From Cyberspace to Independence Square: Understanding the Impact of Social Media on Physical Protest Mobilization During Ukraine’s Euromaidan Revolution

ABSTRACT

Cyberspace has dramatically affected social and political movements over the last 10 years. There is anecdotal evidence to suggest a strong role for information communications technologies (ICTs) in a changing landscape of organization and mobilization for large-scale social and political movements in countries around the world. This paper analyzes the 2013–2014 Euromaidan revolution in Ukraine using big data collected from open source content including blogs, Twitter, Facebook, YouTube, forums, and news sites and mixed methods and assess the ramifications and impact of social media on social mobilization. This work compares digital and physical engagement directionality and finds the impact of social media on physical protest turnout is significant and leads to increased numbers of protestors in the streets. The analysis also highlights means of developing and assessing social mobilization through the use of linguistically and regionally categorized social media.

KEYWORDS:

• Internet
• social mobilization
• protest
• revolution
• Ukraine
• social media

The impact of social media on revolutions has grown in importance in recent years. Much of the literature to date has centered on the importance (or lack thereof) of technology’s influence on citizen-led movements. The impact of social media is either found to be significant or suspect in its true ability to affect regime change (Howard, Agarwal, & Hussain, 2011; Shirky, 2011, December). There is little hard data to confirm or disprove either side of the assessments above. Anecdotally, the use of social media appears from the sidelines of any particular mobilization to heighten the interest of individuals both within a country and those external to it. The volume of messages generated serves as a stand in for value and importance with limited consideration as to the actual impact of any given social media engagement.

By contrast, the present work approaches the influence of social media with skepticism and data to present a dynamic analysis for where and when digital tools help to affect real-world change and where these tools fail. The picture is not black and white, it is clouded with shades of gray and technology is by no means a silver bullet. Yet by developing a rigorous and contextualized understanding of when and where the use of technology enhances social movements, this paper provides insight into how technology and people interact to develop successful movements to affect regime change.

This work focuses on Euromaidan and leverages large-scale data and machine learning. I examine the Euromaidan movement in Ukraine across geographic and linguistic divides, both within Ukraine and internationally. For the below analysis, I leverage data from blogs, twitter, Facebook, Sina-Weibo, YouTube, forums and news sites. I also integrate data from Global Database on Events, Language and Tone (GDELT) to compare user-generated sentiment to physical turnout at protests. Data were examined in three languages, English, Russian, and Ukrainian using native linguists and focus groups for confirmation of assisted machine learning sentiment analysis. The result is a combined effort to sort through complex causal frameworks of online and offline behavior.

Embedded within this research effort are several puzzles. First, do ICTs help or hurt in the development of social movements? Do they create social cohesion or discord in the creation of a social movement? Second, why is so much of the content generated in foreign languages during social movements? Who are the target audiences of this foreign language content generation? And how does the generation of content in foreign languages affect the organization of physical movements to achieve domestic social change? Third, what happens first? Does large online social media engagement predate physical turnout in the streets or does physical turnout lead to increases in online content generation? Combined, these are perplexing questions with both policy and political implications. These questions have implications for future organizers of revolutionary and social movements as well as international actors who might seek to support such movements.

This research fills a gap in the literature by picking up where direct observational case analyses and more limited data samples have left off. Here a holistic picture of events using real time data is presented. This research does not obviate the work of previous studies but adds to it. The paper proceeds in seven sections below. Section one provides background on the Euromaidan movement and its importance. Section two briefly addresses the breadth of the existing literature on social movements. Section three places the literature in the context of existing studies on social media mobilization. Section four presents the methods used and why they were chosen. Section five applies the methods and analyzes the events that transpired in Ukraine. Section six provides a concise discussion on the importance and value added of the results. Section seven concludes by tying together the themes across the literature and the analysis within this work to chart a path forward for assessing the impact of social media on social mobilization.

Social mobilization in ukraine: a background

This paper focuses on a specific social movement from Ukraine that transpired from November 2013 through March 2014. On November 21, 2013 the President of Ukraine, Viktor Yanukovych, contrary to the expectations of many, failed to sign an association agreement with the European Union and instead decided to focus on strengthening economic ties between Ukraine and Russia (Kelley, 2013; Poltorakov, 2015). The failure to sign the European Union Association agreement was a political siren call for the revolution that was to sweep Ukraine. The causes of Euromaidan cannot simply be attributed to the failure of the Ukrainian President to sign an international agreement, but are the result of a combination of factors including rampant corruption, police intimidation, harsh laws (which have since been repealed), and a cozying up to Russia at the expense of European aspirations (D’Anieri, 2006; Kudelia, 2014; Velychenko, 2007) . Beyond these issues a substantial concern highlighted by Shveda and Park (2016) is substantial economic grievance within Ukraine that in the months prior to revolution peaked at nearly 42.1% unemployment in youth between the ages of 14 and 35. As many previous works have noted, the failure to sign the Association Agreement and Deep Comprehensive Trade Area (DCFTA) agreement was a spark that ignited simmering grievances that spanned economic and corruption concerns, to issues related to the rule of law (Shveda & Park, 2016) .

Euromaidan not only resulted in hundreds of thousands of Ukrainians filling the streets in cities across the country, they also flooded the Internet with news, comments, calls for help, and cries for organized mobilization (Onuch, 2015). They mobilized a diversity of resources both nationally and globally to aid in a political and social upheaval that has had far reaching ramifications both for Ukrainian and European security. The episodic nature of Euromaidan’s contentious performances and novel repertoires of the movement included the seizure of government buildings, mobilized self-defense forces and a blue and yellow piano where Ukrainian musicians pounded out soulful tunes. Olga Onuch (2015) found through rapid surveys during the protests a remarkable diversity of protest participants representing both those who speak Russian and those who speak Ukrainian. Additionally, Onuch found a diversity of motivations for the movement across age groups and protestor types. Moreover, Onuch and Sasse (2016) contend that the revolution took place in 6 distinct phases during which different groupings of individuals and organizations exerted power within the revolution.

Euromaidan was waged on two fronts, on the streets of both the capital Kyiv and other major Ukrainian cities, and online. But to understand the origins of the movement, its evolution and its novel repertoire of contentious performances requires understanding the underlying nature of the Ukrainian experience. Onuch and Sasse (2016) data from rapid surveys indicates also indicate a lack of unified revolutionary coherence within various groups, including across age, region, political class, and political or Ukrainian aspirations. Their findings challenge the ability to foster or even track a unified set of terms or keywords in an online environment. A diversity of linguistic repertoires associated with divergent organizational structures, goals and outcomes limits the validity of data collected during the movement selected on the basis of particular terms identified through interactions with individuals on the ground. Despite the dynamism identified by Onuch and Sasse, this work builds from a more consolidated linguistic starting point for identifying the language popularly utilized online during the revolution. The terms defined below were selected through conversations with NGOs and organizations developed through work at the National Democratic Institute for International Affairs.

