, dedicated channel to satisfy China’s media censorship laws in exchange for limited broadcast
rights in China (Flitton, 2011). In the former case, for instance, DiMaggio (2009) studied news coverage
of the 2003 Iraq War and found that corporate mass media outlets focused less on public interest-oriented
coverage and more on maximizing profits. One consequence is a narrower range of voices presented in
the media (Jenkins 2004). An example, again in the case of the Iraq War, is the general failure of corporate media to cover anti-war perspectives and to disproportionately emphasize soft news stories, which
they tended to view as having greater “news value” and hence as more profitable (Shinar 2003). Herman
and Chomsky (2002), in turn, find that if a given news media outlet angers the government, it can be subtly excluded from access to important information. Because this can hurt an outlet’s bottom line, it creates
an incentive to minimize the financial threat by supporting (editorially and in news reporting) government
7
and corporate policies. Finally, Kellner finds that corporate media played a consistent and consequential
role in promoting the war and anti-terrorism policies of President George W. Bush.
The empirical implication is that larger media conglomerates are more likely than their independently owned counterparts to emphasize relatively less controversial (and more profitable) soft news
over more controversial (and hence risky) hard news. This, in turn, implies less criticism of government
policy. After all, such companies are particularly beholden to government officials for favorable market
policies as well as for authoritative information about national or international policy (Bennett 1990).
Hence, the interests of ownership within larger media networks are more likely to reflect an emphasis on
entertainment over politics (Davis & Owen, 1998; Zaller, 1999), and avoidance of conflict with powerful
governmental interests (Herman and Chomsky 2002).
Further, in order to exploit the market power that economies of scale provide, outlets within a
larger network are likely to specialize, rather than attempt to appeal to the median consumer. This is analogous to the logic of the median voter theorem, which holds that in a two party system with a single leftright ideological dimension and single-peaked preferences, parties will tend compete for the median voter, who will be located in the center of the political spectrum. However, as the number of parties increases, each party has an increasing incentive to locate itself in a distinct space along the left-right political
dimension, as they can no longer all compete effectively for the median voter. In other words, as the
number of parties increases, parties become more ideologically specialized in order to maximize their
vote shares.
The analogy to media content is straightforward: as the number of outlets within a network increases, the incentive for each individual network member to compete for the median consumer recedes.
Such a strategy would see network co-members competing with each other for the same readers. Instead,
a network can optimize its total audience by having individual outlets within the network specialize – geographically or substantively – and offer content that appeals to individuals all along the “media prefer-
8
ence” dimension. An upshot of this specialization is that niche-oriented newspapers face fewer incentives
to report on general-interest topics, like public policy and war. Our first set of hypotheses follows.
•
H1: As the ownership network within which a newspaper is located increases in size, the newspaper will offer (a) less coverage of international conflict, and (b) less hard news and more soft news.
Additional research suggests that editorial (Gilens and Hertzman, 2000, Snider and Page 2003a)
and news content – both print (Snider and Page 2003a) and broadcast (Snider and Page 2003b) – tend to
follow the economic interests of media ownership. For instance, in their study of the 1996 Telecommunications Act, Snider and Page (2003b) found that 100% of editorials on the Act offered by newspapers
whose parent companies earned a high percentage (at least 20%) of their revenue from broadcast television supported granting free broadcast spectrum to TV broadcasters. Conversely, 100% of editorials on
the same topic presented by their counterparts whose parent companies earned lower percentages (less
than 20%) of their revenues from broadcast TV editorialized against the policy.
Such ownership effects extend beyond editorial content. Research shows that increases in ownership concentration – that is, control of a larger number of outlets by fewer owners – leads to reduced diversity in content across a variety of media (Peterson and Berger 1975 and 1996, Bielby and Bielby 2003,
Bagdikian 1985, Jenkins 2004). Affiliated outlets, in turn, are likely to make editorial choices more similar to outlets within the network than outlets located within other networks, or independent outlets.
An expectation of within-network similarity extends logically from our first hypothesis. If a
newspaper belongs to a large network – that is, a network containing many newspapers – it is unlikely to
devote much page space to foreign policy or hard news, since it will feel pressure to pursue a niche media
market strategy. Other papers within the same large network will face the same market incentives, and
will be unlikely to cover foreign policy or have a hard news focus for the same reason. Because two
9
newspapers belonging to the same large network will likely both be niche papers, they should both have
lower scope and less hard news. By the same logic, two newspapers belonging to the same small network
– that is, one containing relatively few newspapers – should both have greater scope and more hard news.
They simply have less incentive to pursue niche strategies. More generally, however much foreign policy
coverage they offer and whatever the prevailing focus, newspapers owned by the same parent company
should produce very similar types of coverage. Our second hypothesis follows.
•
H2: The content of newspapers within the same ownership network is likely to be more similar in
(a) scope and (b) focus than is the content of newspapers in separate ownership networks.
Circulation
The size and composition of a newspaper’s audience, traditionally captured by its circulation, is
likely to influence news content in important ways (Gentzkow & Shapiro 2007). The political economy
literature frequently employs aggregate newspaper circulation statistics as a proxy for the volume of information available to the public (Besley & Burgess 2002; Adsera et al 2003; Keefer 2007). The argument is that larger newspaper circulation implies a better-informed population. This logic, however, has
more to do with presumed exposure than assumptions about content.3
Zaller (1999) offers a different theoretical perspective that more directly links audience magnitude with news content. Arguing from the logic of Downs (1957), Zaller develops his “Theory of Media
Politics” based on the notion that market size drives the relative emphasis on hard vs. soft news content.
3
While identification of an independent circulation effect is rendered difficult by the endogeneity
of this variable to state policy, demographics and a host of other environmental conditions, its immediate
impact on daily editorial decisions is more theoretically tractable.
10
He finds in a series of empirical investigations that in larger and more competitive markets, TV and
newspaper outlets have a more limited hard news focus than in smaller and less competitive markets, and
that each new entrant into a market tends to have a more limited hard news focus than the existing market
actors. The reason is that viewers tend to prefer softer news than journalists prefer to provide. Hence,
news outlets tend to provide the highest hard-to-soft news ratio that the market will tolerate.
