State-controlled Media and Foreign Policy:

Analyzing Russian-language News

Mark David Nieman∗ Elena Labzina†

Abstract
Authoritarian regimes frequently employ state-owned media to frame and explain
their domestic and foreign policies, often referencing national-markers and invoking
social identities to justify their actions. We focus on how the Russian government
used state-owned media to build public support prior to its invasions of Georgia and
Ukraine. Specifically, we analyze the vocabulary structure of news broadcasts for
Russia-24, a state-owned news channel, and compare it to that of Dozhd, an inde-
pendent news source. We adopt a placebo approach to separate event-driven coverage
from state-directed propaganda and identify changes in the nature of news coverage
using a change-point model. We find that Russian state-owned media significantly in-
creased its coverage of Georgia and Ukraine in the months preceding Russia’s military
interventions, beyond what would be expected by purely event-driven coverage. More-
over, this increase in coverage is associated with an increase in discussion of traditional
Russian geopolitical rivals, such as the US.

∗
Assistant Professor. Department of Political Science. University of Toronto. Corresponding author:
mark.nieman@utoronto.ca.
†
Commercial Software Engineering. Applied Machine Learning Team, Microsoft len4ick@gmail.com.
Introduction

State-owned media outlets provide authoritarian regimes with an invaluable resource to

frame events and policy decisions.1 Recent research has evaluated authoritarian regimes’
efforts to limit dissent (King, Pan and Roberts 2013, 2017), influence public opinion during
political crises (Windsor et al. 2017; Munger et al. 2018), or manipulate negative economic
news (Rozenas and Stukal 2018). We build on this research to evaluate the state-owned
media outlets with the goal of framing military interventions.
Authoritarian regimes often employ state-owned media to explain and justify their do-
mestic and foreign policies. These policy frames are most effective, however, when they
are consistent with existing conceptions of the state and its pre-established social identi-
ties (Miskimmon, O’Loughlin and Roselle 2014, 2017). State-sponsored media accounts of
current events, in particular, are generally constructed to remain consistent with these pre-
existing narratives. The narrative of actions for both the government and foreign states can
be placed set of familiar tropes, which can often be condensed into characterizations such as
whether either state is a superpower, rising power, regional power, ally, rival, subordinate or
some combination (Wendt 1994; Thies and Nieman 2017).
These tropes help contextualize, justify, and make sense of either the regime’s, or a rival
state’s, foreign policy actions. Lo (2004), for example, contends that Russian leadership
views itself as a barrier against a barbarian West, while Lavrov (2006) and MacFarlane
(2006) argue that Russia considers itself a great power central to the international order
with a hegemony in the post-Soviet space. From this vantage point, Russian military action
in Eastern Europe fulfills the conflict management obligations of a regional or great power.
These characterizations and attributed identities provide a framework that both clarifies and
1
We refer to “state-owned media” as those state-owned outlets that engage in propaganda on behalf of
the governing regime. This definition excludes publicly owned media that adheres to standard journalist
practices, such as the BBC or NPR.

1
constrains the ways in which a government can effectively cast friendly or unfriendly foreign
governments to their population.
The goal of the paper is to investigate the use of state-media to frame militarized interven-
tions. Advances in textual analysis have created new research opportunities for systematic
analysis of mass media coverage, including manipulation of coverage for the purposes of fur-
thering specific goals of the political elite. McManus (2014, 2016) and Kydd and McManus
(2015), for example, examine statements by US presidents to study how leaders signal re-
solve. Textual analysis has also been widely used to investigate how closed regimes convey
(and sometimes manipulate) information for both domestic and foreign audiences. Lucas
et al. (2016) and Windsor (2018) use textual analysis to study the process of Islamic rad-
icalzation, while Windsor, Dowell and Graesser (2014) and Dowell, Windsor and Graesser
(2015) investigate linguistic patterns of persuasion utilized by authoritarian leaders during
crises.
We use these tools to analyze the Russian government’s use of state-owned media to frame
the public’s perception of issues and events. We argue that, in contrast to independent media
outlets whose coverage is determined by occurrence/salience of events, state-owned media
outlets may be used to shape public response/perception as, or even before, a state imple-
ments a specific policy. We develop a web-scraping application that allows us to extract
and analyze vocabulary structures of both state-owned and independent Russian-language
media. The goal is to explore how the Russian government projects both its self-perceptions
and its perception of “the West” onto its citizens. By focusing on Russian-language sources,
rather than relying on secondary sources or English translations, we directly access govern-
ment efforts to frame issues for their domestic audience. We find evidence that Russian
state-owned media does initiate coverage of targets in the months preceding military inter-
ventions. The results have immense practical value as a technique to analyze the signaling
behavior of autocratic regimes.

2
 In what follows, we analyze the vocabulary structure of Russia-24, a major state-owned
news channel, and Dozhd (TVRain), an independent news channel, in order to separate
event-driven coverage from state-directed propaganda. We use a change-point model to
identify shifts in the nature of Russian media coverage. We find that Russian state-owned
media significantly increased its coverage of Georgia and Ukraine in the months preceding
Russia’s military interventions. This increase in coverage was often supplemented and asso-
ciated with an increased focus on traditional Russian political rivals, such as the US.
In the next section, we discuss how state-owned media influences political outcomes.
We develop a foreign policy theory of state behavior and media framing, and apply this
account to the case of Russia. The following section discusses our data collection technique
and displays preliminary evidence. Finally, we describe and present the results of our data
analysis, and conclude by summarizing the results, implications, and next steps.

State-owned Media

State-owned media alter the manner in which individuals can process information and affect
political behavior. Countries where the state significantly controls the media substantially
increase the costs to citizens to gather independent information. In the absence of counter-
balancing information, large segments of the population can be influenced by a centralized
and orchestrated account of policy or events.2
Autocratic regimes frequently use media to cement their own political power, even as
the media environment has become more commercial and less centralized since the 1990s.
Autocratic regimes are incentivized to restrict media freedom to promote news favorable to
the regime and restrict news dis-favorable to it. Autocratic regimes generally have several
2
Individuals with low political sophistication may change their attitudes when presented with a counter-
attitudinal message, whereas politically sophisticated individuals mediate messages that counter their pre-
existing beliefs (Zaller 1992; Redlawsk 2002; Taber and Lodge 2006).

3
strategies to do this. First, in order to reduce public dissatisfaction, autocratic regimes can
act prevent the public from knowing of policy failures. A downside to this strategy, however,
is that it risks associated with undermining the state capacity to respond to environmental
and economic crises by restricting credible information flows of bureaucratic performance to
the central regime (Egorov, Guriev and Sonin 2009).
Second, autocrats can overwhelm the public with counter-messaging. For example, Chi-
nese government officials frequently respond to online criticism by raising other issues, rather
than responding to specific criticisms (King, Pan and Roberts 2017). Likewise, Maduro used
emotionally charged populist rhetoric to distract from poor economic news (Love and Wind-
sor 2018). Similarly, pro-Maduro legislators used Twitter to pose competing narratives in
order to obfuscate criticisms posed by opposition officials during the anti-Maduro protests
(Munger et al. 2018). A downside of this that it can make rallying support for policies dif-
ficult, as some segments of the public may become disenchanted and largely ignore politics
altogether.
Third, regimes exert some degree of control over the content of news broadcasts. This ap-
proach is an effort to partially gain the benefits of information and public interest, while still
minimizing the risk of collective action against the regime (Egorov, Guriev and Sonin 2009;
Gunitsky 2015; King, Pan and Roberts 2013). Rather than directly managing content via
government-scripted stories and pre-publication censorship, autocratic regimes have increas-
ingly relied on self-censored media to support regime goals (Hassid 2008; Rozenas and Stukal
2018; Schimpfossl and Yablokov 2014; Stockmann and Gallagher 2011)3 , as well as providing
a higher-quality product on state-owned and state-affiliated media to maintain viewership
(Moehler and Singh 2011; Stockmann and Gallagher 2011).4 Media self-censorship relates
to the choice, distortion, and emphasis of stories covered (or not covered) (Lee 1998).
3
The threat of coercion, of course, is necessary in order to enforce self-censorship.
4
Oates (2007) argues that state subsidies offer government’s leverage over media coverage and content.

4
 Self-censorship arises from one, or a combination, of mechanisms. First, self-censorship
can result from uncertainty regarding what content may result in punishment of journalists
and managers. In China, for instance, reporters temper how aggressively to pursue stories
because they are unsure how far they can push a story without facing arbitrary and harsh
punishment (Hassid 2008, 422). Second, the government may make clear that certain topics
that are taboo to cover at all. Coverage critical of Vladimir Putin, for instance, is not allowed
on state-aligned networks (Schimpfossl and Yablokov 2014).5 Third, self-censorship can
result from journalists responding to financial reward structures that avoid raising content
that the government considers problematic. Finally, media may report negative news, but
take steps to limit negative fallout or blame of the regime. In Russia, for example, negative
news stories are often attributed (either directly or indirectly) to foreign governments or
global processes (Rozenas and Stukal 2018).
The effect of each of these mechanisms is to produce media coverage that toes the party
line. At the same time, state-owned media are able to take advantage of relatively high
programming quality, and subsequent viewer interest, resulting from reporter creativity.6
Even non-political shows, including ostensibly entertainment programs, such as legal shows,
can be used to push the government’s agenda (Stockmann and Gallagher 2011, 453,456).7
Independent, private-owned media, conversely, are often disadvantaged in financing and may
have lower programming quality, on average, than state-owned media outlets (Coyne and
Leeson 2009; Vartanova 2012).
That state-media reflect their government’s positions and policies is important because
5
These strategies can serve as substitutes. China has consistently been one of the world’s leaders in
jailing journalist, but seldom executes journalist. Russia, on the other hand, has consistently seen the most
journalists killed (Committee to Protect Journalists 2017).
6
The existence of multiple state-aligned channels in Russia, for instance, ensures that they must compete,
to some degree, for viewers. Thus, to keep viewer interest, they cannot aim solely at brainwashing, as viewers
will become bored and change the channel (Schimpfossl and Yablokov 2014, 308).
7
Huang (2015) argues that, in addition to direct attempts to influence public opinion, the constant
presence of government signals strength in effort to maintain political order.