Ukraine is a fascinating case study in civil disobedience because it has experienced several significant political movements in the last decade and a half. Ukraine differs from many other countries both by its geography and its tumultuous history over the last several hundred years (Plokhy, 2015; Sakwa, 2016; Wilson, 2000). Ukraine straddles the divide between Europe and Eurasia, and, consequently, European Union and Russian influence (Beisswenger, 2007). Even the word Ukraine connotes meaning that few other countries’ names possess, derived from the word ukraina, it translates closely to mean “the borderlands.”

There are few other countries that suffered more in the 20^th century than Ukraine. Ukraine, between 1917 and 1921 failed to establish itself as a successful independent state and was subsequently broken up amongst the newly re-established Poland and the USSR. Between 1932 and 1933 Ukraine suffered approximately 4.5 million deaths due to forced starvation in what has become known as Holodomor or the “hunger-extermination” at the hands of the Stalinist regime (Plokhy, 2016). Less than a generation later, World War II ravaged the country resulting in upwards of 6.8 million additional fatalities or 16.3% of the population. If Ukraine’s early 20^th century misery and woe weren’t enough, it also experienced what is regarded as one of the two largest nuclear disasters in history in the meltdown of reactor four at Chernobyl. The accident killed more than 36 people, irradiated thousands more and rendered large swaths of Ukraine’s famously rich agricultural areas unable to participate in the global economy due to contamination.

War and other man-made disasters are only the start of Ukraine’s problems. Ukraine has been marred by extreme corruption in its post-soviet era. Transparency International gave Ukraine a score of 25 out of 100 and ranks it 144 out of 177 countries globally.¹ Endemic corruption remains a significant source of political grievance within Ukraine. While corruption is rampant the population is highly-educated with near 100% literacy and high secondary and post-secondary educational attainment. Corruption, government ineffectiveness, a highly-educated population combined with low or stagnant economic growth over the last 5 years,² deteriorating public infrastructure particularly outside of the major cities combine to seemingly establish a grievance-based cause for potential social mobilization in opposition to established political elites. Yet the resultant protests extended beyond grievance-based causes for protests directly into new social movement theories based on ideational notions of civilizational place. Even the name of the movement itself “Euromaidan” implies a civilizational construct that differed from previous protests such as the Orange Revolution. The Orange Revolution of 2004, while inclusive of a European place for Ukraine were intrinsically centered on a Ukrainian political awakening (Jekel Cik, 2007). By Contrast, the Revolution of Dignity did not have a single focal point like the 2004 Orange revolution, but was instead multifaceted and included a mix of concerns (Popova, 2014).

Ukraine is a complex and dynamic nation having been cobbled together and torn apart by various empires over its history. The process of forming modern Ukraine crosses national, civilizational and linguistic lines. These lines are very often examined through the challenges of ethnicity and language in Ukraine (Pop-Eleches & Robertson, 2018). Although some scholars contend that language and ethnicity are poor lens through which to view politics and identity in Ukraine (Kulyk, 2011), here the widely held contention that ethnicity and language matter is examined both for data purposes and based on a breadth of historical literature on the evolution of Ukraine to its modern political state. This is necessarily a resort to parsimony in an effort to more fully understand a specific mechanism, social media, in a complex social movement.

In addition to being ethnically, religiously, and linguistically complicated, Kyiv, the capital of Ukraine, is located above 50 degrees North. During the winter it is not uncommon for the country to experience significant amounts of snow and long periods of sub-zero temperatures. It was under these harsh environmental conditions that protestors who took to the streets.

During the revolution, Ukraine, a country of 44.43 million people³ and an online population of 19 million people⁴ produced more than 2 million geographically locatable unique pieces of content on social media networks related directly to the events happening on the streets of Ukraine from November 20, 2013 until March 1, 2014.⁵ The volume of geographically identifiable content is all the more surprising when regime attempts at suppressing the movement are taken into consideration (Brantly, 2014). Ukraine’s social movement fits well within existing conventional literatures on social and political mobilization taken up in the next section.

The mechanics of social mobilization

Often scholarly analysis of various political and social movements focuses on various structural causes of social movements, including the impact of concepts such as social capital, political system, economics of greed and grievance and many more to name just a few (Crawford & Naditch, 1970; Muller, 1972; Snyder & Tilly, 1972) . The literature surrounding social movements has been enhanced in the last two decades with political transitions across Eastern Europe, North Africa, Asia and the Middle East. Each political transition has added new insights into the complex mechanisms that function to facilitate large-scale social and political change. More recent debates surrounding social and political movements highlight the role of technology in facilitating, influencing or even making possible large scale social and political organization. Despite substantial previous analysis, there has been little systematic data to adequately support what role technology and cyberspace writ large plays in such movements. This paper deconstructs the casual mechanisms and relationships of the Ukrainian movement to derive a clearer picture for scholars and practitioners as to the role and impact information communications technologies (ICTs) and, in particular, social media have on large scale social and political movements.

Questions of cyberspace mobilization take root in historical social mobilization theories based on class, societal integration, and individual interest aggregation and evolutionary belief structures (Tilly, 1998). Here the focus centers on the root concepts associated with episodic contentious politics (McAdam, Tarrow, & Tilly, 2008). Episodic contentious politics are defined as:

“episodic, public, collective interaction among makers of claims and their objects when (a) at least one government is a claimant, an object of claims, or a party to the claims and (b) the claims would, if realized, affect the interests of at least one of the claimants.” (McAdam et al., 2008)

Episodic behaviors incorporating information communications technologies, in particular social media and new media, influence social mobilization both on and offline. The analysis below is constrained to episodic movements rather than continuous political organization and rhetoric around normalized processes of states or similar political entities. By honing the focus of analysis, events that are of contention are isolated. These events affect domestic change through the engagement of both domestic and international audiences (to include diaspora, foreign political and social entities).