Increased competition thus drives news down-market, as new entrants compete for viewers who
want less hard news than the existing market provides. Consequently, as a media market fragments, with
new entrants competing for smaller niches of the overall audience, they will tend to offer consumers relatively more – and more profitable (Shinar 2003) – soft news and relatively less hard news. To the extent
circulation reflects market share, outlets with larger circulations should offer relatively more hard and less
soft news than their smaller counterparts.
Finally, higher-circulation newspapers tend to have greater financial resources, larger staffs, more
pronounced international footprints and, by extension, greater capacity to produce original foreign policy
news coverage (Chang & Lee 1992, Markham 1961). According to the Newspaper Association of America4, circulation accounted for 27% of aggregate US newspaper revenue, well above advertising and second only to the cover price of print editions. The implication is that as circulation rises, so too do revenues, and hence resources. Along these lines, Soroka (2003) finds that larger newspapers, like the New
York Times in the United States and The Times in the United Kingdom, offer more foreign policy coverage than their smaller counterparts, which tend to focus more on domestic news. This combination of
market incentives and resources yields our final set of predictions:
4
http://www.naa.org/trends-and-numbers/newspaper-revenue/newspaper-media-industry-revenue-
profile-2012.aspx
11
•
H3: All else equal, as circulation increases, newspapers will offer (a) more coverage of international
conflict, and (b) more hard news and less soft news.
News Coverage and Media Ownership Data
To test these propositions, we introduce new data on news coverage during four recent international security crises: (1) the 2011 Libyan uprising and NATO-led intervention, (2) the 2003 U.S.-led invasion of
Iraq, (3) the 2001 U.S.-led operations in Afghanistan, and (4) the 1999 NATO-led intervention in Kosovo. Beyond their historical significance as instances of major power war in the post-Cold War era, these
cases represent three distinct types of armed conflict: third-party interventions (Libya, Kosovo), a preventive war (Iraq), and a war of retaliation (Afghanistan). Their selection enables us to examine how patterns
of news coverage vary across and between different types of conflict, and over time. The data are based
on a corpus of 598,981 international newspaper articles published in the weeks and months before and
after the launch of military operations.5 While news coverage appears in various forms of electronic and
print media, we confine our current focus to newspapers due to their international prevalence as primary
sources of information on political, economic and social events, and our ability to collect a consistent and
representative data sample across the largest possible set of countries.
Figure 1 shows the countries included in each of the four datasets. For each country, we conducted a census of all daily and weekly newspapers listed in the electronic databases Lexis-Nexis and ISI
Emerging Markets. We identified a universe of 2,765 unique and active (i.e. in press at time of conflict)
5
The Libya corpus includes 198,291 articles published by 2,031 newspapers in 102 countries. The
Iraq corpus includes 278,361 articles published by 2,310 newspapers in 63 countries. The Afghanistan
corpus includes 79,321 articles published by 2,189 newspapers in 52 countries. The Kosovo corpus includes 43,008 articles published by 1,813 newspapers in 33 countries.
12
newspapers, excluding weekend supplements, inserts, evening editions and associated materials.
[Figure 1 about here]
Ownership network data
Using a combination of industry organization listings (e.g. Audit Bureau of Circulations), international news media guides (e.g. Mondo Times), financial databases (e.g. WorldScope), annual company
reports and the websites of individual news organizations and their parent companies, we collected information on each newspaper’s average daily circulation figures and ownership. We were able to find circulation data for 2272 (82%) of the newspapers in our data set and successfully identified the primary
shareholders of all 2765.6 Based on these data, we constructed the ownership network shown in Figure 2,
where red dots represent parent companies and light blue squares represent individual newspapers. For
each of the four conflicts, we modified the network to account for the entry and exit of newspapers, and
occasional changes in the ownership structure.7
[Figure 2 about here]
We used these ownership data to calculate network-lagged versions of coverage variables (discussed below), as well as measures of network size (number of other newspapers owned by each media
company). We define an independent newspaper as a “singleton” in the network, with no other newspa6
In each of the four cases, we collected circulation and ownership data for the most recent year
with available data. Following La Porta et al. (1999) and Djankov et al. (2001), we identified legal entities
(families, corporations, holding companies, political parties, governments) that own majority voting
stakes in each newspaper.
7
The 2011 network includes 2,031 newspapers and 789 owners; the 2003 network includes 2,310
newspapers and 840 owners; the 2001 network includes 2,189 newspapers and 809 owners; the 1999
network includes 1,813 newspapers and 594 owners.
13
pers owned by its parent company (network size of 0). We define a corporate-owned newspaper as one
that shares a parent company with at least one additional newspaper in the network.
Our sample includes 122 (6%) publicly owned newspapers, 555 (26%) privately owned independent newspapers and 1438 (68%) corporate newspapers.8 Multinational conglomerates own 775 (28%)
newspapers, with newspaper holdings in multiple countries. Independent outlets constitute 100 percent of
all newspapers in 23 countries, but countries with the highest absolute numbers are the U.S. (89, or 27%),
Russia (76, or 36%), and the U.K. (51, or 9%). Countries with the most state-owned newspapers include
Russia (62, or 30%), Jordan (5, or 36%), Egypt (5, or 29%), and Tunisia (4, or 44%). Corporate-owned
newspapers dominate the market in the U.K. (485, or 84%), U.S. (213, or 64%), the Netherlands (69, or
48%) and Australia (68, or 54%).
Foreign policy news coverage data
For each newspaper, we collected every unique article archived in Lexis-Nexis or ISI, containing
the terms “Libya,” “Iraq,” “Afghanistan” or “Kosovo” (in English or the newspaper’s source language)
and published within a distinct time window for each conflict, spanning the weeks and months immediately prior to and following the beginning of U.S.-led military operations.9 The text corpus was multilin-
8
We define a newspaper as independent if it shares a parent company with no other paper.
9
For Libya, the time window was between 18 December 2010 and 20 October 2011. These dates
mark, respectively, the day of first protests in Tunisia following Mohamed Bouazizi’s self-immolation –
generally accepted as the beginning of the Arab Spring – and the death of Muammar Gaddafi. For Iraq,
the window was between 20 December 2002 (three months prior to military operations) and 28 April
2003 (three weeks after the fall of Baghdad). For Afghanistan, the window was between 10 September
2001 (the day preceding the 9/11 terrorist attacks) and 27 October 2001 (three weeks following the launch
14
gual, including articles in the native language and – where available – in English. Where the articles were
in a language other than English, we used statistical machine translation (Google Translate API) to convert them to English.10
We used supervised learning methods to automatically classify the content and tone of each article according to a sample of reference texts coded by a team of research assistants (see online appendix).
