Even before the onset of Russia's war against Ukraine in 2014, Twitter, Facebook, and especially Vkontakte¹ played an important role in protesters' mobilization and self-organization (Tucker et al., 2014; Onuch, 2015). While Ukrainian citizens had already protested against electoral fraud in the 2004 presidential election, the protests in 2013/2014 became even more heated. This time, citizens gathered in Independence Square (Maidan) to protest the Ukrainian government's decision to withdraw from an anticipated trade and association agreement with the European Union. On the one hand, the protests of 2004 and 2013/2014 are similar in the sense that Russia's influence on Ukraine was being negotiated. On the other hand, both protests differ remarkably in terms of the state of technological development. Gruzd and Tsyganova (2015) eloquently depict this development in their paper on the social media involvement of pro- and anti-Maidan groups in Ukraine during the 2013/2014 protests:

After Vkontakte became known for being heavily monitored by the Russian government (Menn, 2022), new platforms became substitutes for coordinating the Ukrainian resistance. This trend started during the Maidan protests when Facebook and Twitter were used extensively to coordinate anti-government protests (Ronzhyn, 2014) and attract and mobilize the international community by tweeting in English (Tucker et al., 2014). The migration to Twitter grew dramatically during the full-scale Russian invasion of Ukraine in February 2022, when the micro-blogging service became a platform to disseminate information, influence campaigns, and mobilize resistance (Chen and Ferrara, 2022). For instance, Ukraine's minister of foreign affairs, Dmytro Kuleba, used Twitter to ask “foreigners willing to defend Ukraine and world order as part of the International Legion of Territorial Defense of Ukraine . . . to contact foreign diplomatic missions of Ukraine” (Kuleba, 2022).

Because of its crucial role for the Ukrainian public and its use as a tool of conflict communication (Fisentzou, 2019; Jungblut, 2020), researchers have also investigated Twitter's role in Russia's warfare (e.g., Nikolayenko, 2019). In fact, Russia's strategy has been identified as a form of hybrid warfare, i.e., military activities are being supported by cyber disinformation campaigns that are aimed at spinning the narrative of the war, confusing the enemy, distracting allies, and “achiev[ing] strategic advantage at minimal cost” (Snegovaya, 2015, p. 9). In this context, studies have tried to shed light on the dissemination of Russian political disinformation on Twitter (Nikolayenko, 2019; Linvill and Warren, 2020; Zannettou et al., 2020). According to their findings, the Russian Internet Research Agency made attempts to impact the political agenda of foreign nations between 2014 and 2018, showing “surprising consistency in mean output” (Linvill and Warren, 2020, p. 8). Additionally, the dissemination of pro-Russian imagery on Twitter was found to coincide with real-world, polarizing political events in Western democracies, such as right-wing rallies (Zannettou et al., 2020), suggesting the targeted anti-Western nature of their creation. This “anti-Western information war” (Tsygankov, 2018, p. 6) reflects Russia's evolving perception of NATO from a potential partner to a renewed military threat, as noted by Tsygankov (2018). According to Russia's perspective, NATO is attempting to change regimes across the world, including Ukraine, which had undergone the colored revolutions with the assistance of the United States' democratization strategy (Tsygankov, 2018). This view portrays NATO as an indirect party of Russia's war against Ukraine and may have spread to international social media discussions.

Considering Twitter's significant role in the past 9 years, especially in Ukraine's efforts to attract and mobilize the international community and in Russia's hybrid, anti-Western warfare, the micro-blogging platform offers a promising avenue for investigating how NATO's role in the context of the 2022 full-scale Russian invasion of Ukraine is being perceived and justified in different language communities. We will investigate these perceptions and their justifications using framing analysis.