The Revolution of Dignity (Euromaidan) in Ukraine at the heart of this analysis is rooted in what is referred to as a contentious performance (McAdam et al., 2008; Tilly & Tarrow, 2007). The contentious performances that transpired in Ukraine are inclusive of demonstrations defined as “the orderly passage through a public space of an organized collectivity on behalf of some claim, identity, or program” (Boix & Stokes, 2007). Beyond demonstrations, there is substantial evidence of individual civil society and political elite involvement at various stages of the social movement and playing varying roles in the development and relative success achieved in the Euromaidan saga (Onuch & Sasse, 2016). Due to the inclusion of both contentious performances, such as demonstrations, and the involvement of civil society and political elites, Euromaidan is constitutive of a social movement consisting of:

“a sustained challenge to power holders in the name of a population living under the jurisdiction of those power holders by means of public displays of that population’s worthiness, unity, numbers, and commitment” (Tilly, 2008) .

Social movements are complex and are inclusive of many moving parts. However, for simplicity, this examination focuses on the impact of social media and other ICTs at the nexus of what Brett Rolfe refers to as “electronic repertoires of contention” and what Tilly and Tarrow simply refer to as repertoires of contention (Rolfe, 2005). Within these repertoires of contention, this analysis establishes a relationship between the physical and the digital. Close examination of the construction and maintenance of Euromaidan’s development and organization demonstrate that it extended beyond contained contention and included transgressive contention, a condition in which not all actors involved in the movement were previously established (Onuch & Sasse, 2016) . The development of new actors throughout the period of contention fostered collective action that innovated or diverged from previous episodes (McAdam et al., 2008; Onuch & Sasse, 2016). The development of repertoires across the physical-digital divide likely facilitated transgressive contention. The increasingly transgressive nature of the movement resulted in what Charles Tilly (2008) generalizes as a reaction to central government actions to limit the evolving collective claims making of claimants. By limiting collective claims through the physical use of force in the streets, digital means of surveillance such as mobile phone tracking, and manipulations of the media, the central government further increased the public’s willingness to engage and further develop the social movement.

Social media can alter the conventional resource environment for social mobilization and make possible, through ease of access and low cost, new forms and scales of resource mobilization. Historically, mobilization for social movements has been considered a challenge due to financial, media, labor, organization and legitimacy issues (Tilly, 2008) . In particular, a lack of resources constrain social movements, who often struggle to gain access to various forms of media to promote and frame messages to draw in recruits for mobilization (Gamson & Wolfsfeld, 1993) . Social media minimizes some types of resource constraints and challenges state control over information resources such as the media. Moreover, many of the resources necessary to conduct a successful social mobilization are interactive and can influence one another. Social media, unlike traditional, print, radio and television can serve as a foundation upon which to organize, plan and mobilize everything from physical protestors to monetary or physical resources necessary for a given movement.

Often a state attempts to minimize and downplay social mobilization within conventional media to mitigate its elevation to the level of significant collective action. Some studies have identified the active use of media and Internet as constraining mechanisms (censorship, prosecution for particular types of content, filtering, etc.) in China (King, Pan, & Roberts, 2012). China is a powerful example of the use of state control to manage media and identify those forms of information that individuals and organizations might utilize to mobilize. While China manages its information environment in near real-time, the same is not true of most other countries due to limited governmental resources. Whereas, Chinese media and Internet controls deny potential resources for collective actions, most other states lack this capacity. States that place great emphasis on removing content can create significant barriers limiting access to information and thereby limit mobilization both online and offline.

To combat state efforts to create resource deficiencies and prevent collective action for social mobilization individuals, organizations and dissident political elites actively attempt to circumvent these efforts. Movements capable of circumventing state censorship efforts and other forms of communication restrictions through the use of novel tools to capture both domestic and international audiences can (although it is not a certainty) do so to enhance their access to resources (Jardine, 2018). Increased access to resources elevates the power (i.e. ability to influence) of an organization or movement relative to the state (Gamson & Wolfsfeld, 1993) . The result is a struggle over framing of an issue or idea. Gamson and Wolfsfeld (1993) identify three elements that affect movement actors: (1) standing, the consideration given to a movement by traditional media; (2) framing, the elevation of the central idea or issue of a given group; and (3) sympathy, the likelihood that the group will garner sympathy from relevant members of the public. The alleviation of resource deficiencies and the establishment of a repertoire of contention offers a promise for sustained social movement by increasing access to potential human and financial capital.

Sustaining a social movement over the long term comes with numerous challenges. As resources, both physical and virtual, are secured, the challenges of collective action become increasingly problematic. Mancur Olson (1965) established that despite increased resources and a strong repertoire of contention, the rational behavior of any given individual is to free-ride on the on the collective actions of others. Although the free-riding problem can be minimized through creative interactions via novel social media campaigns, the ability to move individuals from a virtual to a physical environment remain challenging under the best of circumstances. Moreover, as a movement becomes more successful, it is necessary to understand the ideational-cultural attributes that motivate individuals to participate in a given social movement beyond the virtual world.

The free-rider problem within social mobilization is substantial, but it is only one of the many challenges faced in mobilizing a population. Bruce Bimber (1998) highlights many of the challenges across the literature in determining the value and impact of the Internet in establishing a significant political movement. He identifies a populist claim that often rests at the core of Internet studies indicating that communication capacity can function as a limiting factor in political engagement (Bimber, 1998). Bimber (1998) notes that although the functional information capacity of humans has been increasing steadily for quite some time with the advent of mass printing, radio, television and other information sources, the ability to synthesize information from complex and divergent sources remains a function of capacity and motivation of individuals. The inattentiveness of individuals and their desire to seek out information unevenly results in a contestation of the notion that a controlled information environment is circumvented simply through provision of new information communications technologies.

There is little doubt that social media and other related technologies are adding tools and resources to the mix of social movements and likely adding resources and mechanisms by which individuals and groups can more efficiently and effectively develop repertoires of contention. What is unclear, however, is how impactful these new technologies are on the achievement of social movement goals. Although social media and new media can both foster transgressive movements that expand a social movement beyond a core cadre of individuals and alleviate some resource constraints, their fundamental value to any given social movement is likely to be conditional on a variety of factors. To establish the importance of social media on mobilization, it is necessary to assess the impact of these new tools and techniques as applied to prior social movements. The next section picks up where this one leaves off and briefly examines a sampling of the findings from many recent analyses conducted on movements associated with the “Arab Spring.”

The impact of social media on social movements in the Arab awakening and beyond

The Arab Spring was a diverse series of semi-connected social movements across North Africa and the Middle East. The analyses of the various movements provide divergent accounts of the impact of technology on social mobilization. These accounts are important when assessing the impact of social media on social mobilization. Case analyses on the impact of social media on mobilization during the Arab Spring diverge widely. While the relative value of social media diverges, all cases generally agree there was some impact.