5
the public often has equal or higher levels of trust in state-owned than independent media
outlets (Moehler and Singh 2011). One implication is that viewing independent media does
not always affect public opinion when it contrasts with state-owned media (Gehlbach and
Sonin 2014; Shi, Lu and Aldrich 2011). Relatively high trust, combined with a lack of well-
resourced competition, mean that state-owned media can effectively relay the government’s
policy positions to the public.

Foreign Policy and State-owned Media

State-owned media can be used to frame and justify a government’s policy position. Con-
sistent with China’s increasingly cooperative foreign policy towards the US, the state-owned
People’s Daily was more positive in coverage towards the US than independent media (Stock-
mann 2011). State-owned media can also be used by the government to solidify or coordinate
policy with domestic elites. Gunitsky (2015) argues that mass media, and more recently so-
cial media, is used to coordinate the interests of central and local elites. He notes that
Russia, in particular, has used mass media to central power and limit the authority of local
politicians.
State-owned media often manipulate news stories by attributing their causal effect to
direct or associational actors (Rozenas and Stukal 2018). This is made easier in the case of
international news and foreign policy, as most viewers have less information/interest on such
topics relative to news related to domestic/local issues. Thus, they are more susceptible to
framing/influence on topics related to international and foreign policies.
The ability of a regime to project policies onto its population, however, is not absolute.
Even with state-owned media casting government policies in a favorable light—or attributing
negative outcomes to external forces—a government faces some restrictions when presenting
information to the public. Both internal and external foes must be easily recognizable if
state-owned media is going to be able to frame opposition to a government policy as non-

6
patriotic, or place blame on negative policy outcomes. Foes must be familiar tropes and
scapegoats, widely recognized as ‘enemies of the state’ whose invocation is immediately
associated with opposition to the national interest and the nation’s values. States, in other
words, are constrained and framed by the messages and identities that a they have previously
invoked.
To account for the constraints imposed by national identities and value systems, we
adopt a foreign policy approach to analyzing state behavior towards its neighbors. Foreign
policy approaches consider both international and domestic features that serve to struc-
ture and constrain the choices available to a government regime (Holsti 1970; Thies and
Breuning 2012). Through interactions with other states, state leaders adopt perceptions of
themselves and assign identities to other states (Walker 1992; Wendt 1994; Breuning 2011).
The specific roles/identities that a state assigns to itself—i.e. its national role conceptions
(NRCs)—reflect domestic elite preferences and their expectations of appropriate state behav-
ior.8 Adopted and assigned identities/roles provide social constraints on a state’s behavior
through limiting the strategies a leader can even conceive of undertaking (Thies 2013; Nie-
man 2016b; Thies and Nieman 2017). These identities can be used by political actors, such
as autocratic elites, to create narratives of the past in a strategic manner in order to pro-
mote their policies, downplay shortcomings, and manage expectations of future behavior
(Miskimmon, O’Loughlin and Roselle 2014, 2017).
Governments can use state-owned media to describe themselves, as well as those states
identified as foreign friends and foes, as part of an effort to generate a well curated and
managed message, or strategic narrative, for the public (Roselle, Miskimmon and O’Loughlin
2014). Governments use their NRCs as part of their strategic narrative, in effort to reify
their own influence and to manage how foreign events are perceived domestically. Thus,
8
The roles that a state assigns to itself may face some domestic contestation from domestic rivals (Cantir
and Kaarbo 2012; Brummer and Thies 2015).

7
states have an incentive to maintain NRCs and frequently use them as a point of reference
for making sense of and understanding unexpected events. By operating within a foreign
policy theoretical framework, we are able to understand how authoritarian regimes can raise
the salience and context of specific issues or interactions with neighboring states within the
broader understanding of a state’s position within the international system. For instance,
following the Maidan protests that resulted in the ouster of Ukraine’s President Yanukovych,
the Russian government described their intervention in Ukraine and support of militant forces
in Donetsk and Luhansk within the broader context of a geopolitical struggle with the US,
rather than simply as an attempt to maintain political control of a traditional ally.
Efforts to unite and control domestic (and sometimes foreign) populations through nor-
mative influence via mass media is a critical component of state sovereignty, and is a cheaper
alternative to state-building and political control than coercion (Warren 2014, 2015). While
independent media tend to be event- rather than policy-driven, i.e. their coverage is driven
by event-occurrence and salience, state-owned media outlets may be used to shape pub-
lic response/perception, even before a state implements a policy (Roselle, Miskimmon and
O’Loughlin 2014; Miskimmon, O’Loughlin and Roselle 2017). While all governments have
an incentive to shape the narrative, especially prior to initiating a conflict, state-controlled
media can be more easily directed to this function than independent media. As such, state-
controlled media can be used to re-assert specific identities in order to support the narrative
that a government wants to emphasize. For instance, governments can justify intervention
in neighboring state by invoking the role of great power, protector of a region or co-ethnics,
or a regional or global balancer.9 Mass communication can also be used to co-opt potential
partisans in neighboring states.
9
While the great power role is well known within the international relations literature (e.g., Chiba, Mar-
tinez Machain and Reed 2014), the regional and co-ethnic protector (Frazier and Stewart-Ingersoll 2010;
Davis and Moore 1997) and regional and global balancer (Levy and Thompson 2010) roles have also been
invoked to explain conflict behavior.

8
 As an example, following the 2013–2014 Maidan protests and subsequent departure of
the pro-Russian President Yanukovych from Ukraine, Russian media repeatedly referred to
a region in Ukraine—stretching from Donbass in the East, to Crimea and along the coast
adjacent to the Black Sea in the South, to Transnistria in Moldova—as Novorossiya. In
the aftermath of the Maidan uprising, Russia claimed that Novorossiya was historically a
Russian territory—citing the conquests of Catherine the Great—and referred to the new
government in Kiev as a fascist junta, attempting to invoke geographical and political divi-
sions from World War II. Such references could potentially be used to provide justification
for intervening in a neighbor in effort to protect co-ethnics, as well as attempt to assert
a greater-Russian identify and generate support from those living in southern and eastern
Ukraine who may view Russian news broadcasts advocating this perspective.10

The Russian Case

In this paper, we focus on the Russian government’s use of media in an effort to preemptively
build support for their international conflicts. Contemporary Russia constitutes a suitable
case for our analysis, as the government directly or indirectly controls much of the media
landscape.11 A privately owned independent channel, Dozhd, is a notable exception.
We argue that governments are constrained in the manner in which they can portray both
themselves and others depending on their pre-existing NRCs. Governments can manufacture
support in the buildup to potential international conflicts by framing the coverage of the
target (Roselle, Miskimmon and O’Loughlin 2014). Thus, we expect to see an increase in
the number of new stories about a target prior to military intervention, as the government
10
Russian media is popular in many post-Soviet states, as it frequently provides higher quality news and
entertainment programming than domestic offerings (Trenin 2009b; Vartanova 2012).
11
Despite the increase in television channels, news coverage is conducted primarily by three national
networks, which are directly controlled by the government (Gehlbach and Sonin 2014). Russia’s high degree
of media control is consistent with theories of information communication and resource allocation (e.g.,
Egorov, Guriev and Sonin 2009).

9
attempts to generate support (or, at least, minimize dissent) for the conflict. In order to rally
support around the government’s cause, state-controlled media frame the dispute within a
narrative favorable to the government and consistent with its pre-existing NRCs. In the
case of regional or rising powers, such as Russia, constructing support for potential conflict
abroad can be relatively easy, if prior NRCs have emphasized an explicit sphere of influence
or a security complex. Governments can simultaneously build nationalist sentiment and
direct this energy against target states (Miskimmon, O’Loughlin and Roselle 2014, 2017).
Governments can direct state-owned media to cover issues and run stories on target states
to raise the salience of these states and magnify grievances with them.
To remain consistent with a state’s NRCs, governments can invoke similarities between
the target of their aggression and other enemies of the state, such as other traditional rival
powers. This can be done by stating that a target is in danger of potentially moving under
the influence of a traditional rival. Doing so can help reduce domestic aversion to the use
of force against co-ethnics. As we demonstrate, state-owned Russian media does precisely
this in the months prior to intervening in Georgia and Ukraine by increasingly referencing
traditional rivals, such as the US.
Russian NRCs, under President Putin, have emphasized the roles of a regional power,
an independent great power, and an aspirant global power (Allison 2008; Lo 2002; Trenin
2009a).12 The creation of the Commonwealth of Independent States (CIS) and Eurasion
Economic Union (EEU) have even been described as evidence of neoimperial or hegemonic
role in the post-Soviet space on the part of Russia (Rahr 2007; Frazier and Stewart-Ingersoll
2010). In addition, since at least 2006, the Russian government has once again viewed the
US as its primary geopolitical rival (Rahr 2007).
Given these NRCs, one would expect Russia to have an active and interventionist foreign
12
MacFarlane (2006) suggests that Russia, rather than being an emerging power, seeks to reverse its decline
and return to great power status.