Framing is the most prominent mass communication theory of the present millennium (Bryant and Miron, 2004), which is one of the reasons why political conflict researchers have increasingly turned to this theoretical paradigm to investigate mediated portrayals of war (e.g., Dimitrova and Strömbäck, 2005; Jungblut and Zakareviciute, 2019; and the special issue “Framing war and conflict” by Al Nahed and Hammond, 2018). Framing can be defined as the process of selecting certain parts of reality and making them more salient, i.e., “promot[ing] a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation" of events (Entman, 1993, p. 52). Due to this, frames become organizing principles that give a meaningful structure to both media content (media frames; Scheufele, 1999, p. 115), and the world (recipient frames; Matthes, 2014, p. 18). In political conflicts, media frames can advance “particular definitions and interpretations of political issues” (Shah et al., 2002, p. 343) by simultaneously excluding other parts of reality (Entman, 1993). As a result of the framing process, media frames can become socially shared recipient frames that persist over time (Reese, 2001).

Against this conceptualization, frames can play a significant role in armed conflict and war, for instance, by influencing who or what is considered to be the cause of the conflict and what the appropriate response should be (Hammond, 2018). Furthermore, frames can potentially affect the course of conflicts by encouraging people to engage in violent behavior (Hamelink, 2008) or, on the contrary, decrease hostilities and contribute to peace and reconciliation (Bratic, 2008). It is not surprising then that political conflicts and wars such as the Arab Spring or the 2003 invasion of Iraq have generated a rich body of content analyses on how armed conflicts are covered by mass media (Jungblut, 2023). However, not all armed conflicts have attracted equal scholarly attention (see Gabel et al., 2020, for an overview of armed conflicts in Tunisia, Egypt, Turkey, Israel and Palestine, and Ukraine). Most studies on conflict and war framing have looked at the English-language media sphere with very few cross-country and even fewer cross-language comparisons. Existing cross-country comparisons tend to focus on English-speaking countries or media networks. Some notable exceptions include a study by Kolmer and Semetko (2009), where the authors looked at how the Iraq War was framed in the main TV news programs in the United Kingdom, the Czech Republic, Germany, South Africa, the United States, and in the Arab-language Al-Jazeera network. Dimitrova and Strömbäck (2005) looked at the framing of the 2003 invasion of Iraq in prominent Swedish and US newspapers, and Dimitrova and Connolly-Ahern (2007) analyzed the online websites of the elite newspapers in the US, UK, Egypt, and Qatar to understand how they framed the Iraq War. Yang (2003) did a comparative study of the framing of the NATO air strikes on Kosovo in 1999 in Chinese and US newspapers. These studies, however, are exceptions. Much of the existing literature on framing investigates how Western media frame foreign conflicts while essentially neglecting the framing of those conflicts in domestic and non-Western settings (Vladisavljević, 2015).

When it comes to war and conflict framing in online environments, specifically social media, cross-national and cross-language comparisons are even rarer. Little is known about how armed conflicts and their respective conflict parties are framed in social media across language-specific communities. In addition, armed conflicts are frequently depicted with a small selection of frames that focus on violence like military actions and casualties, but not on diplomacy and economy (Fröhlich et al., 2007; Baden, 2014). Filling this research gap seems particularly important when the involved parties are deemed ambiguous by the public, i.e., when framing can determine who is considered a conflict party in the first place, who is blamed for starting or escalating the war, and what justifications are raised for this escalation.

Shared language and identity play an important role in mobilization through collective and personalized action frames (Bennett and Segerberg, 2013). From a social psychology perspective, frames can be understood as cognitive schemata of interpretation that provide individuals with shared labels for events and information, and they can generate support for mobilization (Goffman, 1974, p. 21). In times of war, shared frames have the potential to connect people with similar identities, views, and goals. This kind of frame alignment “is a necessary condition for movement participation” (Snow et al., 1986, p. 464), and can serve as a critical tool for managing both obedience and resistance. However, creating a frame alignment within language communities inevitably leads to a narrative bias tied to national identity. Since framing analysis of armed conflict aims at “identify[ing] media bias that for example can be the result of ethnocentrism” (Jungblut, 2021, p. 1), investigating different language-specific communities and the frame alignment within and between these communities is warranted (Entman, 1991). However, there is a notable paucity of research examining how various language groups discuss armed conflicts on social media. As Dimitrova and Strömbäck (2005) have already argued eighteen years ago:

Despite eighteen years of research, this quote is still very much relevant since personalized action frames in different languages are currently being used by the pro-Ukrainian Cyber Army to streamline tactics for resistance on social media (Maikovska and Canevez, 2023). In fact, national identity and nationalism have become significant aspects of conflict communication on Twitter (Gabel et al., 2020). One of the reasons for this development is that sharing a language on Twitter can be used strategically to demonstrate one's affiliation to a nation that speaks that main language, particularly for non-English tweeters (Takhteyev et al., 2012). Therefore, language choice can display one's allegiance during an armed conflict and help to connect with people who share the same national identity, culture, and values (Sheafer et al., 2014). To avoid naive universalism in our analysis of the framing of NATO's role in the full-scale Russian invasion of Ukraine, we opt for cross-language comparisons that allow us to detect possible biases that might result from strategic language use and ethnocentrism.

To provide a framework that allows us to analyze how NATO's role in the full-scale Russian invasion of Ukraine is framed across different language communities on Twitter, we combine framing theory with the interpretive repertoire framework proposed by Baden and Springer (2014, 2017). Although originally developed to measure frame diversity, we argue that Baden and Springer's framework can be adapted to capture the framing of armed conflicts. To that end, Figure 1 illustrates our theoretical argument on how to combine frames and interpretative repertoires to analyze armed conflicts such as the Russian invasion of Ukraine. It also visually depicts the coding scheme used in this framing analysis and guides the following sections.

In communication studies, framing analysis of news coverage of armed conflicts has a long history. As a result, the use of framing theory to analyze the portrayal of NATO in the context of the Russian invasion of Ukraine on Twitter appears to be the rational choice. Traditionally, frames are composed of four elements that combine to form a specific meaning or viewpoint (Entman, 1993): They define problems (problem definition), diagnose causes (causal attribution), make moral judgments (evaluation), and suggest remedies (treatment recommendations, p. 52). Breaking down the meaning of war-related tweets into these four elements has several advantages for analyzing communication during armed conflicts. Most importantly, it increases the ease of identifying patterns within these messages (e.g., how different language communities attribute blame resulting from divergent evaluative standards or situational problem definitions) and therefore contributes to a more nuanced understanding of how the context and consequences of the conflict are perceived on Twitter. For example, by looking at the problem definition element, one can recognize the different issues that are being argued over and perceived as being at stake in the conflict. Next, the causal attribution element gives away the public perception of the underlying causes of the conflict, which can help to identify potential areas of misunderstanding or false narratives and to develop more effective information campaigns. Similarly, treatment recommendations can provide an understanding of what possible solutions to the conflict are being suggested on Twitter, as well as how these solutions are being assessed. Finally, the evaluation element offers insight into how personal and situational values inform judgments about the conflict. We will explain and discuss these four elements in greater detail in the following sections.

The aim of this study is to investigate how NATO's role is being framed in the context of the 2022 full-scale Russian invasion of Ukraine in Russian-, English-, and German-speaking Twitter communities during the initial four months of the invasion. Given the research gaps in the comparative analysis of social media conversations about armed conflicts, we pose three research questions. These questions explore how blame and responsibility attributions toward NATO result from divergent problem definitions and evaluations between different language communities:

Social media conversations about the role of NATO can vary widely, influenced by various factors such as the discussant's home country's historical background and political affiliations. Even within a single language community, there may be significant variation in opinions expressed, as citizens from diverse national and cultural backgrounds contribute to the conversation. For example, English-speaking Twitter users may represent many voices from within NATO, particularly from the US and the UK, who generally support the alliance. Since the US is a major global power and its leaders have historically been strong advocates for NATO, it is not surprising that the majority of US citizens also support the alliance, as evidenced by the NATO Annual Tracking survey (nato.int, 2022). Similarly, most NATO member states also show support for the alliance (nato.int, 2022), which likely influences the English-language discussions on Twitter about NATO. However, the invasion of Ukraine may lead to a greater representation of Ukrainian voices in the English-language Twitter sample as Ukrainians strive to draw the attention of the international community to the invasion. Nevertheless, in the broader context, it is expected that these voices will be a minority within the English-language community.