There are diverging perspectives on the significance of social media’s impact on the events within each of the countries experiencing large scale social mobilizations. For instance, Wilson and Dunn (2011) found little indication that digital media was significant in mobilizing in Tahir Square. Nuancing the debate further, John Alterman (2011) notes that one of the more significant aspects of social media in influencing social behavior was its role in facilitating identity development within activists and their ability to conceive of themselves as such. However, Alterman (2011) also notes that many studies to date have used improper methodological frameworks and thereby biased their analyses with preconceived assumptions that are not supported by fact. Marc Lynch (2011) was more circumspect in his analysis, indicating that technology was one of many potential drivers of change and provides an optimistic assessment that technology is impacting the political environments in the Middle East.

Perhaps the most damning assessment of the use of social media comes from Brym, Godbout, Hoffbauer, Menard, and Zhang (2014) when they wrote “commentary (on social media’s importance) says more about the triumphalist biases of the mass media in the USA than about the actual sources of protest in Egypt in 2011”. Specifically, they note that protests in the Egyptian case were long in coming and the result of the pre-existing grievances formalized in networks of civic associations (Brym et al., 2014). In contrast, Victoria Carty (2015) is more optimistic in her analysis of the impact of technology on social movements. Carty contends: “that at least initially digital media – a new and critical resource – provided new venues for expressing grievances, an activity that was previously unimaginable in these relatively shielded and oppressed societies”. Lastly, Andy Carvin’s Distant Witness (2012), provides a rousing account of the diverse connection and impact social media had on global understanding and engagement in the movements occurring in Egypt, Tunisia, Libya and elsewhere. In his work he establishes a level of significance in the ability of crowds to mass report on events in real time. Within this mass reporting, he identifies a potential shift in how scholars understand and interpret modern social mobilizations.

Nuance and a rigorous approach to understanding causal relationships are important in determining the effect of social media on social movements. The popular accounting of revolutions often involves the correlation of technology use and movement outcome with a systematic theory-driven approach to understanding the role technology plays in civil disobedience and the potential for regime change. It is further important to recognize that there are factors involved in the success or failure of a moment that extend beyond the individual protestor or even domestic political environments. A spectrum of external actors influences movements through various explicit, implicit and unforeseen mechanisms of support (Bob, 2001). These actors can be political or commercial, domestic or foreign.

A civic movement that topples dictators and overthrows corrupt regimes cannot be done through digital means alone. Digital citizen mobilization is insufficient to effect on the ground policy and regime change in most instances because it lacks the visceral and kinetic actions that bodies in the streets conveys to even the ineptest of governments. Successful protest movements need a multi-vectored approach that leverages digital networks, human networks, and what can best be described as sneaker networks that stand at doorsteps of governments and demand change. A thousand tweets are unlikely to have the same impact as a thousand protestors banging on trashcans has. Despite good intentions, the digital power of citizens is not equal to their physical presence. And for the foreseeable future this is likely to remain true.

How impactful is social media in social movements?

If social media begins to mitigate resource imbalances, fosters organization and engagement, how important is it to social mobilization? To assess the value of social media empirically requires understanding who is creating content, where they are creating it, in what language they are creating it and what the impact of this content is. Below I list four hypotheses that develop the relationship of social media and social mobilization in Ukraine. In developing the relationship, the objective is to begin moving beyond case analysis to the utilization of machine learning and quantitative analyses. Quantitative analysis is necessarily informed by and dependent upon robust antecedent case analysis. Each of the hypotheses below is tailored to understanding social mobilization over the duration of the Ukrainian Revolution of Dignity. Although these hypotheses are not generalizable to all social mobilizations, the mechanisms they expose in the utilization of social media to foster repertoires of contention will likely have relevance beyond the Ukrainian case and can be tested for in subsequent work.

Building a social movement that is transgressive requires the establishment of broad-based support within a population. Isolating those roadblocks to the formation of a transgressive movement first requires understanding how populations within a country relate to various attempts at mobilization. Of specific interest is the relationship of Ukrainian citizens to digital attempts to mobilize. Much has been made of the strong regional and linguistic fractionalization within Ukraine (Barrington & Herron, 2004; Metzger, Bonneau, Nagler, & Tucker, 2016). Yet, the role and impact of social media in understanding the divisions between Ukrainian and Russian, Eastern and Western Ukraine are underexplored. The ethnic, religious, geographic, and economic distribution of Ukraine has long been contentious. Cultural theories of social movements indicate that social movements can be heavily influenced by how individuals identify themselves (Giddens, 1991). The collection of social media by language and geographic region allows for the movement to be examined not only in aggregate, but also in real-time to assess the importance of language and region on the unity of the movement across the country. The first two hypotheses utilize posts by day, by language and by region to assess the divisions within the movement.

Hypothesis 1: proposes that individuals generating content online in the Ukrainian language were likely to be significantly more supportive of the Euromaidan movement than their counterparts in Russian and English. This first hypothesis establishes a connection between linguistic cleavages and support for movement goals and objectives. To assess whether the movement was truly impactful across Ukrainian society requires an understanding of who is engaging and supporting the movement. This hypothesis seeks to determine whether social media helps or hurts efforts to transgress linguistic barriers and foster an inclusive movement. If the movement was unable to transgress language barriers, its overall effectiveness nationally might be reduced and signal potential conflicts or challenge the ability to achieve proximate movement goals. It might also expose opposition to social mobilization within certain populations that might hinder the overall goals of mobilization.

As Russian, Ukrainian and English-speaking individuals express themselves online, they highlight the contours of support for the movement and across society more broadly in a way that simply looking at the volume of posts or physical turnout in the streets cannot assess. Because Ukraine is linguistically fragmented, language serves as a tool by which to measure aggregate societal support for mobilization and in particular the utility of mobilization online.

Prior to Euromaidan, Ukrainian identity was still in flux nationally (Korostelina, 2013). Events such as the Orange Revolution helped to codify and foster the national identity to a point, but surveys conducted in the years and months prior to the revolution indicate significant fractionalization nationally. A September 2013 survey of Ukrainian public opinion by the International Republican Institute found that 17% of individuals in Ukraine identified as ethnically Russian. In addition to identifying as ethnically Russian, 40% of the population said they spoke primarily Russian in the home (IRI, 2013). Because accounts by various scholars indicate a quasi-East-West language divide in Ukraine between Ukrainian and Russian, it is pertinent to understand how this divide carries over into online sentiment for a European leaning movement. Studies following Euromaidan further indicate a move towards national identity consolidation (Kulyk, 2016).