10
policy towards neighboring states (Thies and Nieman 2017). Russia is directly or indirectly
involved in a number of outstanding territorial disputes: the status of Abkhazia and South
Ossetia with Georgia, the annexation of Crimea with Ukraine, the status of Transnistria
with Moldova, a border dispute with Estonia, and possession of the Southern Kurils with
Japan. Russia also has played an active role supporting government regimes or aiding ethnic
minorities that are aligned with it politically, using a variety of hard and soft power tools
(McManus and Nieman 2018; Truckos 2021). Russia has been invited to send troops to help
defend a neighbor’s territorial integrity in the cases of Tajikistan and Uzbekistan (Gibler
and Sewell 2006). It has also been involved in managing the conflict in Nagorno-Karabakh
region between Armenia and Azerbaijan and, more recently, in Syria. Russia has provided
material support to governments within its orbit, often helping to prop up a regime when
facing domestic opposition (Chyzh and Labzina 2018), most recently in Kazakhstan in 2022.
Lastly, Russia has also expressed concern about the welfare of Russian-speaking populations
in Estonia, Latvia, Lithuania, Moldova, and Ukraine.
To justify such a foreign policy, the Russian government has created a narrative in which
they frame many of these disputes and actions as part of a larger conflict between itself and
the US. NATO expansion, for example, has been argued to serve a destabilizing influence on
Russia, especially as Russia is increasingly dissatisfied with its global status and faces eco-
nomic difficulties (Gibler 1999). It is therefore unsurprising that Russia viewed and framed
the conflict with Georgia within the broader context of US efforts to undermine Russia’s
great power status by its involvement in the colored revolutions and NATO expansion (Tsy-
gankov and Tarver-Wahlquist 2009). Similarly, Russian officials accused the US of aiding
and orchestrating the Maidan protests, and used this claim to justify potential intervention
in the crisis (Higgins and Baker 2014).
To investigate Russia’s use of state media in support of their active foreign policy, we
examine a time series of Russian media coverage of Ukraine (and, more specifically, Crimea)

11
 Figure 1: Change in Georgian and Ukrainian Affinity towards Russia and the US

Note: UN voting similarity data from Bailey, Strezhnev and Voeten (2017).

and Georgia. The conventional framing of both Georgia and Ukraine as drifting away from
Moscow and towards Washington arose, in part, from real policy shifts by the two former
Soviet states. As evidenced in Figure 1, which displays the annual UN voting similarity
scores of Georgia and Ukraine with Russia and the US, respectively, between 2000 and 2015,
it is clear that the foreign policy preferences of Tbilisi and Kyiv have slowly been moving
away from those of Moscow.13 While both Georgia and Ukraine maintained ideological scores
closer to Russia than the US throughout the time period, the trend in each has been towards
convergence and eventual transition to the US.14
13
Data on UN voting similarity are obtained from Bailey, Strezhnev and Voeten (2017).
14
Changes arising from domestic policy does not, however, necessarily correlate positively with changes in
the degree of support signaled by either Russia or the US. In fact, as we show in Figure 7 in the Appendix,
Russian support for both Georgia and Ukraine remained mostly constant between 2000–2012, while US

12
 Efforts to keep both Georgia and Ukraine in Russia’s political orbit are often referenced
as justifications for the intervention. In the case of Georgia, Russia intervened in Georgia in
the summer of the same year that Tbilisi made a dramatic shift towards the US. Conversely,
Ukraine had been steadily moving towards the US since the 2004 Orange Revolution, even
during the presidency of the Russia-supported Yanikovich.
In each case, the change in foreign policy affinity towards Washington and away from
Moscow likely affected both the quantity and content of coverage provided by Russian state-
owned media more than that of independent outlets. State-owned media often discussed
Ukraine and Georgia as satellite states, squarely within Russia’s sphere-of-influence. The
warming of Ukraine and Georgia relations with the US, a traditional foe of Russia15 , was
viewed skeptically by Russia, with ‘US aggression’ and Western expansion into Russia’s
neighbors previously being invoked as a cause of Russia’s more active foreign policy within
the region (Gibler and Sewell 2006; Rahr 2007; Aksenyonok 2008; Nau and Welt 2013).
We expect that state-owned media news stories discussing the US were increasingly linked
and negatively associated (directly and indirectly) with the changing foreign policy orienta-
tions of Russia’s neighbors. We expect this negative associations to be especially pronounced
immediately prior to Russian military interventions into Georgia and Ukraine, as the gov-
ernment sought to build support and justify its policy actions. Conversely, independent
media were less likely to frame the Russian government actions in a positive light. Rather,
independent media is more likely to focus on observed events to drive its news coverage, as
well as to provide a wider range of political analysis regarding the government actions.
Our focus on Russian coverage on Ukraine and Georgia is appropriate because we are
interested in changes in Russian state-owned media coverage. We expect coverage of each
state to shift, from relatively low salience and mostly positive descriptions as Russian sub-
support actually decreased. Measures of major power support are obtained from McManus and Nieman
(2018).
15
Thies and Nieman (2017, 86-88) list “US rival” as one of Russia’s core foreign policy NRCs.

13
ordinates, to relatively high salience mostly and negative descriptions as adversaries of the
Russian state. Both the Ukrainian and Georgian cases experience interventions in the time
series of Russian news coverage in the form of Russian military involvement. Russian in-
terference in Ukraine began with the appearance of “little green men” on 27 February 2014.
After initial denials, Russian President Putin admitted that the forces were, in fact, Russian
troops (Walker 2015a). This military intervention culminated in the formal annexation of
Crimea by Russia on 18 March 2014. In addition, Russian “military tourists” entered Eastern
Ukraine in March 2014 (Walker 2015a). In the case of Georgia, Russia sent military forces
into South Ossetia and Abkhazia, two breakaway regions supported by Russia, on 8 August
2008 and 9 August 2008, respectively. From a methodological perspective, these interven-
tions indicate that there are periods of stability in the relations of both the Ukrainian and
Georgian governments and Russia, punctuated by significant animosity.
The variation in coverage related to these military interventions allow us to explore
how Russian state-owned and independent media differ in several ways. First, we expect
to observe increasing new stories involving both the target of Russian aggression, as well
as its traditional rivals, by state-owned media prior to the onset of military intervention.
State-owned media can be more easily utilized to push a narrative that is favorable to the
government in the build-up to a militarized intervention. In contrast, independent media
is more events-driven and fluctuation in coverage should coincide more directly with events
on the ground. In particular, increases in state-owned media coverage should precede the
Russian military interventions in Crimea on 27 February 2014 and Eastern Ukraine in March
2014. We also expect increases in media coverage prior to the Russian intervention into
Georgia on 8 August 2008. This account leads us to the following hypothesis:
Hypothesis 1: Ukraine and Georgia feature more frequently in Russian state-owned news
stories prior to military intervention.
Second, to enhance the effectiveness of the government’s narrative, state-owned media

14
coverage is expected to operate within the pre-existing NRCs. In the case of Russia, the
familiar account of an aggressive and threatening traditional rival, such as the US, immedi-
ately creates a context by which Russia is justified to intervene in the affairs of its neighbors.
Rather than being perceived by the public as aggressive actions on the part of Russia, within
the narrative of a threatening enemy, military intervention can be justified as a defensive
measure. These expectations lead to the following hypothesis:
Hypothesis 2: The US features more frequently in Russian state-owned news stories prior
to Russian intervention in Ukraine and Georgia.

