Us vs. Them as Structural Equivalence: Analysing Nationalist Discourse Networks in the Georgian Print Media

Nationalist discourse has been identified as a driving factor in the causal chain linking democratization to the likelihood of ethnic conflicts, due to its nature of polarizing us against others along ethnic lines. However, we lack systematic knowledge of the structure and dynamics of this polarization. Adding to the established practices of analysing in-group/out-group divisions in ideological and political discourses, this article proposes an innovative way of measuring the divisiveness of nationalist discourse using social network analysis. Instead of looking at direct nationalist interaction between actors, deeper discursive structures are found by analysing indirect relationships across actors, based on their nationalist interaction with third parties. In this manner, it is possible to identify whether certain actors form structurally similar clusters, based on whom they direct their nationalist appeals to, how intense these appeals are towards specific actors, and what other groups they are targeted by themselves. By applying the measure of Structural Equivalence to the original data on nationalist appeals obtained from the quantitative content analysis of the Georgian print media across the 20 years of its democratization (1991–2012), this article shows that the actor structure of nationalist discourse conveys information on group polarization. Further, it demonstrates that the divisiveness of this discourse in Georgia became sharper during electoral periods, with the first two elections after independence being particularly dangerous in this regard.


Introduction
Conflict researchers, who have found evidence that electoral periods in democratizing countries increase the risks of violent conflicts, identify nationalist discourse in media as a major mechanism driving the causal process between the two variables (Cederman, Gleditsch, & Hug, 2013;Mansfield & Snyder, 2005;Snyder, 2000). According to this argument, relative liberalization of the public sphere, and an increasing degree of political participation in post-authoritarian states, incentivize both old and new actors to mobilize resources-including their control of media-to win mass support. Carrying the legacies of authoritarian political culture, these actors do not have strong policy platforms they can rely on. Therefore, nationalism becomes the "universal" cate-gory to appeal to (Snyder & Ballentine, 1996). But, apart from creating feelings of cohesion within the members of a national in-group, appealing to national sentiments also risks creating aversion towards the national or ethnic 'others.' Whether targeting groups within or beyond state borders, these appeals are particularly dangerous if saturated with hostile references. It is in these circumstances that nationalist polarization between us and them occurs, which can contribute to conflict escalation under the conditions of weak political institutions in democratizing countries (Cederman, Hug, & Wenger, 2008;Mansfield & Snyder, 1995).
Despite focusing extensively on the importance of nationalist discourse, this strand of the literature offers little systematic knowledge about its structure and dynamics. As a result, we also lack empirical evidence suggest-ing the intensification of divisive nationalist discourse during electoral periods. Based on an analysis of the print media of Georgia-a typical case of democratization and conflict-this article addresses the problem by identifying empirically if, how and when the nationalist discourse in media is at its most divisive.
To this end, taking the structural perspective and extending on the methodology of discourse network analysis (Leifeld & Haunss, 2012), nationalist discourse is conceived in this article as a social network wherein actors are engaged in nationalist interaction, i.e., sending, and targeted by, nationalist appeals. This proposition, which relies on the approach to the study of nationalist discourse by Abzianidze (2020), implies an innovative way of conceptualizing and measuring the in-group/out-group structure of this discourse, based on the structural similarities and differences of the actors sending nationalist appeals to, and receiving them from, the same other actors. Rather than simply focusing on dyadic interaction between actors, my approach is to bring in third parties and look at the indirect relations between two actors involved in the nationalist discourse, based on the patterns of their interaction with these third parties. In network analytic terms, this implies using the method of structural equivalence analysis to tease out the hidden structures of nationalist discourse networks.