Bruce Etling (2014) also conducted a similar survey of sentiment in Ukraine in 2014, while his analysis focused the sourcing of the materials being posted rather than the language or geographic regions in which they were posted. His work provides valuable insights into the other attributes of the social and political dynamics transpiring online during the Euromaidan movement. What differentiates Etling’s analysis from the present analysis is the logic of the support for the movement as well as being partially rooted in linguistic identification. His analysis is highly complementary to the investigation of Hypothesis 1.

It should further be noted that most Ukrainians are bilingual. Yet, despite being bilingual their willingness to engage online establishes a linguistic choice that possibly frames the way in which they interpret events. While it is possible that a native Russian speaker could post supportive messages in Ukrainian and, therefore, diminish the fidelity of this hypothesis, such shifts in posting behavior across languages should be visible in diminished Russian content generation in Eastern oblasts and disproportionally higher volumes of posts in Ukrainian in Eastern oblasts. Neither of these features was visible in the data collected. Yet, the point remains, in a bilingual country, assessing post sentiment and volume by language suffers from a potential weakness in that individuals may switch between languages.⁶

Hypothesis 2: Individuals generating content in Western Ukraine were likely to be significantly more supportive of the Euromaidan movement than their counterparts in Eastern Ukraine. Hypothesis 2 further disaggregates the West-East language divide and emphasizes geographic content distribution and subsequent support for movement goals. This is intended as a measure of support for the cultural-ideological cleavage argument made in hypothesis 1 and highlights geographic cleavages.

Keck and Sikkink (1998) develop the argument that often informal transnational advocacy networks will develop typically along horizontal patterns of communications that facilitate non-hierarchical information transmission in situations where the rapid dissemination of information to a broader community of interest is important. They further find that due to governments being the primary violators of rights, such transnational networks are leveraged to establish international pressure on domestic rights violators (governments). In Ukraine the use of Berkut, Ukraine’s internal security service special police forces, which are highly militarized and the use of digital surveillance and media suppression tactics, increasingly violated the human rights of Ukrainian protestors. As a result of these rights violations, Ukrainians turned to networks outside of Ukraine. To reach beyond Ukraine individuals began posting substantial volumes of content in English.

Hypothesis 3: Content generated in English that was geographically located in Ukraine was likely to predate and result in significantly increased content development within other countries capable of exerting geopolitical influence on Ukraine. Hypothesis 3 relates to Gamson and Wolfield’s (1993) argument examined in the literature review and focuses on the elicitation of non-indigenous support and the control of a narrative. It is also illustrative of Keck and Sikkink’s (1998) use of transnational networks to pressure domestic government engaged in rights violations. By generating content supportive of the movement in a non-native language, protestors are able to draw international attention to their repertoire of contention. It is expected that significant increases in internationally generated content will follow domestic content creation but that the level of support within this content will be significantly lower than indigenous support.

Lastly, what is the relative value or impact of social media on physical protest turnout or the impact of protests on social media? As highlighted above there are various scholars who have substantial reservations regarding the true value and impact of social media on social and political mobilization (Morozov, 2013; Shirky, 2011, December). It is clear from the spikes in social media around events that there is something going on but there remains a degree of uncertainty as to the directionality and importance of the relationship.

Hypothesis 4.1: Significant online content generation is likely to have predated all physical mass mobilization movements. Hypothesis 4.1 examines the ability of online content generation to affect physical world movements. This hypothesis attempts to determine directionality. If online content generation is of significant value, then calls to mobilize should precede mobilization. This is particularly true in Ukraine, where both government and opposition parties mobilize large numbers of individuals onto mass-transit systems to reach the capital for protests (Åslund, 2009). Other scholars within the broader literature on social movements and previous studies on Ukraine indicate social media as a means of reducing mobilization costs (Garrett, 1998). The literature suggests that lowered resource thresholds enable turnout. According to hypothesis 4 there should be indications of mobilization online prior to physical turnout in the streets. While many of the largest mobilizations occurred on weekends, not all weekends saw protest activity. If social media content generation does lead to physical mobilization there should still be significance in the relationship between content generation in advance of specific events.

Hypothesis 4.2: Physical protests are likely to have a significant impact on the generation of online content. While seemingly highly correlated with Hypothesis 4.1, 4.2 is the inverse of the relationship and proposes that the driver of social media is not the act of mobilizing for action, but rather a response to mobilizations that have already transpired. Hypotheses 4.2 does not necessarily diminish the value of social media and mobilization, it is simply a result of social mobilization not the instigator of it.

Methods

To understand cyber and physical mobilization and the relationship between the two – this study analyzed 10,254,266 pieces of which 5,780,365 pieces of content generated in English, Russian and Ukrainian were deemed directly relevant to the analysis. The primary keyword used for searching all content was Euromaidan (English), Євромайдан (Ukrainian), Евромайдан (Russian), respectively. This keyword within a Boolean structure was the only mandatory term in the search query in each of the three collections of content. All keywords were translated for each of the three languages, facilitating uniform assessment of online mobilization consistently across languages and attributes of the movement: Euromaidan, #Euromaidan, nazi, Fascist, America, West, Freedom, Democracy, Hope, Peace, Russia, Poland, Europe, Germany, invasion, crimea, kyiv, kiev, protest, Ukraine, United States, foreign, yanukovych, pravy sektor, #prayforukraine, #pray4ukraine, ukraineukraine. The keyword set was established through observational analysis of online conversations amongst Ukrainian nationals and directed face-to-face and phone conversations with Ukrainian nationals from various educational, economic, and geographic locations. Initial keywords were identified with assistance from host country nationals in Kharkiv, Kyiv, Dnepropetrovsk, and Lviv Oblasts, National Democratic Institute and International Republican Institute host country nationals in the Kyiv, Kharkiv, and Odessa. Further linguistic analysis was used by creating word clouds of the most common words linked to the Euromaidan movement in each of the languages. Thus, the informed words from individuals were fed into Crimson Hexagon and used to derive common word groupings in the forms of world clouds. These word clouds were generated through the automated linking of content in posts and locations over the duration of the protest. Although the initial word set was quite substantial, the scope of the final words chosen was necessarily limited and designed to provide a focused effort centered on specific terms of relevance to the Euromaidan revolution that encompasses both support for and against the movement.