Data Collection

We focus on Russian-language state-owned media outlets to analyze the Russian govern-

ment’s manipulation of domestic news coverage aimed at maintaining consistency with pre-
existing NRCs. Our focus on analyzing primary sources in their original language offers
several advantages. By analyzing primary sources, we can assess the impact of Russian me-
dia coverage more directly than if we relied on secondary sources compiled by academics, or
English-language state-owned media, such as Russia Today. Secondary sources may reflect
ex post rationalizations or critiques in light of subsequent events, while English-language
outlets may present an account intended for foreign, rather than domestic, audiences.
Analyzing textual data in the original Russian compared to a translation also provides
several advantages. First, while translations are often a practical choice, they are generally
a second-best option, as the process of translating a text can only introduce error, compared
to working with the original. The second reason is more specific to the current application;
the Russian language has a highly inflectional morphology. For example, Moskva, Moskvy,
Moskve, Moskvo, Moskvu are translated as the same word Moscow in English. Working
directly with the original language aids in training the application to recognize each inflection

15
as the same proper name when stemming, reducing the risk of inadvertently missing data.
Our corpus is news headlines from Russian media websites. We focus on analyzing
headlines as they are explicitly written with the intent to succinctly inform the reader of the
content of a news report, thus best characterizing the salient focus of the report. Moreover,
headlines reach, and potentially influence, the broadest set of viewers of a report. For
example, in television reports, headlines are often featured on the screen during the entire
duration of a report; hence, even passive viewers not closely following a report are likely
to exposed to the featured message, such as people in airports or public service offices who
play news programs in the ‘background’. Moreover, and importantly in a country as large
and ethnically diverse as Russia, those who are not completely fluent in Russian are more
likely to understand a headline than audio content. As an example, Table 1 presents ten
headlines (and their English translation) from Russia-24, a major Russian media site, on
1 March 2014.
Our corpus is drawn from two sources. The first one is the main state-controlled Russia-
24. Russia-24, formerly Vesti (“news” in Russian), is a state-owned news channel. Russia-24
promotes the official position of the Russian government and serves as one of the Kremlin’s
propaganda devices that serves to shape the public opinion. Though banned in Ukraine,
Moldova, and Latvia, the network is popular among Russian-speaking minorities in several
neighboring formerly Soviet republics.
The second source is Dozhd (“Rain” in Russian), the only private independent Russian
television channel. As an independent media outlet, Dozhd is not privileged to private gov-
ernment information. Thus Dozhd provides purely events-driven news coverage and functions
as a placebo to the state-owned news coverage in our analyzes.16 While there may be differ-
ences in the degree of domestic vs. international coverage between the two news sources, we
16
In a 2015 article in Walker (2015b), Dozhd’s editor-and-chief described the channel’s mission as an
independent news source: “Nobody needs counter-propaganda. Our goal is not to fight anything, our goal is
just to give people who don’t have real information the thing they don’t have.”

16
 Table 1: Examples of Russia-24 headlines: March 1, 2014
Simferopol~ski a roport vnov~ naqal prinimat~ resy
Simferopol Airport started receiving flights again
V stolknoveni v Har~kove postradali 110 qelovek
In collision in Kharkov, 110 people got injuried
Turqinov mobilizuet ukrainsku armi
Turchynov mobilizes Ukrainian army
Vvod vosk v Krym i otzyv posla: rexenie { za prezidentom
Introducing troops in Crimea and recalling ambassador: the decision is up to the President
Pan Gi Mun ne dast kommentariev po Ukraine do razgovora s Putinym
Ban Ki-moon will not comment on Ukraine before talking with Putin
\Berkutovcev" voz~mut v moskovsku polici
“Berkuvcy” will become part of Moscow police
Lider \Pravogo sektora" obratils k Doku Umarovu za pomow~
Leader of “Right Sector” appealed to Doku Umarov for help
Peskov: Putin primet rexenie o vvode vosk na Ukrainu, ixod iz razviti situacii
Peskov: Putin will make decision on the introduction of troops into Ukraine, based on situation
Fedor Bondarquk: situaci na Ukraine { urok dl Rossii
Fedor Bondarchuk: situation in Ukraine is a lesson for Russia
Timoxenko priletit v Moskvu, qtoby obsudit~ situaci v Krymu
Tymoshenko will fly to Moscow to discuss the situation in Crimea
Note: Each headline is reported in the original Russian with an English translation below it.

would expect the trend lines of news coverage to remain relatively consistent in the absence
of government framing.17
We collected our data by scraping all news headlines from the archives available on the
web pages of Russia-24 and TVRain.18 To do this, we developed a web-application in the
C# programming language to perform this task. First, our application parses the HTML
source of the web-pages. To program our parsing algorithm, we first manually identified tags
embedded within headlines and sequentially requested historical information as far back in
time as possible. Once the application had collected all headlines, we selected proper names.
To do this, we first collected all capitalized words. Next, we stem these words by training
17
In particular, while Dozhd generally provides less foreign coverage than Vesti, Dozhd has provided
important events-driven coverage of the war in Ukraine that is often not reported on state-owned media. For
example, the detention of Dozhd correspondent Timur Olevskiy in May 2015 after reporting on the arrest
of two Russian soldiers captured in Ukraine (Walker 2015b). The report contradicted the official Russian
government argument—repeated by state-owned media—that there are no Russian troops in Ukraine.
18
Russia-24: http://www.vesti.ru/news/ and Dozhd: https://tvrain.ru/news/.

17
the application to treat all inflections of a word as the same word. This is a critical step
while analyzing the text in Russian because, as previously noted, it is a highly inflectional
language. Finally, we checked that all our words are the proper names of interest (country,
city names, surnames) and discarded the rest.
We then count the frequency of proper names in Russian news headlines from featured
on either media outlet’s website to identify the salience of that subject (Hopkins and King
2010).19 We assume that the more frequently a key word appears in the headlines of state-
owned media, the more salient the government wants the topic to be to the public.20 Overall,
we examine a total of 585,615 headlines from Russia-24 for the period 1 January 2006 to
30 April 2016 (4,355,263 words). We review a total of 116,895 headlines from Dozhd between
24 February 2011 to 20 September 2016 (1,048,362 words). Data were collected at the daily
level, but we aggregate the word counts to the monthly level for our analysis in order to
include other covariates.

Changes in Russian News Coverage

We begin with a preliminary examination of the data. The figures report the frequency of a
word count as a proportion of all proper names in the corpus in a given month.
Figure 2 focuses on the frequency of the words Ukraine, Kiev, and Crimea. We focus
on Ukraine and Kiev, as the the frequency of these words may be expected to increase in
response to the Maidan protests, given Russia’s long-standing interest in, and proximity
to, Ukraine. We focus on Crimea because it was occupied and annexed by Russia. Exam-
ining the frequency of these specific words also let us compare media focus on unexpected
19
We rely on counts rather than n-grams for several reasons. First, since the location of a word’s placement
does not matter in the Russian langunage, n-grams are not especially useful. Second, the placement of stories
near one another is consistent with the concept of selective attribution via association. This is especially
true in the case of passive viewers looking at news headlines/chyrons.
20
Hopkins and King (2010) note that observing counts of key words have consistently been sufficient to
extract substantively meaningful features of texts (see also Grimmer and Stewart 2013, 273).

18
 Figure 2: Word Frequencies–Ukraine from Russia-24

Note: Frequency as a proportion of all proper names in the corpus in a given

month. The blue vertical line indicates the start of Maidan and the red vertical line
the annexation of Crimea.

compared to expected events. The Maidan protests and the eventual overthrow of the pro-
Russia Yanukovych regime, for example, would qualify as unexpected events, as the Russian
government had no hand encouraging either event. The invasion of Crimea, on the other
hand, is an expected event, as a military invasion requires premeditation on the part of the
government. In contrasted to unexpected events, state-owned media coverage may presage
expected events, as the government attempts to preemptively frame how the public will
respond.
A preliminary analysis of the data reveals that frequencies of Ukraine and Kiev increase
at the same time as the start and continuation of the Maidan protests, as would be expected
with news coverage of unexpected events. Frequencies of Crimea, on the other hand, begin to
dramatically increase in February, the month prior to the bulk of the Russian invasion and

19
 Figure 3: Word Frequencies–Ukraine from Dozhd.

Note: Frequency as a proportion of all proper names in the corpus in a given

month. The blue vertical line indicates the start of Maidan and the red vertical line
the annexation of Crimea.

two months prior to the formal acknowledgment of Russian forces and subsequent annexation.
Frequencies of the US and EU also increase at this time.
As a point of comparison, Figure 3 displays the word frequencies for Ukraine, Kiev, and
Crimea from Dozhd. Frequencies of Ukraine and Kiev increase with events in Maidan and
the removal of the Yanukovych regime. Frequencies of Ukraine, Kiev, and Crimea all peak
in March, reflecting the timing of actual events in the conflict.
These data provide initial support for our theoretical expectation. Yet, a simple time
series reporting the proporttion of our key words as a function of all proper names within
news headlines does not tell us if their observation is systematic or they can explained by
other factors. The counts could result, for example, from events-driven coverage, rather than
government-directed coverage. In the next section, we systematically analyze the data using

20
a change-point model to identify sudden changes in its underlying data generating process.
In order to separate event-driven coverage from government-directed coverage, we control
for Dozhd (TVRain), an independent news source, when estimating counts of the frequency
of key words. By controlling for event-driven news coverage, we are able to identify when
state-owned media begin to focus on specific topics. We expect that state-owned media will
increase coverage their coverage of the targets of Russian aggression prior to the onset of
military action, as the government attempts to rally support and justify its behavior.

Data Analysis

In contrast to events-driven coverage of independent media, we expect that the Russian

government increases state-owned media coverage of target states prior to military interven-
tion. We also expect that state-owned media increases its coverage, and hence the salience,
of auxiliary issues—such as as its rivalry with the US—to frame the imminent militarized
conflict within the structure of existing NRCs and government-promoted narratives.
In order to identify sudden changes in the structure of the data, we employ an endogenous
Baysian Markov chain Monte Carlo Poisson Change-point model (Chib 1998; Park 2010).21
The estimator is appropriate because (1) our variable of interest is a count of the frequency
of our key words in a given month and (2) it is designed to identify structural breaks in the
underlying data generating process of the frequency of word counts. Change-point models are
a class of time series analysis in which data are generated from multiple temporal regimes
(or states) and the primary quantities of interest are the number and timing of changes
in temporal regimes (Martin, Quinn and Park 2011, 12). In other words, the goal of the
change-point model is to identify the optimal number of sub-periods within a time series, and
21
Change-point models have previously been used to identify time-dependent effects for predictors of
militarized conflict (Park 2010; Nieman 2016a; Thies and Nieman 2017).