'Us' vs. 'Them' in Political and Ideological Discourses: Established Practices
The general nature of the intergroup relationship has been identified and largely discussed by the social psychological literature (e.g., Hogg & Abrams, 1998;Tajfel & Turner, 1986). While the latter is mostly focused on behavioural aspects of intergroup relations, Billig (2009) has emphasized the roles of text and talk in the reproduction of group identities. Critical discourse analysts have further studied patterns of ideological reproduction by embedding the social psychological understanding of intergroup relations in discourse analytic theories and methods (Fairclough, 1995;Wodak, de Cillia, Reisigl, & Liebhart, 2009). These studies mostly focus on, for example, references to pronouns, such as 'us,' 'them,' 'we,' 'they,' or to other deictic words such as 'here,' 'there,' etc. Also, by deconstructing a text content qualitatively, they try to link discourse structures to power structures ( van Dijk, 1993van Dijk, , 2012. While all of these studies acknowledge the importance of the actors, few of them, if any, analyse systematically the actor structure of the discourse, which leaves patterns of actor interaction in the nationalist discourse understudied. More recent studies of political discourses have addressed these challenges by extending the scope of the material analysed, both in terms of its amount and the information coded, thus allowing more formalized analyses over time. Particularly important in this regard is the research on political claims analysis (Koopmans & Statham, 1999), which combines the quantitative rigor of protest event analysis with the qualitative depth of political discourse analysis. However, the relational aspects between senders and addressees of claims, and the resulting discourse structures, have not been greatly explored, apart from work on the debates around EU enlargement by Adam (2007). To address the lack of attention to the relational properties of political discourses, Leifeld (2017) suggested the discourse network analysis methodology, which combines the structural depth of critical discourse analysis and a more formalized social network analysis. It employs the methodological toolbox for a systematic analysis of the discursive interaction of actors based on the concepts they refer to, and the stance they take towards issues in a policy domain. Thus, it allows the identification of the discourse coalitions in policy debates, and the analysis of patterns of discursive interaction within and between those coalitions.
While this strand of the literature has substantially advanced the systematic study of the relational structures of political discourses, its application to nationalist discourse and, especially, its divisiveness, can be limited. The reason for this is that policy discourse is a debate in its own right, wherein different actors might take different stances towards different issues. Yet nationalist discourse is more a discourse of hegemony rather than a debate. It is clear that, within this discourse, members of ethnic and/or national groups will always take the stance in favour of their own groups, whatever the issue concerned. Therefore, with the purpose of identifying the cleavage of the ethno-national 'self' against the 'other' in the discourse, this article demonstrates how the methodology of discourse network analysis can be extended from using it for studying policy debates to its utilization for ideological discourses such as nationalism.

New Prospects for Analysing Nationalist Discourse
In this study, nationalism is defined as a "doctrine that people who see themselves as distinct in their culture, history, institutions, or principles should rule themselves in a political system that expresses and protects those distinctive characteristics" (Snyder, 2000, p. 23). It is a way of speaking about a nation, its boundaries, interests, and aspirations (Brennan, 1990;Calhoun, 1997), thus, manifested in both text and talk (van Dijk, 1998, p. 193). Divisiveness is characteristic for the nationalist discourse, just as it is the case with other ideological discourses (e.g., populist discourse). The major source of this divisiveness is the emphasis on the distinction between the in-group (us) and the out-group (them). Typical to this emphasis is the expression of in-group affinity and distrust, or sometimes even hostility and hate, towards an out-group. The latter can frequently result in the expressed desire to exclude the members of this out-group from certain rights or resources. In this way, groups manage to reaffirm their in-group cohesiveness and thus ensure the constant re-production of their national identities. Although, frequently, it is not the open and con-frontational "flag-waving" version of nationalism, but rather the everyday "banal" form that constitutes the reproduction of national belonging (Billig, 1995). The banality of it lies in the fact that these expressions of nationalism mostly remain unnoticed-for example, a flag on a country's parliament building, which is rarely noticed by passer-by citizens. Discursive manifestations of this banal nationalism are not benign, either. The everyday usage of such banal words as 'us,' 'them,' 'we,' or 'they,' as well as that of deictic expressions, such as 'here, ' 'there,' or 'now,' can enhance the process of nationalist re-production (Billig, 1995, pp. 93-127). I argue that there is yet another, so far undiscovered, manifestation of banal nationalism hidden in the actor structure of the discourse.

Theoretical Logic of the Argument
This article argues that if members of an in-group make repetitive negative statements towards the same outgroup(s), so that these members are aware of each other's statements, it can strengthen the bonds within the in-group members vis-à-vis that/those out-group(s). Applying this logic to the nationalist discourse means that when different members of a nation or an ethnic group repeatedly make nationalist statements publicly, and direct those statements towards the same other groups, it can, on the one hand, intensify the feeling of in-group cohesion, and on the other hand, aggravate the aversion towards the respective out-groups. In times of nation-building, political and social actors who have access to the public sphere, and who need to legitimize their position as part of the elite, respond to the group prototypicality demands by setting and/or enhancing the boundaries of their in-group and delimitating it from everybody else (van Knippenberg, 2011). In this process, some out-groups might be addressed in a neutral way, while others can be represented as existential threats to the group self-continuity, and thus be portrayed as enemies (Sani, Herrera, & Bowe, 2009). When we talk about the re-production of the in-group cohesion through the nationalist discourse, the exact content of their nationalist appeals, and how actors address each other, is certainly of great importance. However, by focusing on dyadic interactions between actors, we might be missing deeper structures of the discourse, which can have a strong impact on in-group cohesion.