Specifically, we used wordscores, a supervised learning method used to locate statements in a predetermined issue space (Laver et al., 2003; Lowe, 2008). The algorithm uses information from texts
whose positions on some policy dimension are assumed known (“training set”) to learn about a second set
of texts whose policy positions are unknown (“test set”). The training set is used to generate a score for
each word, measuring the relative rate at which that word appears in each training text. These scores are
then used to scale the documents in the test set, by taking a frequency-weighted average score of the
words they contain.11
of U.S. air strikes). For Kosovo, our data range from 20 December 1998 (three months prior to the NATO
intervention) to 24 April 1999 (one month from the start of the NATO bombing campaign).
10
Although Google Translate uses statistical methods based on bilingual text corpora (using a train-
ing set of 200 billion words from United Nations materials), rather that grammatical or rule-based algorithms, this approach is well-suited for the automated content analysis techniques employed in this paper
and discussed below – which rely on natural language processing that discards grammar, stop words (e.g.
“a”, “the”, “and”) and word order, producing an unordered array of terms (a “bag-of-words” model).
11
Formally, let R be a set of reference texts included in the training set (e.g. a pair of news articles:
one with a human interest focus and one without). Each text in r∈R is assigned a position on dimension d
(coverage type), denoted Ard. For example, Ard = 1 if article r has a human interest focus, and Ard = 0 otherwise. Let Fwr be the relative frequency of word w in text r, as a proportion of the total number of words
15
Based on instructions and examples provided in a codebook (see online appendix), our research
assistants classified 400 randomly selected texts by four sub-dimensions of coverage focus. These dimensions included (a) human interest, which places an emphasis on human needs, concerns or achievements,
(b) personality, which emphasizes the personal story, motivation or feelings of a political, military or civilian person, (c) military, which focuses on the execution of the foreign policy on the ground by armed
forces, and (d) policy, which includes any discussion of the content of a foreign policy, outside of military
operations. We treated these dimensions as non-mutually exclusive (i.e. coders could check all that apply), with human interest and personality indicative of a soft news focus, and military and policy indicative of hard news.
We used the first 50 documents to train the human coders, comparing their classifications against
each other and creating a “gold standard” of the same 50 documents classified by the authors. On the basis of this evaluation set, the coders were then given feedback on their performance and any obvious irregularities or systematic sources of error evident from the sample. We held the remaining sets of 350
documents constant across the research assistants to assess intercoder reliability. In addition to the four
categories mentioned above, we asked the coders to indicate whether each text was (a) ambiguous or otherwise presented a tough call for a given category, (b) a particularly clear, unambiguous example of a
in the text. Let Pwr = Fwr/∑r(Fwr) be the probability that we are reading text r, given the occurrence of
word w. The wordscore is defined as the expected position of a text on dimension d, given that we are
reading word w: Swd = ∑r(PwrArd). This statistic is an average of the a priori reference text scores Ard,
weighted by the probabilities Pwr. Let K be a set of texts included in the test set. The scores calculated for
the training set are used to estimate the position of any new text k∈K on dimension d: Skd = ∑w(FwkSwd),
where Fwk is the frequency of scored word w in document k and Swd is that word's score in the original
training set.
16
given category, or (c) incomprehensible, missing or written on a topic other than foreign policy. We implemented this procedure separately for articles on Libya, Iraq, Afghanistan and Kosovo.
For each variable in each conflict, we considered four measures of training set intercoder reliability: (a) percent agreement, (b) Fleiss’ Kappa (c) Kendall’s W, and (d) Krippendorff’s Alpha, with bootstrapped confidence intervals. The first measure has an intuitive interpretation as the proportion of documents in the training set, for which both coders gave the same value. The other three measures explicitly
account for chance agreement among multiple coders, and test the null hypothesis that agreements can be
regarded as random. Every test demonstrated positive and highly significant agreement between coders,
meeting or exceeding professionally acceptable levels of intercoder reliability. For instance, coders assigned identical values to between 72 and 94 percent of all documents, with Alpha statistics ranging from
.52 to .83 (or .61 to .82 with “tough calls” removed), where a value of 1 indicates perfect agreement and 0
indicates that all agreement is due to chance. We provide a more detailed overview of intercoder reliability in the online appendix.
To train the wordscores algorithm, we used the coded training texts to select a subsample of “ideal type” reference documents, which at least one of the coders considered a clear example of a given category, neither considered a tough call or incomprehensible, and which was assigned the same value (e.g.
both human interest) by the two coders. Following this step, the wordscores algorithm assigned each article in the corpus a score for the four dimensions of coverage focus. To reduce the dimensionality of our
quantity of interest, we created a composite measure of hard news from the subcategories as follows:
HARDNEWS k = Policy k + Military k - Human interest k – Personality k
(1)
where the variables on the right side of the equation are scores assigned to each document k on the four
dimensions of coverage focus. These individual scores range from 0 (e.g. no policy focus) to 1 (e.g.
17
strong policy focus), and the aggregated relative hard news score ranges from -2 (i.e. all focus is on human interest and personality) to 2 (i.e. all focus is on policy and military topics).
[Figure 3 about here]
Figure 3 shows word clouds for articles from each conflict that received high (i.e. 99th percentile,
in blue) and low (i.e. 1st percentile, in red) relative hard news scores. The scale of each word reflects the
relative frequency with which it appears in articles on each side of the spectrum. Consistent with our theoretical concept, articles that received high hard news scores focused on military operations or broader foreign policy concerns. Terms like “aircraft,” “attack” and “nato” featured prominently here, while the Iraq
articles, for example, disproportionately mentioned weapons of mass destruction, inspections, and North
Korea – potentially because that country withdrew from the Non-Proliferation Treaty in early 2003, just
as public debate about Iraq was intensifying. Articles with low hard news scores had a heavier humaninterest focus on the plight of non-combatants (i.e. “children”, “student,” “refuge”).
We aggregated the scores from the atomic level (individual articles) to panel data, where the unit
of analysis is a newspaper-day and the coverage variable (HARDNEWSit) is the corresponding average
daily score of the measure. To balance the daily panels, we added a dummy variable, PUBit, coded 1 if
newspaper i published an article on Libya on day t, and 0 otherwise.