21
estimate parameters within each sub-period.22 One advantage of a Bayesian change-point
model is that, in contrast to a frequentist technique such as a Chow test, the model allows
us to calculate measures of uncertainty regarding the specific timing of change-points, i.e.
are the sub-periods distinct or do they gradually change from one sub-period to the next.23
Identifying the timing of change-points allows us to evaluate our first hypothesis concern-
ing changes in the number of stories about a target prior to Russian military intervention.
Parameters on covariates are estimated within each of these temporal regimes and are free
to vary in both effect and direction between them. Examining the effect of there covariates
allows us to assess our second hypothesis concerning whether mentions of the US are asso-
ciated with increases in the number of stories of a target state during a temporal regime in
which a militarized conflict occurred.
The estimator is a Markov model with hidden states and restricted transition properties
(Chib 1996, 1998). Changes from latent states follow a first-order Markov process:
 
 p11 p12 0 ··· 0 
 
 

 0 p22 p23 ··· 0 

 
 .. .. .. .. .. 
p= 
 . . . . . ,

 
 .. .. 

 . . 0 pm,m pm,m+1 

 
 
0 0 ··· 0 1

where pi,j = Pr(st = j|st−1 = i) is the probability of moving to regime j at time t given that
the regime at time t − 1 is i, and m is the number of change-points. We specify our Bayesian
Poisson change-point model as

yt ∼ Poisson (λt ) , t = 1, . . . , T
22
A Poisson model with covariates does not adhere to a known conditional distribution. Our estima-
tor addresses this by taking the log of time between successive events. Fruhwirth-Schnatter and Wagner
(2006) demonstrate that doing so transforms the Poisson regression into that of a linear regression with log
exponential (1) error.
23
Parameter estimates for the Poisson model are drawn from the posterior distribution of the entire state
space; that is, the degree of precision of the estimated change-points affect estimates for the parameters in
the Poisson model (Park 2010; Nieman 2016a).

22
 λ ∼ x0t βm , m = 1, . . . , M

with prior distributions for the Poisson parameters and temporal regime transitions of

βm ∼ N (0, 10), m = 1, . . . , M

pij ∼ Beta (α, β) , m = 1, . . . , M.

To recover estimates for the model, we estimate 20,000 MCMCs, after discarding a burnin
of 10,000 iterations.
We use Bayes Factor to assess model fit and identify the optimal number of change-points
(Chib 1996; Park 2010). Bayes Factor compares two models, treating one as the baseline
m(y|Mi )
model and the other as an alternative model. That is, BFij = m(y|Mj )
where the numerator
is the marginal likelihood for the baseline model Mi and the denominator is the marginal
likelihood for the alternative model Mj is the alternative model. We take the log of the
Bayes Factor so that positive values are evidence in favor of the baseline while negative
values are evidence against the baseline model (Gill 2009, 209).24
Finally, we include several other variables in the model that may influence the frequency
of our variables of interest and that may affect the timing of a structural break. We in-
clude counts of the frequency of US and EU in the Poisson model to account for Russia’s
NRCs, as both have been cast as traditional rivals to Russia by the Russian government.
We also include the level of inflation that Russia experiences in a month. Inflation serves
as an indicator of the status of the economy. Economic struggles are likely to be of in-
terest to the population and receive media attention at the expense of international news.
Monthly inflation data are obtained from the OECD (2016). When appropriate, we include
an indicator variable for the months when the Maidan protests took place. The inclusion
of this dummy variable accounts for the immediate effect of the protests on news coverage.
Any long-term effect, of course, would be captured by a structural break leading to a new
24
Kass and Raftery (1995) provide additional guidelines for interpreting Bayes Factor comparisons; values
between 3-20 offer some support for the baseline model while values >20 provide strong support for the
baseline model.

23
temporal regime. Lastly, in both the Ukraine and Crimea time series, we include a control
variable accounting for observations of the dependent variable from the independent media
outlet Dozhd.25 The inclusion of this variable explicitly separate ‘events on the ground’ from
propaganda, in that independent media can only report what is observable while not being
privy to government plans. This variable cannot be included in the Georgian time series,
however, as unfortunately the Dozhd data are not available at the time of Russian military
intervention.

Empirical Results

We analyze three different dependent variables: the frequency of Ukraine, the frequency of
Crimea, and the frequency of Georgia in Russia-24 headlines. We use a Poisson change-
point model to identify when there is structural change in the frequency of our dependent
variable. In each case, we begin by using Bayes Factor to identify the number of change-
points that best fit the data. Next, we visualize the posterior density for any identified
temporal regimes, as well as posterior probability density of a change-point in a given year,
to identify the timing of structural breaks. Lastly, we report summaries of the posterior
parameter estimates for each endogenous variable within each identified temporal regime.

Ukraine

We begin by identifying the optimal number of change-points/temporal regimes present in

the data. We do this using Bayes Factor to compare the marginal likelihood across models
with m change-points. These results are reported in Table 2. They indicate that M1 has
25
As Dozhd has a larger number of total headlines on a monthly basis, we include the dependent variable
as a proportion of all headline words. An advantage of this approach is that it eases calculations of the
marginal likelihood. Parameter estimates are similar regardless of whether the count or proportion measure
is used. The correlation of proportion and count measures are greater than r = 0.9 for Ukraine and Crimea
for both Russia-24 and Dozhd.

24
 Table 2: Comparing Poisson Change-point Models of the Frequency of
Ukraine in Russia-24 Headlines.

Log(Bayes Factor) M0 M1 M2 M3 M4
M0 0.00 -1833.35 -1832.71 -455.20 -418.8
M1 1833.35 0.00 0.64 1378.10 1414.53
M2 1833.35 -0.64 0.00 1377.46 1413.89
M3 455.20 -1378.10 -1377.46 0.00 36.40
M4 418.8 -1414.53 -1413.89 -36.40 0.00
 
m(y|Mi )
Note: log BFij = m(y|M j)
where BFij is the logged Bayes Factor for models
Mi and Mj . Columns are Mi and rows are Mj .

the best model fit.26 We now focus on the results from this model.
Figure 4 presents the posterior density for the three temporal regimes, as well as posterior
probability density of the change-point in a given year. Figure 4 identifies a change-point
in February 2014. The results indicate that there is a structural break in the data between
January and February 2014. This structural break suggests that the determinants of how
frequently Ukraine appeared in news headlines changed. Figure 4 also indicates that this
break occurred quite sharply. The local means between the temporal regimes increase dra-
matically (38.1 compared to 236.4). A change-point at this location, along with an increase
in the frequencies of Ukraine appearing in the headlines, is consistent with our expecta-
tions expressed in hypothesis 1, that Russian state-owned news stories feature target states
more frequently prior to military intervention. In other words, Russian state-owned media
increased their news coverage before their military intervention occurred.
Lastly, Table 3 displays the summaries of the posterior distribution for each estimate
associated with the covariates. Of primary interest is the regime following the February
2014 change-point—the second temporal regime. The coefficient on US is positive and
significant at traditional levels, after being negative in the first temporal regime. It is worth
26
The results from the Bayes Factor suggest that M1 and M2 are difficult to separate. Both M1 and
M2 , however, identify a change-point in February 2014 and provide nearly identical parameter estimates for
the post-February time regime. See appendix for the results from the M2 model.

25
 Figure 4: Identifying Change-points in the Frequency of Ukraine in Russia-24 Headlines.

Posterior Regime Probability

1.0
0.8
0.6

State1
State2
0.4
0.2
0.0

0 10 20 30 40 50 60
Posterior Density of Regime Change Probabilities
1.0
0.8
0.6
Density

0.4
0.2
0.0

0 10 20 30 40 50 60

Change-point in February 2014. Local means: 38.1, 236.4.

noting that these results hold after accounting for expected media coverage related to the
events taking place in a neighboring state (via the independent news Dozhd variable). The
results indicate that state-controlled media became more likely to write about the US in
the second period, as would be expected if these outlets sought to justify Russia’s military
actions in Ukraine as part of a defensive action against the US. Overall, these results offer
support for hypothesis 2, which expected that stories featuring the US would be associated
with increases in headlines of Ukraine.

26
 Table 3: Poisson Change-point Es-
timates for Frequency of Ukraine in
Russia-24 Headlines.

Feb 2011– Feb 2014–

Feb 2014 Apr 2016
Variable Mean/SD Mean/SD
Count US -0.0027 0.0008
0.0009 0.0004
Count EU -0.0008 0.0002
0.0012 0.0006
Inflation -0.0318 -0.0227
0.0181 0.0040
Maidan 1.0338 -0.1019
0.0725 0.0672
Dozhd 0.5309 36.8044
3.1904 2.5092
Constant 4.1726 5.3283
0.1954 0.1023
Note: Estimates of the mean and stan-
dard deviation are drawn from the
posterior distribution. Results from
MCMCs of 20,000 runs after discarding
burnin of 10,000 draws.