Instead, I propose to understand and, thus, operationalize the structure of 'us' vs. 'them' in the nationalist discourse in terms of the groups of structurally equivalent actors. This implies shifting our attention from who attacks whom to what relationships actors have with each other, based on their interaction with other actors. More specifically, actors involved in the nationalist discourse might not have direct relationships with each other (i.e., sending statements to each other, in this case), but they can be linked indirectly through the patterns of their interaction with third parties. These indi-rect relations can be defined by the similarity (or dissimilarity) of actors in whom they target their nationalist appeals at, in how intensive their appeals are towards specific addressees, as well as in being targeted by the same other actors.
Nationalist discourse in the present study is constituted of instances of appealing to exclusion or expressing hostility, whereby at least two different ethnic or national groups are concerned. Therefore, drawing on the work by Maoz, Kuperman, Terris and Talmud (2006), we can think of this similarity as the structural affinity among actors involved in the nationalist discourse-i.e., actors who are structurally similar in sending nationalist appeals to and receiving them from the same other actors form the group, members of which share the same (in this case, exclusionary/hostile) attitudes towards the groups that are structurally dissimilar from them. At the same time, if we have the information on the ethnic and/or national attribution of the actors to be analysed, we can also find out if the clustering patterns of these groups follow ethno-national lines and, thus, form ethnic/national in-groups against out-groups. In this way, we will be able to tease out the 'us' vs. 'them' structure of the nationalist discourse systematically.
The graphical representation of this argument can help us to understand its logic better. Figure 1. presents confrontational discursive interaction between actors representing three hypothetical groups labelled as A, B, and C. Members of group A (a 2 and a 11 ) are structurally similar because, on the one hand, they make confrontational statements against the members of the same other groups, in this case c 1 and b 2 , and, on the other hand, are themselves targeted by the same other actors. Actors c 1 and b 2 are structurally similar as well in that they share patters of interaction with other actors; however, they are dissimilar from a 2 and a 11 . In this way, the structurally similar actors a 2 -a 11 and c 1 -b 2 form two different clusters. If this type of confrontational interaction occurs repeatedly in a public arena, it can strengthen the bonds between a 2 and a 11 , as well as between c 1 and b 2 . Even more importantly, it can exacerbate the polarization between these two clusters (i.e., a 2 -a 11 and c 1 -b 2 ).
It is then the goal of the empirical analysis to identify whether the data on nationalist appeals collected from the Georgian print media in 1991-2012 yield the groups of structurally similar and dissimilar actors, and, if this is the case, what lines these group divisions follow-are the clusters of structurally similar actors in the discourse divided along ethno-national lines?

Nationalist Polarization between 'Us' vs. 'Them' and Elections
Embedding these theoretical propositions back to the election-nationalism-conflict nexus is instrumental to improve our understanding of the relationship between elections and the divisiveness of nationalist discourse. Both large-N studies, as well as case studies which find Figure 1. Network of confrontational discourse among the hypothetical groups, a graphical representation of the argument. Notes: Capital letters stand as labels for hypothetical groups, while the respective lowercase letters represent members of these groups.
that electoral periods are associated with the increased likelihood of violent conflicts, assume that nationalist discourse intensifies during these times (Cederman et al., 2013;Mansfield & Snyder, 2005;Snyder, 2000). Their theoretical argument builds on the peculiarities of political competition in newly emerged states that have embarked on the path of democratic transition. Unlike in mature democratic countries, where losing elections might merely mean waiting for another try in the next round of elections, here the costs can be much higher for politicians. Losers might face not only marginalization from political power but also oppression. Conversely, given the level of centralization in these countries, winning elections usually implies obtaining near unlimited access not only to political power, but also to the economic resources and coercive forces of a country. These high election stakes and the resulting potential threats can push actors involved in the political competition on either side to instigate violence (Höglund, Jarstad, & Kovacs, 2009;Wilkinson, 2004). During electoral periods, political and military actors frequently resort to confrontational, exclusionary, and divisive nationalist rhetoric (Mansfield & Snyder, 2009). Given these propositions, I expect that the 'us' and 'them' polarization in the nationalist discourse present in the Georgian print media of 1991-2012 will be stronger during electoral periods as compared to non-electoral years. In addition, the first two competitive democratic elections have been identified as the most dangerous, in terms of the likelihood of violent conflicts (Cederman et al., 2013). Actors are supposed to be the most radical during these times due to the uncertainty of the situation (Brancati & Snyder, 2013;Reilly, 2002). These propositions lead me to expect that the divisiveness of the nationalist discourse between the in-group and the out-group should be stronger during the elections at the beginning of the democratization period (i.e., early elections), compared to those taking place later in the period (i.e., late elections).