Control variables
In addition to the variables of primary theoretical interest described above, we collected a series
of country-level controls commonly used in the literature on comparative media systems, including access
to the Internet and democracy scores. We also included a dummy variable, indicating whether an article
was published before or after the onset of U.S.-led military intervention in each of the four cases. Table 1
provides summary statistics on all variables employed in our analyses.
[Table 1 about here]
18
Empirical Analysis
We are interested in how variation in newspaper ownership structure and circulation shapes the scope and
content of news coverage of foreign policy crises. If media concentration stifles substantive political coverage, we would expect newspapers with smaller ownership networks to (a) be more likely to feature
news stories about the foreign policy crises, and (b) to cover such stories with more of a hard news focus
on military and policy issues than a soft news focus on human interest and personalities. We estimate two
sets of regression models:
PUBijkt = logit-1[ α + xijk β + xjkt γ + xkt θ + W·PUBijk,t-1 λ + uk + vj + eijkt ]
(2)
HARDNEWSijkt = α + xijk β + xjkt γ + xkt θ + W·HARDNEWSijk,t-1 λ + uk + vj + eijkt
(3)
where i indexes the newspaper, j indexes the country, t indexes the day, and k indexes the case (i.e. Libya,
Iraq, Afghanistan, Kosovo). PUBijkt is an indicator of whether newspaper i in country j publishes a story
on crisis k on day t, and HARDNEWSijkt is a normally-distributed score with lower values indicating a
relatively greater soft news coverage focus on human interest and personalities and higher scores indicating a relatively greater hard news focus on military and policy issues. The newspaper-level vector of covariates xijk includes measures of network size (or, alternatively, a dummy variable for independent ownership) and circulation (per 1,000 inhabitants). The country-level covariates xjkt include Internet access
(percent of population) and Polity 2 democracy scores (-10 = full autocracy, 10 = full democracy). Daily
covariates xkt include a post-intervention binary indicator (0 = pre-intervention, 1 = post-intervention). W
is a row-normalized connectivity matrix of the ownership network shown in Figure 2. The autoregressive
terms W·PUBijk,t-1 and W·HARDNEWSijk,t-1 represent, respectively, the proportion of co-owned newspapers that featured stories about the conflict on day t-1, and the average HARDNEWS scores of articles
printed by co-owned newspapers on day t-1.
19
We fit two versions of the models in (2-3): one with fixed and the other with random effects. In
the first instance, we account for variation in the dependent variable due to both the unobserved idiosyncrasies of each conflict (uk), and time-invariant characteristics unique to each country (vj). For example,
we may expect a greater baseline level of media attention for all countries and newspapers during the
2003 Iraq War than during the 1999 Kosovo War, or we may expect U.S. newspapers to devote greater
attention than their Australian counterparts to all four cases. If such unobserved characteristics are correlated with the error terms of our models, pooled estimation will produce biased estimates. In the random
effects specification, we account for the possibility that the unobserved heterogeneity across cases (uk), or
in countries’ individual attributes (vj) – such as editorial idiosyncrasies, niche market characteristics and
stylistic norms – is stochastic and uncorrelated with our explanatory variables.
Tables 2 and 3 report coefficient estimates for the fixed and random effect models, respectively.
Figure 4 summarizes the most theoretically relevant relationships.12 Table 4 reports a breakdown of these
results for each of the four cases.13
[Tables 2-4 and Figure 4 about here]
Scope of newspaper coverage
A cursory glance at descriptive statistics reveals initial support for our hypotheses. Newspapers in
smaller networks – with a below-median number of co-owned outlets – were 56.9 percent more likely to
12
The predictions in the left pane of Figure 4 are based on Models 1 and 2 from Table 2; the right
pane uses Models 3 and 4. Predictions are based on k=Libya and j=United States.
13
Predictions in Table 4 are based on Models 1-4, with j=United States.
20
publish a story on any of the four crises than their large-network peers.14 This difference was greatest for
coverage of the Libyan war (73.7 percent) and smallest for Afghanistan (21.2 percent), with differences in
means statistically significant for all four cases.15
Our regression results confirm that media concentration is associated with more limited coverage
of foreign crises (Hypothesis 1a). As Table 4 shows, an increase from 1st to 99th percentile in the ownership network size (1 co-owned newspapers to 148) reduces the likelihood that a U.S. newspaper will publish a report on Libya by 56 percent, all other variables held constant at their means. On any given day, an
average independent U.S. newspaper was 54.7 percent more likely to publish a story on Libya than a
similar corporate newspaper. For a newspaper in a large conglomerate, such as Newsquest in the UK (148
newspapers), the probability of running a story on Libya on an average day was just 1.2 percent (95% CI:
1.2, 1.3). The same statistic for an average independent newspaper in the same country was 2.9 percent
(95% CI: 2.8, 3.1).
The results – from both simple differences in means and regression models – further confirm that
network autocorrelation is quite strong among co-owned newspapers (Hypothesis 2a). Where an aboveaverage proportion of co-owned newspapers ran a story on any of the four crises, an average newspaper
was over three times as likely to also run a story the next day. The associated percent difference in probability of a news story ranged from 200 percent for Afghanistan to 252 percent for Iraq.
We find similar results in our regression models. An increase in the proportion of co-owned
newspapers reporting on Libya from 0 to 1 makes an average U.S. newspaper more than 4 times more
14
Formally, percent change is 100*(E[y|x1]/ E[y|x0]-1), where y is the dependent variable (i.e. PUB
or HARDNEWS), and x1 and x0 are different hypothetical values of the explanatory variables (e.g. below
vs. above-median network size, etc.).
15
p<.001 for all Kolmogorov-Smirnov test statistics.
21
likely do the same on the following day. If all outlets owned by the same parent company published a story on Libya, an average newspaper in that network would have a 31 percent chance of doing the same on
the following day (95% CI: 30, 32). Yet if no other co-owned newspaper ran a Libya story, the probability drops to 5.1 percent (95% CI: 4.9, 5.4). As Table 4 shows, these results were consistent is size and
significance for Iraq, Afghanistan and Kosovo.