These keywords were used to collect all public source content on Twitter, Facebook, hundreds of blogs, forums, news sites, reviews, YouTube descriptions and comments, Google Plus, and Tumblr. All content was posted publicly and was not protected by privacy settings or passwords or other settings meant to constrain the public nature of the content gathered. Data were collected from November 21, 2013 through March 1, 2014. In total, 5,780,365 unique and relevant pieces of content globally were collected and used for this analysis. Within the global sample, there were 2,809,476 unique pieces of content with geographically identifiable location in Ukraine based either on geotagged posts or country of origin noted within an individual’s social media or content account biography.⁷ To collect data, this work leveraged the Crimson Hexagon platform. Crimson Hexagon is a proprietary company founded out of research conducted by members of Harvard’s Institute for Quantitative Social Science.⁸

Upon completion of data collection, the BrightView assisted machine learning algorithm, a feature within Crimson Hexagon, was employed by a two-person multi-lingual team to assign content to discrete categories of Positive, Negative, and Neutral. The two-person team was fluent in Ukrainian, Russian and English and worked together to assign more than 40 pieces of content per category per language. BrightView is predicated on a non-parametric content analysis algorithm designed by Daniel Hopkins and Gary King (2010). Crimson Hexagon’s control tests indicate 92% fidelity with human coding within large scale data allocations.⁹ The study did not rely solely on Crimson Hexagon’s own data validation metrics. In addition to Crimson Hexagon’s metrics for analysis, fluent Ukrainian, Russian, and English speakers provided sample validation indicating approximately 95% fidelity with content allocation within the discrete categories.

In tandem with the online dataset the analysis leverages turnout numbers by day for the duration of the Euromaidan revolution derived from the Global Events Language and Tone Dataset Global Knowledge Graph and Events Dataset. GDELT data is notoriously noisy, with large volumes of superfluous information often included. The noise within GDELT data is due to its automated content collection and analysis engine that collects content from the world’s news media in over 100 languages (Leetaru & Schrodt, 2013). Protest counts for GDELT are determined by protest mentions included within news sources. Because there are multiple counts per day this analysis segments the high, low, and mean values of protest turnout by day were used as a rough approximation of physical protest movement mobilization.

To examine hypotheses 1 and 2 on language and regional divergences in sentiment towards the protest movements, language groups and regional groups were compared using repeated analysis of variance (ANOVA). Repeated ANOVA allows for a within subject repeated measure design (Girden, 1992) . Hypothesis 1 focused content in Russian, English, and Ukrainian generated within Ukraine. Hypothesis 2 focused on geographic regions delineated East and West. The variances in positive opinion by language and geography were each examined independently over the duration of the social movement from 21 November until 1 March for a total of 101 days.

To examine hypothesis 3 on the generation of English content to facilitate transnational network responses, lagged generalized ordinary least squares (OLS), autoregressive moving average (ARIMA) and Granger Causality models were employed. OLS and ARIMA provide the most theoretically applicable measures of impact and allow for reasonable forecasting of content generation in Ukraine relative to content generation in foreign nations. By contrast, Granger causality indicates whether previous values of English content generated in Ukraine led to statistically significant changes in English content generated external to Ukraine (Freeman, 1983).

The addition of ARIMA and Granger causality models further adds robustness to claims of impact and strengthen the argument within hypothesis 3. The resultant model meets all of the requirements of the Gauss-Markov assumptions (Pickup, 2015). English content geographically generated in Ukraine was compared to content generated in the United States, the United Kingdom, Russia, Spain, Canada, Sweden, Turkey, Ireland, Denmark, Poland, Italy, Australia, France and Germany using a t-1 lag. This paper assumes a unidirectional linear relationship between social media posts in Ukraine in English and posts in English in other countries around the world. The logic of this argument as highlighted in the hypothesis above is that events happening in Ukraine must first be transmitted to foreign audiences. While reverse directionality is possible and there is likely some endogeneity between posts generated external to and internal to Ukraine particularly in the later stages of the protest movement, this endogeneity was found to be minimal because events transpired in Ukraine first and actors in Ukraine had proximate access to information to be disseminated prior to foreigners. This argument at its most basic is both geospatial and temporal.

Hypotheses 4.1 and 4.2 focusing on the importance and directionality of social media in physical protest turnout and vice-a-versa was examined using an autoregressive moving average (ARIMA) model to account for minor autocorrelation and the dependent variables were squared to eliminate minor trending and create stationarity. Additionally, Granger Causality tests were run to establish a foundation for causality. These tests do not however indicate the impact of posts on protest turnout, merely that they had a significant impact. The relationship between posts and turnout was examined bi-directionally over the duration of the 101-day protest movement with limited missing data resulting in between 97 and 95 observations. Relationships between social media posting and physical turnout to protest were examined within with a lag of t-1 in an attempt to isolate the relationship between the two variables. Lagged times beyond t-1 were also examined, but no significance was found, and the results are therefore omitted from the below analysis. Null lags are rejected due to the data being collected and analyzed by day rather than within day. Hypothesis 4 is a continuous model over the duration of the movement. The relationship between posting time and turnout will be discussed in the results section.

Results

Assessing the linguistic fractionalization of Ukrainian social mobilization

When cross-group variance in positive opinion is examined within the volume of posts over a 101-day period using Ukrainian territorial geographic parameters the results indicate there is significant variance between positive content being posted between the Ukrainian, Russian, and English languages. This finding confirms hypothesis 1 and supports scholarship highlighting linguistic fractionalization within Ukraine and media reports on the divided nature of the conflict among those who primarily use Russian and those who primarily use Ukrainian. In Ukraine, there is a well-defined divergence in political preference and association demarcated along linguistic lines. This has borne out in voting returns over the last several election cycles, with members of the Party of Regions receiving substantially higher support from Russian speakers, whereas opposition parties have traditionally received higher levels of support from Ukrainian speakers. These results indicate that support for a movement overturning the political faction largely elected by Russian speakers receives substantially less support by individuals who posted in the Russian language. The results should not come as entirely surprising and are in line with anecdotal and historical studies on ethnolinguistic fractionalization in Ukraine (Barrington & Herron, 2004) .

Figure 1 illustrates the distribution percentages of positive content generated by language over the duration of the social movement. As is clearly evident the trend lines between the languages over time indicate substantially different levels of positive sentiment for the movement. While the Ukrainian trend line remains above 60% for the duration of the movement, the others only occasionally break 60% positive sentiment.

Figure 1. Positive sentiment distribution percentages over Euromaidan duration.

Figure 2 illustrates the variance between the linguistic groups. The within group variance is significant at an $\alpha =.05$

Figure 2. Repeated ANOVA of positive sentiment across languages.