The Story of 'Us' vs. 'Them' in Georgia
Georgia declared independence in 1991 after nearly 70 years of being part of the Soviet Union. Before then, Georgia experienced only three years of being an independent republic, between 1918 and 1921. Therefore, when, after the dissolution of the Soviet Union, Georgia became independent and embarked on the path of democratization, questions were wide open regarding what constituted the nation, who was part of it and who not, and, maybe most importantly, how the ethnic understanding of 'Georgianness' could be coupled with the multi-ethnic reality of the Georgian state. Elites, who had the access to the public sphere, faced the task of defining the boundaries of the Georgian nation and that of delimitating it from everybody else (all the out-groups), thereby identifying who posed a threat to the well-being of the in-group and which other groups could be perceived as friends. This process did not start with the declaration of independence, but instead had begun more than a decade before when the Georgian National Independence movement gained momentum. Already in the 1970s, a group of literati from academia and the cultural intelligentsia launched an intellectual effort to define 'the Georgians' by recalling the glorious past of the nation, identifying its 'spiritual missions,' as well as laying out its political interest of self-determination and ways in which it could develop into a future democratic state. Ethnicity was the central defining category here. Obviously, part of the process was to label other ethnic groups and nations in relation to the in-group. The number one enemy of the nation at that time was clear (the Soviet Union), and independence was equated with the existence of the nation. These ideas, gradually trickling down to the general public, gained massive importance by the end of 1980s, with the relative opening of the public sphere in the perestroika era.
The more realistic the prospect of the dissolution of the Soviet bloc became, the more the range of outgroups was diversified. First, as the legal successor of the Soviet Union and the one holding back Georgian independence, sometimes even with the use of the force, Russia was evolving into the enemy, as it was the major threat to the self-continuity of the in-group. Second, emerging tensions with the ethnic groups residing in the autonomous regions within the Soviet Socialist Republic of Georgia and demanding independence from the Georgian state, created another stratum of out-groups-Abkhazians and South Ossetians-who threatened 'our' political interests, inasmuch as 'their' demands for independence intimidated the in-group's understanding of its territorial integrity. Related to this, Armenian and Azerbaijanian ethnic minorities residing in Georgia also began to be perceived as a potential threat, in part due to the fear of a potential repetition of the Abkhazian and South Ossetian scenarios. Several factors might explain why these fears emerged towards these and no other minorities living in the country: These two groups were (and still are) the largest ethnic minorities, territorially concentrated in two different regions of Georgia; they both were linguistically and religiously different from Georgians; and, more importantly, both of these groups had kin-states that, similar to Georgia, at the time were undergoing the process of nation-building as well as political selfdetermination, which implied defining national and territorial borders in relation to 'others.' Third, already in its initial phase, the Georgian National Independence movement had defined the new orientation of the emerging state as Western democratic; therefore, the hope was that the countries of Europe and the USA would stand by the in-group in its fight for independence and a democratic future. However, an emphasis on the uniqueness and particularity of Georgian culture and traditions, and its difference from Western cultures (and others), were not uncommon either. Last but not least, becoming an independent state also implied the definition of the in-group´s relations with all the other states on the international arena.
It must be taken into consideration that the process of nation-building in Georgia was accompanied with complex political dynamics during the period under study . Right after the dissolution of the Soviet Union, three conflicts erupted almost simultaneously-a governmental conflict that led to a coup d'état (December 1991-January 1992, and two ethno-territorial wars in the autonomous regions of Abkhazia andSouth Ossetia (1990-1993). After nearly a decade of 'frozen' conflicts, 2008 marked the outbreak of a new wave of violence. While Russian interests and its involvement in the wars of the early 1990s was obvious, the war of 2008 turned into an explicit interstate armed conflict between Georgia and Russia. The so called August War in 2008 resulted in the recognition of Abkhazia and South Ossetia as independent states by Russia.