Finally, higher circulation appears to make crisis coverage significantly more likely (Hypothesis
3a). A newspaper with above-median circulation was over three times as likely as a below-median circulation paper to publish a foreign policy crisis story on an average day. The percent difference – for high
vs. low circulation papers – ranged from 278 percent for Iraq to 1,028 percent for Kosovo, where the daily frequency of coverage was just 0.8 percent for below-median circulation papers, but 9.2 percent for
above-median ones.
Our regression models further show that the chances of publication were almost twice as high in a
newspaper with a circulation in the 99th percentile compared to one in the 1st percentile – rising from 5.3
percent (95% CI: 5.1, 5.7) in a U.S. newspaper with a daily circulation of 1.5 copies per 1,000 population
to 10.5 percent (95% CI: 10, 11.1) in a paper with 1 daily copy per person. The newspaper with the highest daily circulation in the world – Japan’s Yomiuri Shimbun, at 10,025,000 copies per day – had a 33.5
percent chance of running a Libya story on an average day (95% CI: 29.8, 37.4).
Relative hard news focus
In addition to explaining whether media outlets say something at all about a foreign policy crisis,
ownership structure is a significant predictor of what they say. Both descriptive statistics and the results
from Figure 4 and Table 4 are consistent with the view that concentrated media ownership is associated
with sensationalized coverage of human interest and personality-oriented themes (Hypothesis 1b). The
daily relative hard news score for newspapers with a below-median network size was significantly higher
22
than for newspapers with an above-median network size – 2 percent higher across the four cases and as
much as 3.8 percent higher in news coverage of Iraq, before controlling for other factors.
Once again, we see similar patterns in the model-based results shown in Table 4. Compared to
their independent counterparts, newspapers owned by large conglomerates offer relatively less hard news
focus on military or public policy issues. An increase in network size from 1 to 148 co-owned newspapers
reduces the expected relative hard news score by 3.6 percent for a U.S. newspaper covering Libya. The
contrast was starkest in the 2001 Afghanistan case, where a similar counterfactual produced a 30.9 percent decrease in the relative hard news score.
As predicted, coverage focus is also shared across co-owned newspapers (Hypothesis 2b). Newspapers whose co-owned peers had above-average relative hard news scores (on the previous day) received
average daily scores almost twice as large as those whose peers had below-average scores. This percent
difference ranges from 94 percent for news coverage of Afghanistan to 99.5 percent for Kosovo.
Our models confirm that the network autoregressive parameter on the relative hard news score is
positive and highly significant, indicating that a typical newspaper is more likely to cover hard news
themes if co-owned outlets did so on the previous day. For Libya coverage, an increase in the average
relative hard news score among co-owned newspapers from 0 to 0.74 (1st to 99th percentile) is associated
with a 2.2 percent increase in a U.S. newspaper’s relative hard news score on the following day. For coverage of Afghanistan in U.S. papers, the corresponding increase in the relative hard news score was 20.4
percent.
Finally, high-circulation newspapers offer considerably more hard news (relative to soft news)
coverage than their lower-circulation counterparts (Hypothesis 3b). An increase in a U.S. newspaper’s
circulation from the 1st to the 99th percentile (1.5 to 1,000 copies per 1,000 population) increases the relative hard news score by 0.5 percent for Libya coverage, and by 4.1 percent for Afghanistan coverage.
23
Country-level variation in news coverage
Beyond subnational sources of variation, foreign policy news coverage is driven in no small part
by within-country differences in political regime and access to alternative sources of information. As
countries become more democratic, newspapers in these countries become significantly more likely to
publish a story on the foreign policy crises. Yet the same within-country changes in democracy have no
discernible effect on the content of news coverage. After accounting for newspaper-level sources of variation, relative hard news scores appear not to depend on regime type.
Greater access to the Internet has a more perverse effect, increasing crisis coverage, but also making such coverage less policy-oriented. Our results suggest that competition from electronic and social
media may drive newspapers to devote less space for foreign policy in general – since such stories don’t
sell many papers – and more sensational treatment of these topics in cases where they do decide to cover
them. Oddly enough, access to more sources of information does not in and of itself make a population
better informed about foreign policy. If anything, we find that it drives news coverage down-market, toward the softer topics and themes that consumers tend to prefer.
Conclusion
Using four recent international conflicts as examples, we set out to explain two types of variation
in newspaper coverage of international conflict: daily decisions to publish a news story on a developing
crisis, and the type of coverage – hard vs. soft news – given to the story. We hypothesized that much of
this variation can be explained by differences between and within media organizations. All other things
equal, we expected a newspaper’s coverage of foreign policy and its focus on hard news to decline as the
size of its ownership network increased. We expected co-owned newspapers to feature coverage similar
in scope and focus. Finally, we expected to find greater breadth of coverage and emphasis on hard news,
relative to soft news, in higher-circulation newspapers.
24
To test these predictions, we developed a unique global content analytic dataset on newspaper
coverage of multinational conflict-related events in Libya, Iraq, Afghanistan and Kosovo. The results,
summarized graphically in Figures 4, show strong support for these hypotheses. Consistent with our theory, newspapers located within smaller networks, as well as independent papers, were significantly more
likely to report on a crisis than their larger networked counterparts. Such outlets also published stories
with greater emphasis on hard news issues of military operations and policymaking relative to softer topics emphasizing human interest or personality angles. Also consistent with the theory, we find compelling
evidence of ownership network effects: ceteris paribus, newspapers owned by the same parent company
tend to adopt more similar frames and perspectives than their separately-owned counterparts. Finally, the
extent of crisis reporting and hard (relative to soft) news coverage tends to increase with newspaper circulation.
Unless we are willing to accept the assumption that the volume, content and tone of information
communicated between leaders and citizens is unfiltered by the characteristics and incentives of its messengers, then the determinants of variation in news coverage of international conflict should be a critical
concern for scholars and practitioners alike. In recent decades, market forces, rather than regulatory regimes or national policy preferences, have increasingly determined media ownership structures. The driving force behind this movement toward media deregulation has been a belief that liberalization (that is,
deregulation) maximizes market growth and diversity, while the resulting increase in competition benefits
consumers via enhanced innovation and reduced prices. We are agnostic on these questions, which are
beyond the scope of this study. However, one seemingly clear consequence of global media deregulation
has been a trend toward global consolidation of media enterprises.