Crimea

Next, we turn our analysis to exploring changes in the frequency of observing Crimea in
Russian news headlines. We use Bayes Factor to identify the change-point model with the
best fit. Table 4 reports the model comparisons and indicates that M3 has the best model
fit to the data.
Figure 5 presents the posterior density for the four temporal regimes and the posterior
probability density of the three change-points. Change-points are identified in April 2012,
February 2014, and November 2015. The structural break in February 2014, the increase
in local means between the second and third temporal regimes (10.5 and 68.8), and the
sharpness of the break, is consistent with hypothesis 1. That is, there appears to be evidence

27
 Table 4: Comparing Poisson Change-point Models for Frequency of Crimea
in Russia-24 Headlines.

Log(Bayes Factor) M0 M1 M2 M3 M4 M5
M0 0.0 -1282.0 -1749.7 -1894 -1608.6 -1652.1
M1 1282.0 0 -467.6 -612.0 -326.6 -370.1
M2 1749.7 467.6 0.0 -144.2 141.1 97.6
M3 1894.0 612.0 144.2 0.0 285.3 241.8
M4 1608.6 326.6 -141.1 -285.3 0.0 -43.5
M5 1652.1 370.1 -97.6 -241.8 43.5 0.0
 
m(y|Mi )
Note: log BFij = m(y|Mj)
is the logged Bayes Factor for models Mi and Mj .
Columns are Mi and rows are Mj .

Figure 5: Identifying Change-points in the Frequency of Crimea in Russia-24 Headlines.

Posterior Regime Probability
0.8
Pr(St= k |Yt)

State1
0.4

State2
State3
State4
0.0

0 10 20 30 40 50 60

Time
0.4
Density

0.2
0.0

0 10 20 30 40 50 60

Time
0.8
Density

0.4
0.0

0 10 20 30 40 50 60

Time
Posterior Density of Regime Change Probabilities
0.8
Density

0.4
0.0

0 10 20 30 40 50 60

Changes in April 2012, February 2014, and November 2015. Local means: 1.0, 10.5,
68.8, 64.5.

28
 Table 5: Poisson Change-point Estimates for Frequency of
Crimea in Russia-24 Headlines.

Feb 2011– Apr 2012– Feb 2014– Nov 2015

Apr 2012 Feb 2014 Nov 2015 Apr 2016
Variable Mean/SD Mean/SD Mean/SD Mean/SD
Count US -0.0135 -0.0205 0.0208 0.0052
0.0145 0.0059 0.0010 0.0017
Count EU -0.0388 0.0597 0.0086 0.0078
0.0155 0.0074 0.0014 0.0042
Inflation -0.1708 0.2091 -0.0778 0.2119
0.1857 0.3243 0.0088 0.0252
Dozhd 0.0100 0.3423 6.2267 -0.0127
3.1584 3.1670 3.0878 3.1597
Constant 6.0511 0.4909 -0.0513 0.2853
2.4156 1.5503 0.2649 0.6348
Note: Estimates of the mean and standard deviation are drawn from
the posterior distribution. Results from MCMCs of 20,000 runs after
discarding burnin of 10,000 draws.

that Russian state-owned news began to feature stories about Crimea in the lead up to
military intervention.
Finally, Table 5 displays the summaries of the posterior distribution for each covariate.
Our focus is on the third temporal regime, which follows the February 2014 change-point.
The coefficient on US is positive and significant at traditional levels, after being negative in
the previous temporal regime. The results hold even after controlling for the expected media
coverage from an independent outlet (via the Dozhd variable). The results are consistent
with those which focused on Ukraine, and again offer support for hypothesis 2, suggesting
that Russian state-owned media invoke the US in effort to justify Russian actions as part of
a defensive effort against a traditional rival.

29
 Table 6: Comparing Poisson Change-point Models for
Frequency of Georgia in Russia-24 Headlines.

Log(Bayes Factor) M0 M1 M2 M3
M0 0.0 -387.1 -465.0 -416.4
M1 387.1 0.0 -77.9 -29.2
M2 465.0 77.9 0.0 48.7
M3 416.4 29.2 -48.7 0.0
 
m(y|Mi )
Note: log BFij = m(y|M j)
where BFij is the logged
Bayes Factor for models Mi and Mj . Columns are Mi
and rows are Mj .

Georgia

In our last set of analyses, we focus on changes in the frequency of observing Georgia in
Russian news headlines. We identify the change-point model with the best fit using Bayes
Factor. Table 6 reports the model comparisons and indicates that M2 has the best model
fit to the data.
Figure 6 presents the posterior density for the three temporal regimes and the posterior
probability density of the two change-points. Change-points are identified in Aug 2008 and
Nov 2008. The structural break in August 2014 is the same month as Russia’s intervention.
The increase in local means, from 33.5 to 183.3, and the probability of the break, with a 30%
chance that it occurred in July, indicating that increases in Russian news coverage began in
mid to late July, is consistent with hypothesis 1. That is, there appears to be evidence that
Russian state-owned news began to feature stories about Georgia prior to the military crisis
and intervention in the break-away republics of Abkhazia and South Ossetia.
Finally, Table 7 displays the summaries of the posterior distribution for each covariate.
Our focus is on the second temporal regime, which follows the August 2008 change-point.
The coefficient on US is positive and significant at traditional levels, after exerting a negative
effect during the previous temporal regime. These results again offer support for hypothesis 2.

30
 Figure 6: Identifying Change-points in the Frequency of Georgia in Russia-24 Headlines.

Posterior Regime Probability

0.8
Pr(St= k |Yt) State1

0.4
State2
0.0 State3

0 20 40 60 80 100 120

Time
0.6
Density

0.4
0.2
0.0

0 20 40 60 80 100 120

Time
Posterior Density of Regime Change Probabilities
0.6
Density

0.4
0.2
0.0

0 20 40 60 80 100 120

Changes in Aug 2008 and Nov 2008. Local means: 33.5, 183.3, 14.0.

The above results provide strong support for each of our hypotheses. In all three times
series—Ukraine, Crimea, and Georgia—there is a change-point in the month prior to Russia’s
intervention. We also find that the local means increase after each of the change-points.
Taken together, this indicates, consistent with hypothesis 1, that Russian state-owned media
increases the news coverage of military targets prior to intervention, beyond what would be
expected by looking just at ‘event-driven’ news.
In addition, in each of the time series we find that the frequency of the US appearing
in headlines is associated with an increase in Ukraine, Crimea, and Georgia as well. This is
consistent with our expectation that Russian state-owned media would increase its coverage

31
 Table 7: Poisson Change-point Estimates for Fre-
quency of Georgia in Russia-24 Headlines.

Jan 2006– Aug 2008– Nov 2008–

Aug 2008 Nov 2008 Feb 2014
Variable Mean/SD Mean/SD Mean/SD
Count US -0.0027 0.0214 -0.0018
0.0018 0.0126 0.0008
Count EU 0.0112 -0.0899 -0.0125
0.0036 0.0214 0.0013
Inflation 0.2065 0.5071 0.0629
0.0197 0.3704 0.0096
Constant 1.2195 -2.8420 3.1620
0.1744 4.3175 0.1446
Note: Estimates of the mean and standard deviation
are drawn from the posterior distribution. Results from
MCMCs of 20,000 runs after discarding burnin of 10,000
draws.

of the US at the same time as their coverage of conflict in their near abroad, to remind the
public of the threats and danger arising from a traditional rival. That mentions of the US
covary positively with those of Ukraine, Crimea, and Georgia in Russian state-owned media
headlines is consistent with hypothesis 2.

Conclusion

We argue that authoritarian governments use media to build support in the lead up to for-
eign intervention, beyond what events-driven coverage would entail. Further, we contend that
state-owned media will add context to this by linking target states to other, traditional ene-
mies. We develop a web-scraping application to investigate and analyze Russian-languange
media coverage using textual analysis (over 700,000 news stories) of Ukraine and Georgia
from 1 January 2006 to 30 April 2016. We analyze this data for changes in news coverage
using an endogenous Bayesian MCMC Poisson change-point model.

32
 We find evidence that Russian state-owned media does initiate coverage of targets in
the months preceding military interventions. Specifically, we find that Russian state-owned
media increased the number of headlines with Ukraine and Crimea prior to the onset of
the March 2014 military invasion of Crimea. We also find that Russian state-owned media
increased coverage of Georgia before the August 2008 military invasions in South Ossetia
and Abkhazia. In both cases, state-owned media coverage intensified prior to the onset of
actual fighting; this is in contrast to the events-driven coverage from independent media,
which increased at the same time as military intervention. Further, there also evidence that
Russian state-owned media is more likely to discuss target states as it invokes a traditional
rival, the US. These findings are consistent with our expectations. The results indicate that
the Russian government directs state-owned media to manufacture support for its foreign
military adventures. In addition, it frames these conflicts in a manner consistent with a
pre-conceived national roles; namely, as part of a broader geopolitical struggle with the US.
Our analysis contributes the growing literature on foreign policy approaches and role the-
ory by analyzing primary data sources to evaluate hypotheses about a state’s national role
conceptions. More broadly, our approach offers an innovate way to gather and analyze data
from authoritarian regimes. State-owned media can be used to formulate and test prediction
about foreign and domestic identities of these states. State-owned media, for example, can
be analyzed by outside actors as a signallying device for authoritarian regimes’ foreign pol-
icy actions. This would be especially useful for policymakers, as authoritarian regimes are
less credible than other regiems when sending signals in the international arena. Analyzing
the content of domestic state-owned media, however, could help identify when authoritarian
regimes are bluffing on the international stage, and when they are mobilizing their resources.
Relatedly, future research could also compare differences in content between intended do-
mestic and foreign audiences, and use these divergences to gain traction on separating the
preferences of authoritarian regimes from their messaging to external audiences.