In light with these tensions within and beyond state borders, appealing to nationalist sentiments has not been unknown for political contestation in Georgia. The first president of Georgia, Zviad Gamsakhurdia, for example, frequently referred to the idealization of past glories of the Georgian nation (Jones, 2013, pp. 51-74). As Wheatley (2005, pp. 51-63) argues, Gamsakhurdia was able to exploit the existing fears of Georgian citizens by portraying ethnic minorities as a 'fifth column' acting in Russia's interests. Labelling political opposition, or any other opponent, as an 'agent of KGB,' 'the enemy of the nation,' 'the traitor,' etc., was also not uncommon in Gamsakhurdia's speeches. Some authors also attribute the escalation of violence in the region of Abkhazia to the local power struggles within the State Council of Georgia (the former Military Council) between the warlords and Eduard Shevardnadze, the Head of the Council at that time and future President of the country (Jones, 2013, p. 95;Nodia, 1998, p. 34;Wheatley, 2005, p. 70). This pattern of utilizing nationalist discourse in political competition has persisted even after the Rose Revolution, which brought to power young reformers led by President Saakashvili. Nationalism under Saakashvili's government showed some diversionary tendencies, i.e., constantly emphasizing the threat Georgia faced from Russia, and frequently using these threats to de-legitimize opposition within the country. Perhaps, the most vivid example of this was a fake news report about the start of the war that aired on the TV channel, Imedi, during primetime on March 13 th , 2010. It was widely believed that Imedi was controlled by the government, partly because of its openly pro-government bias, and due to the fact that the broadcaster was run by Giorgi Arveladze, a long-time ally of Saakashvili and a former member of his party and cabinet. The 30-minute fake news report suggested that Russian troops were invading the capital of Georgia, and went so far as to report the assassination of President Saakashvili, all of which sparked mass panic across the country. One of the main messages transmitted by the report was that the leaders of the Georgian political opposition were supporting and legitimizing the Russian intervention. After the report, the television anchor said that its aim was to show what could happen, subtly pointing to the threats opposition posed to the country ("Fake report on renewed war," 2010; Mtivlishvili, 2010).
Bearing in mind this background information about the nation-building process and political struggles that took place in Georgia across the period of its democratization between 1991-2012, the major goal of the study is to identify whether and how this nationalist divisiveness was reflected in the structures of the discourse.

Data and Method
The analysis in this article relies entirely on the original data on nationalist appeals collected during an extensive content analysis of six Georgian newspapers between 1991 and 2012. Information regarding the newspaper selection, sampling, and coding procedures is provided in the Supplementary File. The data consists of 1,186 newspaper articles, out of which 809 were identified as being related to the topics of interest (ethnicity, ethnic or governmental conflicts, and political institutions) and, therefore, further analysed at appeal level. Out of the 4,541 appeals identified in these articles, 13.4% were coded as nationalist. Statements were coded as nationalist if they referred to the exclusion from certain political/civil rights of other actors, or expressed hostile attitudes towards them, and if at least two ethnic/national groups were involved. Therefore, the network of the nationalist discourse in this study by default represents a negative and confrontational interaction between actors.
Given the conceptualization of nationalist discourse in this study, the latter is conceived as a set of elements including: the actual content of appeals; the constellation of actors around these appeals (represented as senders or addressees); and patterns of interaction among actors, as well as between actors and the content of the appeals. Hence, the nationalist discourse in this study is represented as a (discourse) network, wherein actors are engaged in nationalist interaction among each other. As this relational nature of the data requires a method that has the properties of analysing structural aspects, social network analysis is considered to be the most effective approach to be used here. According to the theoretical propositions of the study, the major goal is to identify if certain actors involved in the discourse cluster in groups based on their structural similarities, and whether the boundaries of these clusters follow ethno-national lines. It is the claim of this study that in this way we can identify and measure the divisiveness of nationalist discourse. Structural equivalence analysis is utilized exactly for this reason: to detect the extent of structural similarities between the actors and to map their clustering patterns.
Studying the divisiveness of nationalist discourse using structural equivalence analysis, and social network analysis in general, is innovative because it has never been done before. Applying social network analysis methods for analysing policy discourses has already been proven to be effective (Adam, 2007;Fisher, Waggle, & Leifeld, 2013). Some authors have also demonstrated the usefulness of structural equivalence analysis specifically for studying polarization patterns in policy debates (Fisher, Leifeld, & Iwaki, 2013). This study shows how the method of structural equivalence analysis can be utilized to tease out the hidden structures of nationalist dis-course more effectively, as compared to a mere count of word frequencies or qualitative accounts of small numbers of texts.