We have identified an important unanticipated consequence of such consolidation, owing to the
quite different economic incentives of corporate media conglomerates relative to their smaller-network or
independent counterparts. That is, as media conglomerates produce ever-larger networks of outlets, and
25
spread out across multiple countries, entertainment-oriented news increasingly trumps coverage of public
policy in general, and foreign policy in particular, across an ever larger proportion of global media outlets. A related casualty of media consolidation is diversity in news and editorial content, as co-owned outlets tend to share editorial sensibilities.
To the extent that democracies value a public informed about and equipped to critically assess issues of public and foreign policy, policymakers may wish to reconsider the role of regulatory policy in
shaping the mass media environment. Ownership consolidation appears to hinder the diversity of news, as
well as reducing its public and foreign policy content. For instance, one area where regulatory policy
might come into play concerns market share. We found evidence that increased circulation results in more
coverage of public policy in general, and foreign policy in particular, while independent outlets offer
more diverse news coverage. This suggests that policies aimed at enhancing the potential market shares of
individual outlets, so that they could better compete as independent entities, could have beneficial effects
on both content and diversity.
Our findings also hold important implications for bargaining theories of conflict in general, and
the voluminous literature on domestic audience costs in particular. Rationalist theories of war hold that
information failure is a primary cause of interstate conflict (Fearon 1995, Lake and Rothschild 1996),
while democratic peace theorists argue that the accountability of democratic leaders to their electorates
allows them to peacefully resolve international disputes by, for instance, generating domestic audience
costs (Fearon 1994).
Both perspectives emphasize information transparency as essential to mitigate international conflict. Yet for information to be transparent, it must first be transmitted. The primary mechanisms for doing
so, in turn, are the mass media. Recent research (Van Belle 2000, Choi and James 2006, Slantchev 2006,
Baum 2013) suggests that the media influence states’ conflict behavior in differing ways and to varying
degrees, depending on a variety of structural factors. Our findings suggest that media ownership repre-
26
sents one such structural factor that can influence the nature and extent of information available to citizens, and in doing so potentially influence foreign policy outcomes. That is, states with media dominated
by large conglomerates may have a more difficult time than their counterparts with less highly networked
media outlets in disseminating information about foreign policy to citizens. As a consequence, such states
may be less able to credibly communicate to adversaries during crises.
By utilizing recent advances in data availability, automated text analysis and network statistics,
we offered the first quantitative analysis of foreign policy news coverage on a comparable cross-national
scale and fine-grained level of disaggregation. This is a research area with exciting potential, and we foresee an explosion of applied research on comparative media systems over the next decade that may uncover new insights and challenge existing views on the role of information in international conflict. The preceding study offered an initial glimpse at how this might be done, and what some of these insights might
be.
27
Figure 1: Geographic extent of news coverage data.
(a) Libya, 2011
(b) Iraq, 2003
(c) Afghanistan, 2001
(d) Kosovo, 1999
28
Figure 2. Newspaper ownership network.
Blue diamonds denote newspapers; red dots denote parent companies; text labels denote organizations that own at least five newspapers.
29
Kosovo, 1999
Afghanistan, 2001
Iraq, 2003
Libya, 2011
Figure 3. High vs. low hard news scores.
Hard news (99th percentile) Soft news (1st percentile)
30
Figure 4. Determinants of crisis coverage.
Predicted values shown with solid lines. 95% confidence intervals shown with grey
shading or error bars.
H3: Circulation
H2: Coverage in co-owned H1b: Independent
papers
ownership
H1a: Size of ownership
network
Model 1: Probability of publishing crisis story
31
Model 2: Hard news coverage
Table 1. Summary statistics.
Dependent variables
PUB
HARDNEWS
Network lags
W·PUBi,t-1 (network lag)
W·HARDNEWSi,t-1 (network lag)
Newspaper-level
Network size
Independent newspaper
Circulation (per 1,000)
Country-level
Percent with internet access
Democracy score (POLITY2)
Daily-level
Post-intervention
Obs
Mean
Median
SD
Min
Max
1,261,742
102,426
0.08
0.48
0
0.5
0.27
0.17
0
-0.38
1
1.21
1,253,371
102,257
0.05
0.24
0
0
0.16
0.27
0
-0.16
1
1.01
1,261,742
1,261,742
1,063,318
23.04
0.29
103.89
6
0
31
38.48
0.46
351.67
1
0
0.07
148
1
10025
1,235,272
1,256,582
48.16
7.9
48.78
10
28.17
4.38
0
-10
90.72
10
1,261,742
0.51
1
0.5
0
1
32
Table 2. Coefficient estimates. Fixed effect models.
Dependent variable:
Publication
Hard news score
generalized linear mixed effects
linear mixed effects
(1)
(2)
(3)
(4)
Newspaper-level
Network size
-0.006***
(0.0002)
Independent ownership
Circulation (per 1,000)
Newspaper-day level
W·PUBi,t-1 (network lag)
-0.0001***
(0.00002)
0.001***
(0.00001)
0.468***
(0.010)
0.001***
(0.00001)
2.119***
(0.015)
2.309***
(0.016)
W·HARDNEWSi,t-1 (network lag)
Country-level
Percent with internet access
Democracy score (POLITY2)
Daily-level
Post-intervention
Constant
Case fixed effects
Country fixed effects
Case random effects
Country random effects
Observations
Log Likelihood
Akaike Inf. Crit.
0.016***
(0.001)
0.052***
(0.007)
0.016***
(0.001)
0.063***
(0.007)
0.072***
(0.008)
-2.045***
(0.087)
Y
Y
N
N
1,034,058
-269,045.8
538,315.6
0.070***
(0.008)
-2.638***
(0.088)
Y
Y
N
N
1,034,058
-268,566.1
537,356.3
33
3e-6***
(7e-7)
0.003***
(0.001)
3e-6***
(7e-7)
0.017***
(0.002)
0.017***
(0.002)
-0.0004***
(0.0001)
0.0003
(0.001)
-0.0004***
(0.0001)
0.0003
(0.001)
0.004***
0.004***
(0.001)
(0.001)
0.121***
0.118***
(0.008)
(0.009)
Y
Y
Y
Y
N
N
N
N
65,690
65,690
71,291.6
71,271.88
-142,373.2
-142,333.8
Note: *p<0.1; **p<0.05; ***p<0.01
Table 3. Coefficient estimates. Random effect models.