To test hypothesis 2 and to further isolate the cleavages in Ukraine and support for the Euromaidan movement, regional distributions of sentiment in each of the three major languages were examined. Following distributions by sentiment, the variance in positive sentiment by region was also examined. Kharkiv, Luhans’k, and Donets’k Oblasts constitute eastern oblasts, while L’viv, Ivano-Frankivs’k and Ternopil constitute the western oblasts within the East versus West analytical framework. These six oblasts were chosen to provide maximum values of divergence and are deliberately not representative of the gradation between oblasts within geographic proximity. A future analysis that specifically emphasizes the nature of cultural ethnic cleavages rather than the impact of social media on protests would find benefit from a more nuanced and directed analysis within and across oblasts.

Within the limits of the initial study parameters, these data indicate that the variance between positive support of the Euromaidan movements is significant at $\alpha =.05$. While statistically this difference is significant, the effects of this difference are less pronounced than the linguistic differences between the Ukrainian, Russian and English languages. Both hypotheses 1 and 2 tend to support the linguistic cleavage theories of social mobilization. These results differ somewhat from on the ground analyses that indicated more uniform support across the country.

The breadth of the movement as assessed through sentiment analysis over the duration of the Euromaidan mobilizations has statistically significant variations in support by both regional and linguistic parameters. These indicators demonstrate non-uniform support within domestic populations for the social movement that break down on linguistic and regional lines. These fault lines could indicate challenges in national mobilization for the social movement and potential problems that might arise and did arise after its success. The results challenge the findings of John Alterman (2011) in his analysis of the Arab spring. While he noted that in those social movements social media facilitated identity development within social media, these results indicate that the facilitation of identity might be more nuanced and foster divergent identity groups or ethno-linguistic lock-in in relation to the movement. These hypotheses illustrate a fundamental challenge facing proponents of social media as a means of social mobilization in societies with strong ethnic, regional, or language cleavages.

Engaging in international repertoires of contention

To test calls for international support as a function of the repertoire of contention used by individuals participating in the Euromaidan social movement, 961,494 pieces of content written in English and geographically locatable in Ukraine and 380,495 posts in English geographically locatable in the United States, the United Kingdom, Russia, Spain, Canada, Sweden, Turkey, Ireland, Denmark, Poland, Italy, Australia, France and Germany were examined.

An OLS model was used to identify the relationship between content generated in Ukraine in English on future content generation external to Ukraine. A time series model was not used as the specification meets all the OLS assumptions as identified above (Fox, 2016; Pickup, 2015) . However, to further refine the analysis both ARIMA and Granger Causality Tests were used as well. All three models demonstrated statistical significance. Despite a continuous protest movement, I found no indications of significant serial correlation between social media posts on the movement by day or internationally by day. Tests were conducted with one day lags to isolate the impact of posts and protests across days in the relationship between content generated within Ukraine and that generated external to it.

Both OLS and ARIMA results indicate that domestic content generation in English significantly impacts foreign content generation in English at $\alpha =.01$, both within and across days. The models indicate minor differences in the impact of post volumes generated in Ukraine and those generated external to it. For the OLS model a one post increase in domestic content generation results in a .27 post increase in foreign content relevant to Euromaidan across the sample population between days. By contrast in the ARIMA model a one post increase in domestic content generation results in a .78 post increase in foreign content relevant to Euromaidan across the sample population between days. The results are displayed in Table 1 below. As a final check on the impact of post content generation in Ukraine impact on foreign content generation a Granger Causality Test was utilized and indicated significance at an $\alpha =.05$. Highlighting that content posted in Ukraine in English Granger causes posts in the foreign nations specified. As domestic actors generate additional content in non-local languages foreign actors generate more content as well. While Granger Causality tests indicate significant causal relationships, OLS and ARIMA add further analytical insight and highlight the effect of English language content generated in Ukraine on English language content generated external to it.

Table 1. Predicting Ukrainian to global content diffusion.

	Model 1 (OLS)	Model 2 (ARIMA)
Global Posts
UA Posts	0.277 ***	0.779 ***
	(0.029)	(0.053)
N	101	101

Coefficients; Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01

These results indicate a linkage between domestic and foreign repertoires and perceptions of a social movement. Moreover, posting in a foreign language is unlikely, based on social mobilization theory, to dramatically impact domestic populations, suggesting that posts generated internally are direct calls for support and empathy from the wider international community.

These data also begin to highlight what Brym et al. (2014) note in their commentary on social media during the Arab spring, that domestic posts can influence a foreign echo chamber that fuels a level of support often disproportionate to the support within a nation’s own borders. While international support is clearly important it can often skew the perspective of policy makers external to the nation experiencing social mobilization. Understanding the domestic social media scene is important for the identification of divergent positions within a given population relative to an ongoing social movement.

From cyberspace to independence square

To test hypothesis 4 on whether social media posts lead to more or less physical-in-the-streets turnout, GDELT Data for turnout within Kyiv was collected by day for the duration of the 101-day protest period. Outliers within the turnout data were discarded and the remaining data was averaged. The data comprised 93 days of observations and accounted for dropped outliers. The data were examined using ARIMA and Granger Causality models between days. The directionality of the relationship was also examined to determine whether social media predated physical turnout or vice-a-versa. The data indicate a significant impact of social media on physical turnout on the street across days both within the ARIMA and the Granger Causality models at an $\alpha =.01$. Table 2 shows the results of the ARIMA model and Table 3 shows results of Granger causality. What is clear from all four models is a single direction and significance in the casual relationship between social media content and physical mobilization.¹⁰ By contrast physical turnout did not significantly impact social media content generation at $\alpha =.01$ at time t + 1. The relationship between posts and physical turnout is not uniform and includes wide variances, but the general pattern of online engagement resulting in offline turnout holds. Within day analysis of physical mobilization and social media posting are not possible due to a single day data capture window preventing within day analysis.

Table 2. Predicting online mobilization to physical mobilization.

		Model 3 (ARIMA)			Model 4 (ARIMA)
Physical Turnout			Total Posts
	Total Posts	0.45 ***		Physical Turnout	0.07
		(0.137)			(0.008)
N		96			96

Coefficients; Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01

Table 3. Granger causality wald tests.

	DV	IV	Prob > F
Model 5	Physical Turnout	Total Posts	0.00 ***
Model 6	Total Posts	Physical Turnout	0.83

* p<0.10, ** p<0.05, *** p<0.01

The results indicate a strong relationship between social media posts and physical turnout in the streets. These results are strongly unidirectional and episodic. The generation of social media posts did not result in fully sustained turnout for the duration of the movement, although low-to-moderate levels of turnout remained for the duration of the movement. However, spikes in social media posts in the day prior to large turnout in the streets is statistically significant and offers insight into the mobilizational power of social media to shift protests from cyberspace to physical space.