Nodes are defined as all the actors who are mentioned by the articles either as senders or as addressees of the nationalist discourse. For all the actors coded, the data also includes information on their attribution to ethnic groups or nationality. This article is not interested in how much effect the nationalist appeals of a certain set of actors have on nationalist mobilization, as compared to others. Rather, it is primarily interested in a general nationalist discourse that is present in a public sphere. Therefore, while theoretically the weights of nationalist appeals might vary in terms of their capacity for mobilization, depending on who makes an appeal (Conversi, 1995;Hroch, 1985;Kedourie, 1993;Snyder, 2000), this study does not differentiate empirically between the leading ethnic entrepreneurs and actors on the fringes of a movement. It rather takes 'elite' as a cumulative unit and conceives the fact that an actor is given a voice in media as a proxy of being a potential agent of ethnic entrepreneurship.
The edges are defined as the number of statements with which actors target each other. Thus, the networks, and their respective adjacency matrices, in this analysis are directed and asymmetric. Edges represent not only the existence of a link but also its strength. Entries on a diagonal of the matrix are not meaningful since the occurrence of an actor sending a nationalist appeal to itself is not considered.
The goal of the analysis is to identify the clusters of actors that are related with each other indirectly, based on similarities in who they attack with their nationalist appeals, how intense these attacks are, as well as who they receive attacks from. In network analytic terms, this means that an actor i is similar to an actor j if both i and j are linked to an actor k. Actors i and j do not need to be directly related to each other. Using the method of structural equivalence analysis, we will be able to compare the interaction profiles of all the actor dyads involved in the discourse and thereby determine their structural similarity and dissimilarity (Borgatti, Everett, & Johnson, 2013;Lorrain & White, 1971). Interaction profile here refers to all the out-going, as well as in-coming, ties of an actor. Since the network is directed, the interaction profile of an actor i in the socio-matrix represents all the entries adjacent to this actor both in its respective row and the column. Structural similarity is measured using Pearson Correlation Coefficient. This choice over Jaccard Coefficient and Euclidean distance is guided by two factors. First, the data is not binary but count; and second, the study is primarily interested in the patterns of similarities rather than dissimilarities between the actor profiles.
The structural equivalence analysis proceeds in several successive steps. First, the profile analysis of each actor dyad is conducted by calculating the correlation coefficients between their respective rows and columns in the adjacency matrix X. Second, the resulting correlation matrix C is constructed in which every entry ij represents a correlation coefficient between the interaction profiles of actors i and j-thus, the extent to which these two actors are structurally equivalent. The correlation matrix C is undirected, i.e., symmetric, and therefore the correlation coefficient in the entry (i, j) is the same as that in the entry (j, i). The stronger the correlation between two actors, i.e., the closer the coefficient is to +1, the more structurally equivalent these actors are.
The correlation matrix does not allow the identification of patterns easily, so, in order to separate the groups of structurally similar actors from those who are dissimilar, the next step is to partition them into mutually exclusive clusters based on their degree of structural equivalence. Actors who are structurally more equivalent will be grouped together in a cluster and separated from other clusters comprised of the actors who are structurally less similar (Wasserman & Faust, 2009, p. 376). In order to account for the robustness of the clustering patterns of actors, the analysis in this section utilizes two alternative methods of partitioning-hierarchical cluster analysis and multidimensional scaling. The first method subdivides nodes into subsets of actors who are structurally equivalent at level . The procedure is called hierarchical because partitioning occurs at successively less restrictive values of (Wasserman & Faust, 2009, p. 381). The second method uses spatial technique to represent the structural similarity and dissimilarity of actors based on their spatial proximity in two-dimensional space (Wasserman & Faust, 2009, p. 387). Structurally more equivalent actors are located closer to each other, while less equivalent actors are placed further away, thus forming groups of those who are similar, and separating them from those who are different. This analysis is performed using the social network analysis software, UCINET. All the respective correlation matrices can be found in the Supplementary File.
Further, electoral periods are defined as the election date, three months before, and one month after it. Since the sample contains four months in each year, electoral periods are understood as electoral years, i.e., the years in which elections were held. Non-electoral years are defined as a random sample of four months in each year without elections. In addition, in line with Cederman et al. (2013), early elections are understood as the first two competitive national level elections, which, in the Georgian case, means the elections of 1991 and 1992. All national elections starting from 1995 were coded as late elections.

'Us' vs. 'Them' as Structural Equivalence in the Georgian Print Media
Does the partitioning of structurally equivalent actors involved in the nationalist discourse in the Georgian newspapers yield groups corresponding to the case-specific 'us' vs. 'them'?