Dependent variable:
Publication
Hard news score
generalized linear mixed effects
linear mixed effects
(5)
(6)
(7)
(8)
Newspaper-level
Network size
-0.006***
(0.0002)
Independent ownership
Circulation (per 1,000)
Newspaper-day level
W·PUBi,t-1 (network lag)
-0.0001***
(0.00002)
0.001***
(0.00001)
0.467***
(0.010)
0.001***
(0.00001)
2.120***
(0.015)
2.310***
(0.016)
W·HARDNEWSi,t-1 (network lag)
Country-level
Percent with internet access
Democracy score (POLITY2)
Daily-level
Post-intervention
Constant
Case fixed effects
Country fixed effects
Case random effects
Country random effects
Observations
Log Likelihood
Akaike Inf. Crit.
0.016***
(0.001)
0.044***
(0.006)
0.015***
(0.001)
0.055***
(0.006)
0.072***
(0.008)
-2.884***
(0.110)
N
N
Y
Y
1,034,058
-269,398.9
538,815.9
0.070***
(0.008)
-3.238***
(0.131)
N
N
Y
Y
1,034,058
-268,922.3
537,862.7
34
3e-6***
(7e-7)
0.003***
(0.001)
3e-6***
(7e-7)
0.017***
(0.002)
0.017***
(0.002)
-0.0004***
(0.0001)
0.0004
(0.0005)
-0.0004***
(0.0001)
0.0004
(0.0005)
0.004***
0.004***
(0.001)
(0.001)
0.380***
0.378***
(0.106)
(0.106)
N
N
N
N
Y
Y
Y
Y
65,690
65,690
70,999.35
70,983.47
-141,978.7
-141,946.9
Note: *p<0.1; **p<0.05; ***p<0.01
Table 4. Determinants of crisis coverage, case by case.
Values are percent changes in (a) probability of publication or (b) hard news scores, associated with counterfactual scenario, all other variables held constant at means. 95% confidence intervals in parentheses.
Variable
H1a: Network size
(DV: publication)
H1b: Network size
(DV: hard news)
H1a: Independent
(DV: publication)
H1b: Independent
(DV: hard news)
H2a: Co-owned papers
(DV: publication)
H2b: Co-owned papers
(DV: hard news)
H3a: Circulation
(DV: publication)
H3a: Circulation
(DV: hard news)
Counterfactual
1⇒148 (1st⇒99th
percentile)
1⇒148 (1st⇒99th
percentile)
0⇒1 (corporate
⇒ independent)
0⇒1 (corporate
⇒ independent)
0⇒1 (1st⇒99th
percentile)
0⇒.74 (1st⇒99th
percentile)
1.5⇒1000 (1st⇒ 99th percentile)
1.5⇒1000 (1st⇒
99th percentile)
Libya
-56.8
(-58.9, -54.6)
-3.6
(-4.6, -2.6)
54.7
(51.9, 57.6)
0.5
(0.2, 0.8)
505.2
(490, 521)
2.2
(1.8, 2.6)
95.3
(91.9, 98.8)
-56.8
(-58.9, -54.6)
Iraq
-56
(-58.1, -53.7)
-4.9
(-6.3, -3.5)
52.4
(49.8, 55.1)
0.7
(0.2, 1.1)
436.2
(425, 447)
3
(2.5, 3.6)
90.5
(87.2, 93.7)
-56
(-58.1, -53.7)
35
Afghanistan
-55.8
(-57.9, -53.5)
-30.6
(-39.2, -21.8)
53
(50.2, 55.7)
4.3
(1.3, 7.4)
422.3
(411, 434)
20.2
(16.4, 24)
89.4
(86.3, 92.5)
-55.8
(-57.9, -53.5)
Kosovo
-56.7
(-58.8, -54.5)
-6.3
(-8.1, -4.5)
54.3
(51.5, 57)
0.9
(0.3, 1.5)
492.3
(479, 506)
3.9
(3.2, 4.6)
94.5
(91.1, 97.8)
-56.7
(-58.8, -54.5)
References
Adsera, Alicia, Carles Boix, and Mark Paine. 2003. Are you being served? Political accountability and
quality of government. Journal of Law, Economics, & Organization 19(2): 445-490.
Bagdikian, Ben H. 2000, The Media Monopoly, 6th Ed. Boston: Beacon Press.
Bagdikian, Ben H. 1985. The U.S. Media: Supermarket or Assembly Line? Journal of Communication
35: 97-109.
Baum, Matthew and Tim Groeling. 2009. War Stories: The Causes and Consequences of Public Views of
War Princeton, NJ: Princeton University Press.
Baum, Matthew A. 2013. The Iraq Coalition of the Willing and (Politically) Able: Party Systems, The
Press, and Public Opinion on Foreign Policy. American Journal of Political Science 57(April): 442458.
Bennett, W. Lance. 1990. Toward a theory of press-state relations in the United States. Journal of Communication 40 (2): 103-25.
Besley, Timothy, and Robin Burgess. 2002. The Political Economy of Government Responsiveness: Theory and Evidence from India. Quarterly Journal of Economics 117 (4): 1415-52.
Bielby, William T. and Denise D. Bielby. 2003. Controlling Prime-Time: Organizational Concentration
and Network Television Programming Strategies. Journal of Broadcasting and Electronic Media
47(4): 573-96.
Chang, T.K. and J.W. Lee. 1992. Factors Affecting Gatekeepers' Selection of Foreign News: A National
Survey of Newspaper Editors. Journalism Quarterly 69: 554-61.
Choi, Seung-Whan and Patrick James. 2007. Media Openness, Democracy and Militarized Interstate Disputes. British Journal of Political Science 37 (1): 23-46.
36
Davis, Richard, and Diana Owen. 1998. New Media and American Politics. New York: Oxford University Press.
DellaVigna, Stefano & Ethan Kaplan, 2007. The Fox News Effect: Media Bias and Voting. The Quarterly
Journal of Economics 122 (3): 1187-1234.
DiMaggio, Anthony R. 2009. Mass Media, Mass Propaganda: Examining American News in the ‘War on
Terror’ Lanham, MD: Lexington Books.
Djankov, Simeon, Caralee McLiesh, Tatiana Nenova, and Andrei Shleifer. 2001. Who Owns the Media?
Journal of Law and Economics 46 (October): 341-82.
Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper and Row.
Fearon, James D. 1994. Domestic Political Audiences and the Escalation of International Disputes. American Political Science Review 88 (3): 577-592.