As for the German-language Twittersphere, despite Germany's NATO membership, its relationship with the alliance is more complex, largely due to its history with World War II. The legacy of the Soviet-backed DDR and its eventual union with the West still influences public debates in Germany, which is why discussions about NATO may be more nuanced and reflect varying levels of criticism and support in the German-speaking Twittersphere. Political parties such as Die Linke and AfD, known for their pro-Moscow stance, may further shape the discourse on social media and lead to heated debates between users with differing views. The same is true for Austria, another German-speaking country with an ambivalent relationship with NATO. Austria maintains its neutrality while cooperating with NATO on various levels. However, its citizens generally do not support joining the alliance, and the far-right party FPÖ often aligns with Russian narratives on global issues. Similar to Austria, Switzerland adheres to a policy of strict neutrality in international conflicts, avoiding alignment with any political or military blocs.

In the Russian-speaking community, opinions on NATO may be largely negative, as the majority of Russian citizens view the alliance as a threat to the country's sovereignty and security, with Ukraine joining NATO posing the greatest threat (Levada Center, 2022). The perception of NATO as a tool of Western imperialism, and the memory of Cold War tensions, could also contribute to the public opinion of NATO in Russia and parts of the Russian-speaking world. However, determining the degree to which Russian-speaking communities align with the viewpoints of the Russian government is a challenging task, given the significant variations in their cultural, political, and social affiliations. Additionally, some Ukrainians may use Russian on Twitter to amplify their tweets by tapping into preexisting Russian-language conversations about the invasion. For some, using the Russian language on Twitter could be a means of expressing their frustration or anger with the Russian government and its actions. In fact, after the 24th of February 2022, access to Twitter has been restricted for some Russian users (Culliford, 2022), likely as an attempt to shield them from pro-Ukrainian messages. All of these factors contribute to the complexity of interpreting the Russian-speaking Twittersphere, making it the most challenging community to analyze.

To address our research questions, we used an open-access dataset of tweets in English, Russian, and German that included the hashtag or keyword “Ukraine.” We posit that the language differences in the subsamples correspond to distinctions between language communities, yet we acknowledge that these subsamples may not provide a clear-cut representation of the perspectives of various nationalities or affiliations. The data was gathered by Münch and Kessling via the Twitter Academic API and published on 1 March 2022 to support open science (Münch and Kessling, 2022). We accessed the repository and rehydrated³ its content for the time range from 1 February to 31 May 2022, which allowed us to analyze the portrayal of NATO during the onset of the full-scale Russian invasion of Ukraine. The initial datasets contained several million tweets per language (English N = 23,384,107; German N = 7,395,587; Russian N = 1,808,828), making sample preparation and quality control a significant undertaking. Following the methodological advice of Grundmann (2022) that the exclusion of duplicate texts is needed to reduce noise, we decided to remove all duplicates and exclude all retweets (see also Silge and Robinson, 2017, for this approach on Twitter data). We decided to keep all tweets with unique text content, regardless of whether they were copy-pasted messages or had slight alterations, as long as they tagged different politicians or NGOs with @-mentions. This decision ensured that our results were not biased by deleting tweets that could have been part of a coordinated action by activists, which has been shown to be a strategy to bring attention to the invasion. For instance, Maikovska and Canevez (2023) investigated how pro-Ukrainian activists coordinated in Telegram channels to implement this strategy to circulate content on social media such as Twitter.

We used a keyword-based selection process to identify relevant documents for the manual framing analysis of the full-scale Russian invasion. Using the keywords “NATO,” “Nato,” and “nato” in Latin and Cyrillic, we collected a corpus that contained all tweets that are explicitly related to NATO (English: n = 1.264.362, Russian: n = 16,028, German: n = 253,401). Because this was still too big a corpus for manual content analysis, we took a random sample of 650 tweets per language. In the end, we coded n = 1.900 random tweets, of which n = 650 are in English, n = 650 in German, and n = 600 in Russian⁴.