33
References
Aksenyonok, Alexander. 2008. “Paradign Change in Russian Foreign Policy: The Five-Day
War as a Watershed.” Russia in Global Affairs 6(4):66–76.

Allison, Roy. 2008. “Russia Resurgent? Moscow’s Campaign to ‘Coerce Georgia to Peace’.”
International Affairs 84(6):1145–1171.

Bailey, Michael A, Anton Strezhnev and Erik Voeten. 2017. “Estimating Dynamic State
Preferences from United Nations Voting Data.” Journal of Conflict Resolution 61(2):430–
456.

Breuning, Marijke. 2011. Role Theory Research in International Relations. In Role Theory
in International Relations: Approaches and Analysis, ed. Sebastian Harnisch, Cornelia
Frank and Hanns W. Maull. London: Routledge.

Brummer, Klaus and Cameron G. Thies. 2015. “The Contested Selection of National Role
Conceptions.” Foreign Policy Analysis 11(3):273–293.

Cantir, Cristian and Juliet Kaarbo. 2012. “Contested Roles and Domestic Politics: Reflec-
tions on Role Theory in Foreign Policy Analysis and IR Theory.” Foreign Policy Analysis
8(1):5–24.

Chib, Siddhartha. 1996. “Calculating Posterior Distributions and Modal Estimates in Markov
Mixture Models.” Journal of Econometrics 75:79–97.

Chib, Siddhartha. 1998. “Estimations and Comparison of Multiple Change-point Models.”

Journal of Econometrics 86:221–241.

Chiba, Daina, Carla Martinez Machain and William Reed. 2014. “Major Powers and Mili-
tarized Conflict.” Journal of Conflict Resolution 58(6):976–1002.

Chyzh, Olga and Elena Labzina. 2018. “Bankrolling Repression? Modeling Third-Party
Influence on Protests and Repression.” American Journal of Political Science 62(2):312–
324.

Committee to Protect Journalists. 2017. https://cpj.org.

Coyne, Christopher J. and Peter T. Leeson. 2009. Media, Development, and Institutional
Change. Cheltenham, Gloucestershire, UK: Edward Elgar Publishing.

Davis, David and Will Moore. 1997. “Ethnicity Matters.” International Studies Quarterly
41(1):171–184.

Dowell, Nia M., Leah C. Windsor and Arthur C. Graesser. 2015. “Computational linguis-
tics analysis of leaders during crises in authoritarian regimes.” Dynamics of Asymmetric
Conflict doi:10.1080/17467586.2015.1038286.

34
Egorov, Georgy, Sergei Guriev and Konstantin Sonin. 2009. “Why Resource-poor Dictators
Allow Freer Media: A Theory and Evidence from Panel Data.” American Political Science
Review 103(4):645–668.

Frazier, Derrick and Robert Stewart-Ingersoll. 2010. “Regional Powers and Security: A
Framework for Understanding Order within Regional Security Complexes.” European Jour-
nal of International Relations 16(4):731–753.

Fruhwirth-Schnatter, Sylivia and Helga Wagner. 2006. “Auxiliary Mixture Sampling for
Parameter-Driven Models of Time Series of Small Counts with Applications to State Space
Modelling.” Biometrika 93(4):827–841.

Gehlbach, Scott and Konstantin Sonin. 2014. “Government Control and the Media.” Journal
of Public Economics 118:163–171.

Gibler, Douglas M. 1999. “East or Further East?” Journal of Peace Research 36(6):627–637.

Gibler, Douglas M. and Jamil A. Sewell. 2006. “External Threat and Democracy: The Role
of NATO Revisited.” Journal of Peace Research 43(4):413–431.

Gill, Jeff. 2009. Bayesian Methods: A Social and Behavior Sciences Approach. Second ed.
Boca Raton, FL: Chapman and Hall/CRC Press.

Grimmer, Justin and Brandon M. Stewart. 2013. “Text as Data: The Promise and Pitfalls
of Automatic Content Analysis Methods for Political Texts.” Political Analysis 21(3):267–
297.

Gunitsky, Seva. 2015. “Corrupting the Cyber-Commons: Social Media as a Tool of Autocratic
Stability.” Perspectives on Politics 13(1):43–54.

Hassid, Jonathan. 2008. “Controlling the Chinese Media: An Uncertain Business.” Asian
Survey 48(3):414–430.

Higgins, Andrew and Peter Baker. 2014. “Russia Claims U.S. is Meddling Over Ukraine.” New
York Times 6 Feb. https://www.nytimes.com/2014/02/07/world/europe/ukraine.
html.

Holsti, K. J. 1970. “National Role Conceptions in the Study of Foreign Policy.” International
Studies Quarterly 14(3):260–70.

Hopkins, Daniel and Gary King. 2010. “Extracting Systematic Social Science Meaning from
Text.” American Journal of Political Science 54(1):229–247.

Huang, Haifeng. 2015. “Propaganda as Signaling.” Comparative Politics 47(4):419–437.

Kass, Robert E. and Adrian E. Raftery. 1995. “Bayes Factor.” Journal of the American
Statistical Association 90(430):773–795.

35
King, Gary, Jennifer Pan and Margaret E Roberts. 2013. “How Censorship in China Allows
Government Criticism but Silences Collective Expression.” American Political Science Re-
view 107(2):326–343.

King, Gary, Jennifer Pan and Margaret E. Roberts. 2017. “How the Chinese Government
Fabricates Social Media Posts for Strategic Distraction, Not Engaged Argument.” Amer-
ican Political Science Review 111(3):484–501.

Kydd, Andrew H. and Roseanne W. McManus. 2015. “Threats and Assurances in Crisis
Bargaining.” Journal of Conflict Resolution doi:10.1177/002200271557657.

Lavrov, Sergei. 2006. “The Rise of Asia, and the Eastern Vector of Russia’s Foreign Policy.”
Russia in Global Affairs 4(3):68–80.

Lee, Chin-Chuan. 1998. “Press Self-Censorship and Political Transition in Hong Kong.”
Harvard International Journal of Press/Politics 3(2):55–73.

Levy, Jack S. and William R. Thompson. 2010. Causes of War. Chichester, UK: Wiley-
Blackwell.

Lo, Bobo. 2002. Russian Foreign Policy in the Post-Soviet Era: Reality, Illusion and Myth-
making. London: Palgrave Macmillan.

Lo, Bobo. 2004. “The Long Sunset of Strategic Partnership: Russia’s Evolving China Policy.”
International Affairs 80(2):295–309.

Love, Gregory J and Leah C Windsor. 2018. “Populism and Popular Support: Vertical
Accountability, Exogenous Events, and Leader Discourse in Venezuela.” Political Research
Quarterly p. doi:1065912917749752.

Lucas, Christopher, Richard A. Nielsen, Margaret E. Roberts, Brandon M. Stewart, Alex

Storer and Dustin Tingley. 2016. “Computer-Assistaed Text Analysis for Comparative
Politics.” Political Analysis 23(2):254–277.

MacFarlane, Neil. 2006. “The ‘R’ in BRICs: Is Russia an Emerging Power?” International
Affairs 82(1):41–57.

Martin, Andrew D., Kevin M. Quinn and Jong Hee Park. 2011. “MCMCpack: Markov chain
Monte Carlo in R.” Journal of Statistical Software 42(9):1–21.

McManus, Roseanne W. 2014. “Fighting words The effectiveness of statements of resolve in

international conflict.” Journal of Peace Research 51(6):726–740.

McManus, Roseanne W. 2016. “The Impact of Context on the Ability of Leaders to Signal
Resolve.” International Interactions doi:10.1080/03050629.2016.1171764.

McManus, Roseanne W. and Mark David Nieman. 2018. 56(3):364–378.

36
Miskimmon, Alister, Ben O’Loughlin and Laura Roselle. 2014. Strategic Narratives: Com-
munication Power and the New World Order. Vol. 3 Routledge.

Miskimmon, Alister, Ben O’Loughlin and Laura Roselle. 2017. Forging the World: Strategic
Narratives and International Relations. University of Michigan Press.

Moehler, Devra C. and Naunihal Singh. 2011. “Whose News Do You Trust? Explaining Trust
in Private versus Public Media in Africa.” Political Research Quarterly 64(2):276–292.

Munger, Kevin, Richard Bonneau, Jonathan Nagler and Joshua A. Tucker. 2018. “Elites
Tweet to Get Feet Off the Streets: Measuring Regime Social Media Strategies During
Protest.” Political Science Research and Methods doi:10.1017/psrm.2018.3. Forthcom-
ing.

Nau, Henry and Cory Welt. 2013. “How Russia Sees the World: Domestic Foreign Policy
Debates.” Sigur Center for Asian Studies’ Rising Powers Initiative and the Institute for
European, Russian, and Eurasian StudiesGeorge Washington University Policy Report.