Discussion

The influence of social media in modern protest movements remains a perplexing issue. This work leveraged data relevant to the Ukraine protest movement that resulted in the overthrow of the government. By examining the basis of support within linguistic and regional (geographic) groupings this work illustrates that social media analysis can facilitate a more nuanced understanding of how and where support for social movements in diverse societies arise and are sustained. It also indicates that there is ample evidence that societies similar to Ukraine are not monolithic and often have divergent groups each with differing levels of support for social movements. To develop social movements that facilitate national change through robust repertoires of contention, actors can and should likely leverage data streams from across their country to understand how to involve and develop communities to foster engagement. This is likely to be impactful for a variety of analyses in the future beyond social and political movements and might indicate ways in which to assess support for political parties, insurgent groups or other actor categories.

The data indicate several attributes about Ukraine that might surprise individuals not intimately familiar with its internal complexities. First, although there are different patterns of support for the repertoires of contention associated with the 2013–2014 Euromaidan movement the support diverges primarily along linguistic cleavages and regional cleavages as illustrated by Figures 3 and 4 as represented by social media posting patterns. It should also be noted that while the regional cleavage is not as pronounced, there were significant divergences in posting volume between eastern and western oblasts indicating an imbalance in proportionate support between regions in Ukraine. Western Oblasts produced approximately 100% more social media content over the duration of the protest movement. Generally, however, the data seem to anecdotally indicate a more important role for linguistic cleavages than for geographic ones within Ukraine. This mechanism is highlighted by very strong support for the moment among Ukrainian speakers and very weak support upon Russian speakers.

Figure 3. Posting sentiment distributions by language over the duration of Euromaidan.

Figure 4. Posting sentiment distributions by region over the duration of Euromaidan.

The literature broadly, as addressed above, is in conflict as to the value that social media has on a given social movement. While the data in this analysis are limited, they indicate that the relationship is significant. The unidirectional significance of the online posts being followed by substantial increases in physical turnout for protests appears to validate many of the narratives of the Arab spring. Social media clearly has an impact on physical mobilization that is significant and unidirectional. The Granger causality models indicate significance in the relationship and the ARIMA model allows within a wide margin of error allows for the forecasting of actual physical protest turnout.

These results are largely consistent with Onuch (2015), but also highlight the need for combined analyses that rely not only on social media analysis, but on the ground surveys. Together they highlight the complex dynamics of protest movements. It is further possible that as Internet penetration increases in countries like Ukraine that it that the mobilizational power of social media will grow and further add resources to mobilizations of civilian populations. Future studies on Ukraine will be further helped by laws and regulations put forth by President Poroshenko that limit or restrict Russian social media networks, news and websites. These policy changes have resulted in substantial increases in online participation Western social networks, the same networks from which this analysis drew data.

Conclusion

This work examined the impact of social media in social and political movement known as the Revolution of Dignity in Ukraine (Euromaidan). The literature in this field is robust and the analyses presented above provide evidence to support case studies on the impact of social media in movements during the Arab Spring. Although the analyses contained above help to further identify the linguistic, and geographic divides in supporting the revolution of dignity, it does so via a broad lens that might obscure some of the nuance occurring within various regions. It does, however, indicate that in Ukraine language and geography are important and carry over into online discourse. This is likely not a surprise to scholars of Ukrainian politics or history, however, the value imparted on such analysis should be understood as a single perspective and does not obviate the importance of more nuanced analysis. Moreover, the data highlight the immense value of social media in alerting the international community to the event occurring domestically. The relationship between content generated domestically and subsequent international content generation is a valuable attribute of social mobilization. The biggest takeaway from this work is that Social media, based on the events in Ukraine, can and has had a significant impact in facilitating physical mobilization. If this is true in Ukraine, it is potentially true elsewhere and supports many of the anecdotal claims made during the Arab Spring and elsewhere. Combined the analyses above highlight the importance of ICTs in understanding the attributes of social or political movements.

Despite the robustness of the findings listed above, this is only one case in a broader universe of cases and is constrained by national internet penetration and use by certain groups as well as within day posting metrics and physical turnout by hour. The value of ICTs resides not solely in their ability to provide a societal release valve for grievances, but also in their ability to facilitate physical mobilization for social and political change. The data to assess the impact of ICTs on social and political mobilization are likely to improve in the coming years and will provide more opportunities to assess with increasing rigor the impact these technologies have on mobilization across movements within countries.

Additional information

Funding

This work was supported by the Deparment of Defense OSD Minerva R-DEF.

Notes on contributors

Aaron Franklin Brantly

Aaron Franklin Brantly is Assistant Professor in the Department of Political Science and Hume Center for National Security and Technology Affiliated Faculty at Virginia Tech, Cyber Policy Fellow at the Army Cyber Institute at West Point. He holds a Ph.D. in Political Science from the University of Georgia and a Master's of Public Policy from American University. His research focuses on national security policy issues in cyberspace including big data, terrorism, intelligence, decision-making and human rights. His books include: Cybersecurity: Politics, Governance and Conflict and Cyberspace [2019 Polity],The Decision to Attack: Military and Intelligence Cyber Decision-Making (2016 UGA Press)and US National Cybersecurity: International Politics, Concepts and Organization (2017 Routledge).

Notes

1. Transparency International (https://www.transparency.org/cpi2013/results).

2. The World Bank (http://www.worldbank.org/en/country/ukraine/overview).

3. The World Factbook 2013–14. Washington, DC: Central Intelligence Agency, 2013. (https://www.cia.gov/library/publications/the-world-factbook/index.html).

4. Internet World Stats (http://www.internetworldstats.com/stats4.htm).

5. Geographically locatable content requires a user to allow platforms to know their geospatial location. While more than 10 million pieces of content were generated within the keyword parameters only 2 million had specific geotags indicating their origin point as Ukraine. Collecting geotagged content substantially reduces the overall volume of content available but increases the geographic fidelity.

6. Thank you to Review #1 for raising this point.

7. Post counts disaggregated by sentiment type, language and region can be found at: https://drive.google.com/open?id=0Bx_uTR8BHv0SWGdiSzM5bldpV2M .

8. http://www.crimsonhexagon.com/about/.

9. http://www.crimsonhexagon.com/video-flash/brightview-algorithm-intro/story_html5.html .

10. Special thanks to Reviewer #2 on proposing Granger Causality tests. This suggestion resulted in the dropping of outlier numbers and the re-examination of the data.