The analysis starts by contrasting the clustering patterns of the structurally equivalent actors in the nationalist discourse with that of the non-nationalist discourse. The comparison presented in Figures 2 and 3. demonstrates an immediate difference between the two graphs. Actors involved in the nationalist discourse ( Figure 2) form two clusters clearly delimitated from each other at the level = 0.142.
Looking at the ethnic or national attribution of actors in these clusters, we can easily identify that the right side of the graph is completely homogenous, represented only by the Georgian actors, while the cluster on the left side is rather mixed, comprised of all the other ethnonational groups. The clustering pattern of the nationalist discourse in Figure 2 reflects several essential points: The media gives voice predominantly to the members of the in-group (i.e., Georgian actors) and, therefore, the latter are systematically similar in being the dominant senders of the nationalist appeals present in the media discourse; the members of the in-group are also systematically similar in directing nationalist statements to the same other groups; moreover, they are similar in how intense their nationalist statements are towards specific other actors.
As the values of become restrictive in Figure 2, the cluster of 'them' is being divided into subsequent sub-clusters. The West and international organizationsactors perceived by the Georgian state as potential allies and sources of material and non-material aidare grouped together ( = 0.396). Actors with whom Georgians have experienced ethno-political tensions and violent conflicts (i.e., Abkhazia, South Ossetia, Russia, the Soviet Union, and other ethnic minorities concentrated in Georgia) are grouped together ( = 0.523). As the clustering becomes more fine-grained, the two ethnic groups with whom Georgians have experienced violent conflicts and have not resolved those conflicts to date-Abkhazians and South Ossetians-form one cluster ( = 0.844). It is surprising to observe the actor Georgians on the right of Figure 2 among the out-groups. This, however, can be explained by the fact that, when addressed by nationalist appeals, Georgian actors might be referred in such a cumulative way in order to emphasize the category of ethnicity. Contrary to this picture, the clustering pattern of actors in non-nationalist appeals ( Figure 3) does not reveal any theoretically meaningful groups, i.e., actors are mixed across clusters when considering their ethno-national attribution. Thus, the exploratory analysis of the group structure of the discourse in the Georgian print media shows that compared to the regular, non-nationalist appeals, nationalist discourse entails the actor structure, whereby the members of the in-group are systematically similar in directing their nationalist statements to the same other groups.
To test the robustness of these clustering patterns, the correlation matrices were submitted to multidimensional scaling, an alternative method of partitioning. The results from this analysis are presented in Figures 4 and 5, which compare the spatial distribution of the structurally

e A . i _ m e M C F P L J I M P R P E N R E e r b O e o e o s P i o r e o n i a a o O a F x r o n k s t t R r r p
/ v r e g u t l r n l _ t t n r a h s h u g e P i m s i r e i t k i I . i r g l a e U e s S i c R l e i s n l t i -t D G n e a a z t n r s o a i r d l a l a e a i P o m c n a i i i i v n f C S e a l e r s n c s v C i t i c a a o E a i s i 1 1 1 1 1 1 1 2 2 2 1 1 2 1 2 2 2 Level 8 0 7 3 4 1 2 5 9 3 6 5 2 4 6 2 3 0 7 8 4 9 1 5 1 6 --- Numbers of these levels refer to the degree of association (structural similarity) among the actors in a given cluster. The higher this number, the stronger is the association among actors. The sign 'x' placed between two actors indicates that these actors are clustered together at a given level. The red vertical line is inserted manually in order to illustrate the specific clustering pattern more clearly.   Figure 3. Hierarchical clustering of the structurally equivalent actors in non-nationalist appeals. Notes: Columns represent actors. Rows labelled as 'Level' show the stages of clustering. Numbers of these levels refer to the degree of association (structural similarity) among the actors in a given cluster. The higher this number, the stronger is the association among actors. The sign 'x' placed between two actors indicates that these actors are clustered together at a given level. equivalent actors in nationalist and non-nationalist appeals. Similar to the results of the hierarchical clustering, the comparison here reveals an immediate difference between the two. In the multidimensional scaling picture of the nationalist appeals (Figure 4), Georgian actors alone dominate the lower right corner of the graph, while all the other ethno-national groups of actors are located in the upper left corner. Thus, the divisive structure of the 'us' vs. 'them' discourse is maintained. In contrast, the multidimensional scaling of the non-nationalist appeals ( Figure 5) yields a fully mixed picture, i.e., the spatial distribution of the actors does not form any cluster whatsoever. The next step is to test whether and how the clustering patterns identified above are affected by electoral periods. This analysis will also serve the additional function of ruling out the possibility that the patterns identified in the exploratory analysis are an artefact of the definition of the nationalist appeals in this study, thus validating the proposed method of measurement.