Nieman, Mark David. 2016a. “Moments in Time: Temporal Patterns in the Effect of Democ-
racy and Trade on Conflict.” Conflict Management and Peace Science 33(3):273–293.

Nieman, Mark David. 2016b. “The Return on Social Bonds: Social Contracts and Interna-
tional Conflict.” Journal of Peace Research 53(3):665–679.

Oates, Sarah. 2007. “The Neo-Soviet Model of the Media.” Europe-Asia Studies 59(8):1279–
1297.

OECD. 2016. “Main Economic Indicators.” 2016(10).

Park, Jong Hee. 2010. “Structural Change in U.S. Presidents’ Use of Force.” American
Journal of Political Science 54(3):766–782.

Rahr, Alexander. 2007. “Germany and Russia: A Special Relationship.” Washington Quar-
terly 30(2):137–145.

Redlawsk, David P. 2002. “Hot Cognition or Cool Consideration? Testing the Effects of
Motivated Reasoning on Political Decision Making.” Journal of Politics 64(4):1021–1044.

Roselle, Laura, Alister Miskimmon and Ben O’Loughlin. 2014. “Strategic Narrative: A New
Means to Understand Soft Power.” Media, War and Conflict 7(1):70–84.

Rozenas, Arturas and Denis Stukal. 2018. “How Autocrats Manipulate Economic News:
Evidence from Russia’s State-Controlled Television.” Journal of Politics. Forthcoming.

Schimpfossl, Elisabeth and Ilya Yablokov. 2014. “Coercion or Conformism? Censorship and
Self-censorship Among Russian Media Personalities and Reporters in the 2010s.” Demokra-
tizatsiya 22(2):295–311.

37
Shi, Tianjian, Jie Lu and John Aldrich. 2011. “Bifurcated Images of the U.S. in Urban China
and the Impact of Media Environment.” Political Communication 28(3):357–376.
Stockmann, Daniela. 2011. “Race to the Bottom: Media Marketization and Increasing Neg-
ativity Toward the United States in China.” Political Communication 28(3):268–290.
Stockmann, Daniela and Mary E. Gallagher. 2011. “Remote Control: How the Media Sustain
Authoritarian Rule in China.” Comparative Political Studies 44(4):436–467.
Taber, Charles S. and Milton Lodge. 2006. “Motivated Skepticism in the Evaluation of
Political Beliefs.” American Journal of Political Science 50(3):755–769.
Thies, Cameron G. 2013. The United States, Israel, and the Search for International Order:
Socializing States. New York: Routledge.
Thies, Cameron G. and Marijke Breuning. 2012. “Integrating Foreign Policy Analysis and
International Relations through Role Theory.” Foreign Policy Analysis 8(1):1–4.
Thies, Cameron G. and Mark David Nieman. 2017. Rising Powers and Foreign Policy
Revisionism: Understanding BRICS Identity and Behavior through Time. Ann Arbor,
MI: University of Michigan Press.
Trenin, Dmitri. 2009a. “Reimagining Moscow’s Foreign Policy.” Foreign Affairs 88(6):64–78.
Trenin, Dmitri. 2009b. “Russia’s Spheres of Interest, Not Influence.” The Washington Quar-
terly 32(4):3–22.
Truckos, Judit. 2021. “Comparing Russian, Chinese, and American Soft Power Use: A New
Approach.” Global Society 35(3):395–418.
Tsygankov, Andrei P. and Matthew Tarver-Wahlquist. 2009. “Duelling Honors: Power,
Identity and the Russia–Georgia Divide.” Foreign Policy Analysis 5(4):307–326.
Vartanova, Elena. 2012. The Russian Media Model in the Context of Post-Soviet Dynamics.
In Comparing Media Systems Beyond the Western World, ed. Daniel C. Hallin and Paolo
Mancini. Cambridge University Press New York, NY pp. 119–142.
Walker, Shaun. 2015a. “Putin Admits Russian Military Presence in Ukraine for First
Iime.” The Guardian 17 December. https://www.theguardian.com/world/2015/dec/
17/vladimir-putin-admits-russian-military-presence-ukraine.
Walker, Shaun. 2015b. “TV Rain: Inside Russia’s Only Independent Tele-
vision Channel.”. https://www.theguardian.com/cities/2015/jun/09/
tv-rain-russia-only-independent-television-channel.
Walker, Stephen G. 1992. Symbolic Interactionism and International Politics: Role Theory’s
Contribution to International Organization. In Contending Dramas: A Cognitive Approach
to International Organizations, ed. Martha Cottam and Chih-Yu Shih. New York: Praeger.

38
Warren, Camber. 2014. “Not by the Sword Alone: Soft Power, Mass Media, and the Pro-
duction of State Sovereignty.” International Organization 68(1):111–141.

Warren, Camber. 2015. “Explosive Connections? Mass Media, Social Media, and the Geog-
raphy of Collective Violence in African States.” Journal of Peace Research 52(3):297–311.

Wendt, Alexander. 1994. “Collective Identity Formation and the International State.” Amer-
ican Political Science Review 88(2):384–396.

Windsor, Leah. 2018. “The Language of Radicalization: Female Internet Recruitment to

Participation in ISIS Activities.” Terrorism and Political Violence pp. 1–33.

Windsor, Leah C., Nia Dowell and Art Graesser. 2014. “The Language of Autocrats: Leaders’
Language in Natural Disaster Crises.” Risk, Hazards & Crisis in Public Policy 5(4):446–
467.

Windsor, Leah, Nia Dowell, Alistair Windsor and John Kaltner. 2017.
“Leader Language and Political Survival Strategies.” International Interactions
doi:10.1080/03050629.2017.1345737.

Zaller, John. 1992. The Nature and Origins of Mass Opinion. Cambridge University Press.

39
Appendix

Degree of Russian and US Support signaled for Georgia and Ukraine

Figure 7 shows the degree of support signaled by Russia and the US towards Georgia and
Ukraine, respectively, from 2000 to 2012. Signaled support is a latent variable based on
several foreign policy actions that a major power can take to demonstrate support for a
(potential) protégé, such as forming an alliance, selling arms, deploying troops or nuclear
weapons, conducting joint military exercises, or having a leader visit or offer public state-
ments of support. Signaled support data are obtained from McManus and Nieman (2018).
Figure 7 makes clear that while Russian support for both Georgia and Ukraine are rela-
tively constant, US support for each of the former Soviet republics dramatically decreased in
2008. The continuity in Russian support, and the decrease in US support, holds despite the
gradual shift in foreign policy affinity from Moscow to Washington on behalf of both Tbilisi
and Kyiv displayed in Figure 1.

Ukraine: 2 Change-points

Table 2 could not definitively say that M1 is more likely than M2 (Kass and Raftery 1995).
We report the results from M2 below.
Figure 8 presents the posterior density for the three temporal regimes, as well as posterior
probability density of the change-point in a given year. Figure 4 identifies change-points in
January 2013 and February 2014. Consistent with the previous results (i.e. Figure 4), these
results indicate that there is a structural break in the data between January and February
2014. Figure 8 also indicates that this break occurred sharply. The local means between the
temporal regimes increase dramatically (39.5 compared to 236.4). This change-point and the
increase in local means is again consistent with our expectations expressed in hypothesis 1,
that Russian state-owned news stories feature target states more frequently prior to military

40
 Figure 7: Degree of Support Signaled by Russia and the US.

Note: Signaled support data from McManus and Nieman (2018).

intervention.
Table 8 displays the summaries of the posterior distribution for each covariate. Focusing
on the third temporal regime—February 2014 to April 2016—it is clear that the coefficient
on US is positive and significant at traditional levels. US is also significant in the second
temporal regime—January 2013 to February 2014—suggesting that increasing coverage of
the US by Russian state-owned media was associated with increasing coverage of Ukraine
during the final year of Yanukovych’s government.

41
Figure 8: Identifying Change-points in the Frequency of Ukraine in Russia-24 Headlines.
Posterior Regime Probability
1.0
0.8
Pr(St= k |Yt)

0.6

State1
State2
0.4

State3
0.2
0.0

0 10 20 30 40 50 60

Time
0.8
0.6
Density

0.4
0.2
0.0

0 10 20 30 40 50 60

Time
Posterior Density of Regime Change Probabilities
1.0
0.8
0.6
Density

0.4
0.2
0.0

0 10 20 30 40 50 60

Change-point in January 2013 and February 2014. Local means: 37.3, 39.5, 236.4.

42
Table 8: Poisson Change-point Estimates for Fre-
quency of Ukraine in Russia-24 Headlines.

Feb 2011– Jan 2013– Feb 2014–

Jan 2013 Feb 2014 Apr 2016
Variable Mean/SD Mean/SD
Count US -0.0006 0.0052 0.0008
0.0015 0.0024 0.0004
Count EU -0.0028 0.0086 0.0002
0.0013 0.0050 0.0006
Inflation -0.0185 -0.2525 -0.0227
0.0180 0.1411 0.0040
Maidan -0.0118 1.3483 -0.1002
3.1659 0.1410 0.0675
Dozhd 0.1202 0.2458 36.84861
3.184 3.1636 2.5235
Constant 4.0084 3.3791 5.3286
0.2288 1.1181 0.1013
Note: Mean and standard deviation estimates drawn
from the posterior distribution. Results from MCMCs
of 20,000 runs after discarding burnin of 10,000 draws.

43