Fearon, James D. 1995. Rationalist Explanations for War. International Organization 49 (3): 379-414.
Flitton, Daniel. 2011. Sky TV Trumps ABC in China. WAtoday.com (August 17).
Gentzkow, Matthew. 2006. Media Bias and Reputation. Journal of Political Economy 114 (2): 280-316.
Gentzkow, Matthew and Jesse M. Shapiro. 2004. Media, Education and Anti-americanism in the Muslim
World. Journal of Economic Perspectives 18 (3): 117-133.
Gentzkow, Matthew and Jesse M. Shapiro. 2007. What Drives Media Slant? Evidence from U.S. Newspapers. University of Chicago mimeo.
Gerber, Alan S., Dean Karlan and Daniel Bergan. 2006. Does The Media Matter? A Field Experiment
Measuring The Effect of Newspapers on Voting Behavior and Political Opinions. mimeo.
Gilens, Martin and Craig Hertzman. 2000. Corporate Ownership and News Bias: Newspaper Coverage of
the 1996 Telecommunications Act. Journal of Politics 62(May): 369-386.
Grimmer, Justin and Brandon Stewart. 2013. Text as Data: The Promise and Pitfalls of Automatic Content Analysis Methods for Political Documents. Political Analysis 21(3): 267-297.
37
Hallin, Daniel C. and Paolo Mancini. 2004. Comparing Media Systems. New York: Cambridge University Press.
Herman, Edward. 2002. The Media and Markets in the United States. In The Right to Tell: The Role of
Mass Media in Economic Development, Chapter 4:61-82. Washington, D.C.: The World Bank.
Herman, E. S. & Chomsky, N. 2002, Manufacturing Consent: The Political Economy of the Mass Media,
New York: Pantheon.
Iyengar, Shanto. 1991. Is Anyone Responsible? How Television Frames Political Issues. Chicago, IL:
University of Chicago Press.
Iyengar, Shanto, James Curran, Anker Brink Lund, Inka Salovaara-Moring, Kyu S. Hahn and Sharon Coen. 2010. Cross-National versus Individual-Level Differences in Political Information: A Media Systems Perspective. Journal of Elections, Public Opinion and Parties 20 (3): 291–309.
Jenkins, H. 2004. The Cultural Logic of Media Convergence. International Journal of Cultural Studies 7:
33-43
Keefer, Philip. 2007. Clientelism, Credibility, and the Policy Choices of Young Democracies. American
Journal of Political Science 51(4): 804-821.
Kellner, D. 2005. Media Spectacle and the Crisis of Democracy: Terrorism, War and Election Battles.
Boulder, CO: Paradigm Publishers.
Kalyvas, Stathis and Matthew Adam Kocher. 2009. The Dynamics of Violence in Vietnam: An Analysis
of the Hamlet Evaluation System (HES). Journal of Peace Research 46 (3): 335-55.
La Porta, Rafael, Florencio Lopez-de-Silanes, and Andrei Shleifer. 1999. Corporate ownership around the
world. Journal of Finance 54: 471-518.
Lake, David and Donald Rothchild. 1996. Containing Fear: The Origins and Management of Ethnic Conflict. International Security 21(2): 41-75.
38
Laver, Michael, Kenneth Benoit and John Garry. 2003. Extracting Policy Positions from Political Texts
Using Words as Data. American Political Science Review 97 (2): 311-331.
Lowe, Will. 2008. Understanding Wordscores. Political Analysis 16 (4): 356-371.
Markham, James W. 1961. Foreign News in the United States and South American Press. Public Opinion
Quarterly 25 (4): 248-262.
McChesney, Robert W. 2000. Rich Media, Poor Democracy: Communication Politics in Dubious Times.
New York: The New Press.
Patterson, Thomas E. 2000. Doing Well and Doing Good. John F. Kennedy School of Government, Harvard University, Faculty Research Working Paper Series, #RWP01-001.
Peterson, R.A., and D.G. Berger. 1975. Cycles of Symbol Production: The Case of Popular Music. American Sociological Review 40: 153-73.
Peterson, R.A., and D.G. Berger. 1996. Measuring Industry Concentration, Diversity, and Innovation in
Popular Music. American Sociological Review 61: 175-181.
Rosenstiel, Tom, Amy Mitchell, Kristen Purcell and lee Rainie. 2011. How People Learn About Their
Local
Community.
Pew
Research
Journalism
Project.
Available
at:
http://www.journalism.org/2011/09/26/local-news (accessed 9/26/14).
Schultz, Kenneth A. 2001. Looking for Audience Costs. Journal of Conflict Resolution 45 (1): 32-60.
Shinar, D. 2003. The peace process in cultural conflict: the role of the media. Conflict and Communication Online, 2: 1-10
Slantchev, Branislav L. 2006. Politicians, the Media, and Domestic Audience Costs. International Studies
Quarterly 50 (2): 445-477.
Smith, Alastair. 1998. International Crises and Domestic Politics. American Political. Science Review 92
(3): 623-638.
39
Snider, James H. And Benjamin I. Page. 2003a. Does Media Ownership Affect Media Stands? The Case
of the Telecommunications Act of 1996. Northwestern University, Institute for Policy Research, IPR
Working Paper No. 97-6.
Snider, James H. And Benjamin I. Page. 2003b. The Political Power of TV Broadcasters: Covert Bias and
Anticipated Reactions. Northwestern University, Institute for Policy Research, IPR Working Paper No.
97-29.
Sobel Richard. 2001. The Impact of Public Opinion on U.S. Foreign Policy Since Vietnam. New York:
Oxford University Press.
Sobel, Richard, Peter Furia and Bethany Barratt, eds. 2012. Public Opinion and International Intervention: Lessons from the Iraq War. Washington, D.C.: Potomac Books
Soroka, Stewart N. 2003. Media, Public Opinion and Foreign Policy. International Journal of
Press/Politics 8: 27-48.
Stromberg, David. 2004. Radio’s Impact on Public Spending. Quarterly Journal of Economics 119(1):
189-221.
Van Belle, Douglas. 2000. Press freedom and global politics. Westport, CT: Praeger.
Zaller, John R. 1999. A Theory of Media Politics: How the Interests of Politicians, Journalists, and Citizens Shape the News. Chicago: University of Chicago Press.
40