Figures 6 and 7 compare multidimensional scaling of structurally equivalent actors in the nationalist discourse during electoral years to that of the non-electoral years. In line with the stated expectation, this comparison shows that the nationalist discourse differs in its divisiveness depending on what period we look at. The two camps of 'us' (all the Georgians) and 'them' (everybody else) is clearly visible in the nationalist discourse of the electoral periods ( Figure 6), while we cannot identify the same clustering pattern in the nationalist appeals for non-electoral years (Figure 7). We can see from the structure of the nationalist appeals during electoral years that the West and international organizations are also clearly delimited from the actors with which Georgians have experienced ethno-political conflicts.
The finding reveals that nationalist discourse becomes more intense during electoral periods not only because actors with stakes in elections make nationalist appeals more frequently, but also because the general structure of the discourse becomes more divisive, yielding sharper divisions between the in-group and out-groups. Apart from revealing how nationalist discourse becomes more divisive during elections, this finding also serves as an important validation of the proposed method of measuring the actor structure of the nationalist discourse. As the comparison here was made, not between the nationalist and non-nationalist appeals, but within the nationalist discourse itself, it has demonstrated that the identified patterns do not stem from the rules of coding nationalist appeals in this study. Figures 8 and 9 look deeper into the nationalist discourse during electoral periods and explore whether early electoral periods are more prone to divisive discourse than the later ones. In line with the literature attributing particular risks of conflict to the first and second elections after independence, findings from this analysis show that early elections in Georgia are indeed different from later elections in terms of strong polarization between the national 'us' and 'them.' The spatial distribution of the structurally equivalent actors in Figure 8  shows that the distance between the two clusters is particularly large during early elections. Certain actor clusters also exist in the discourse of late electoral periods (Figure 9). For example, the West and international organizations are in closer proximity with each other than with Georgian actors, or the actors with which Georgians have experienced tensions and/or violent conflict. The same holds true for Abkhazia, Autonomous Regions, Russia and the Soviet Union. Although the external actors here are represented as part of the in-group, a clear-cut delineation between the in-group and out-groups, with the clusters situated so far from each other in Figure 8 indicates that nationalist polarization between 'us' and 'them' is indeed stronger during early electoral periods as compared to the later electoral periods.

Conclusions
Adding to the already established practices of analysing in-group/out-group divisions in ideological and political discourses and extending on the discourse network analysis methodology, this article proposes an innovative way of studying the divisiveness of the nationalist discourse using the social network analysis method of structural equivalence analysis. The practical applicability of this approach was demonstrated by analysing the relationship between the instances of increased political participation during democratic transition and the national-ist 'us' and 'them' polarization in the Georgian print media of 1991-2012. The findings suggest that during electoral periods nationalist discourse intensifies not only in terms of its degree, i.e., nationalist statements become more frequent, but also in terms of its kind, i.e., the actor structure of the discourse yields sharper divisions between the in-group and the out-groups.
There are good reasons to think that the exogenous shock of the 2008 war might have affected the patterns of divisiveness of nationalist discourse in the Georgian print media. More precisely, the clustering structure of the discourse during the next few rounds of elections after 2008 would have probably reflected increased tensions with certain out-groups. However, as the first elections after the 2008 war only took place in 2012 and the data that this study relies on covers the period only up until the end of 2012, it does not provide enough data points to conduct structural equivalence analysis of the nationalist discourse in this period. This limitation of the study needs to be improved through further research of the effects of such factors on the divisiveness of the nationalist discourse.
The implications of this study can be understood in relation to what Billig calls "banal nationalism" (1995). The latter sees nationhood as reproduced daily through ideological habits, which remain unnamed and unnoticed. The divisive actor constellation of the nationalist discourse in the media identified in this study is, thus, yet another manifestation of 'banal nationalism.' It is not readily recognizable for a reader, and thus goes unnoticed. However, if we agree that the news media contributes to the construction of reality through its content, this deep discursive structure, which yields such a sharp division between the national in-group and the outgroups, has the potential to reinforce cohesion among the in-group members, but, at the same time, fire up antagonism towards the out-group members. Yet, the findings from this analysis should be taken only as a starting point. While this study has discovered that the structures of the nationalist appeals in the Georgian print media of the democratization period convey a strong 'us' vs. 'them' polarization, through the actor constellation of this discourse, further empirical research is needed to examine, first, to what extent is the pattern observed here identifiable in other cases, and second, to what extent these discursive structures affect the actual ingroup/out-group polarization among media consumers.