Adhering strictly to the theoretical framework outlined in Figure 1, we created a category system to analyze the framing of NATO in the full-scale Russian invasion of Ukraine on Twitter. The resulting codebook was reviewed by the four authors on English, Russian, and German-language tweets. Whenever uncertainties arose, especially due to language or country-specific context, solutions were discussed, agreed on, and recorded in the codebook. Following multiple rounds of inspection and discussion, we conducted a first intercoder reliability test. The outcomes of the initial intercoder reliability test were satisfactory for only a small number of variables, causing us to heavily revise the codebook. When revising, we paid particular attention to the variables which coded the interpretive repertoires because they performed worst. Specifically, we reviewed the literature once again to refine the definitions and examples of “logic of action” and “logic of evaluation” used in our theoretical framework (these definitions are now used in the section about interpretative repertoires of this paper). However, even with the revised definitions, our pairwise interrater agreement was very low in the second intercoder reliability test (Holsti's CR < 0.50)⁵, while the other categories performed well (Holsti's CR > 0.70). In order to improve the measure once again, we binarized all variables which encoded the interpretive repertoires (i.e., the logic of action and the logic of evaluation) and recoded all tweets while only focusing on these variables. This improved our results significantly. Then, using the final version of our codebook, we performed the third intercoder reliability test on 150 tweets per language with satisfactory results, as shown in Table 4. It is worth noting that three variables, which have been rarely observed in the tweets, namely the “life” logic of action, the “ecological” logic of evaluation, and the topic of sovereignty claims, did not yield a positive value on a chance-corrected indicator (Cohen's κ), despite Holsti's CR suggesting a good fit. Therefore, we caution against interpreting these three measurements without careful consideration.

Our first research question aimed to determine which language-community is most likely to explicitly attribute blame and responsibility to NATO for the full-scale Russian invasion at its onset (RQ1). Regarding the blame attribution, we found that from all tweets that discuss NATO's role in the full-scale invasion explicitly, tweets in German were most likely to blame NATO for the (escalation of) the full-scale invasion (17%), while English-speaking users were least likely to perceive NATO as the instigator (12%). The results of a χ²-test indicated that this group difference was significant, χ²(2) = 8.843, p < 0.05, V_C = 0.070.

In addition to these discussions about NATO being responsible for (escalating) the full-scale invasion of Ukraine, there was also a debate about whether NATO could and should propose a solution to end it. When it comes to assigning responsibility, English speakers who talked about NATO's role in the conflict were a lot more likely to express that NATO is responsible for finding a solution (42%) to the full-scale invasion than German (11%) and Russian (7%) speakers. The differences were significant once again, χ²(2) = 277.46, p < 0.001, V_C = 0.397.

Please note, however, that these percentages are based solely on tweets that mention NATO explicitly and, as conditional probabilities, cannot be considered representative of the overall prevalence of NATO blaming on Twitter. It is important to approach these results with caution since they may not capture underlying patterns if, for instance, users of a particular language tend to blame NATO more often but mention it less frequently than other users in the first place.

Our second research question asked how different problem definitions and evaluations are associated with attributing blame to NATO in the three language communities (RQ2). We fitted logistic regression models to gain an understanding of how different types of blame attributions derive from different problem definitions and from supporting one conflicting party over the other. We proceeded hierarchically by successively introducing two models, beginning with Model 1, which focused on the problem definition (composed of topics and the logic of action) and moving on to Model 2, which incorporated the evaluation, i.e., the expressed support for the conflict parties and the logic of evaluation for that support.

Our findings show that in the Russian-language Twittersphere (Table 5) discussing the military conflict (e.g., troop movements, equipment, and stalemates) was positively associated with attributing blame to NATO (b = 1.33, p < 0.001). Similarly, mentioning interactions between the actors that are based on desires and emotions (e.g., resentment, fear, hatred, or pursuit) was also positively related to blaming NATO (b = 0.82, p < 0.01). However, discussing the topic of violence of war (e.g., war crimes, the aftermath of bombings, or casualty totals) was negatively related to blaming NATO (b = −1.22, p < 0.01). The same was true (b = −1.60, p < 0.001) for the discussions about interactions that involved material exchanges (e.g., production, purchase, borrowing, or consumption). All these relationships remained statistically significant in Model 2 as well, where we added the evaluation variables. Among those, only supporting Ukraine was negatively correlated with blaming NATO, although this relationship did not achieve the conventional level of statistical significance (p = 0.065).

In the German Twittersphere, attributing blame to NATO was negatively linked to both taking a neutral stance (b = −1.81, p < 0.001) and supporting Ukraine (b = −2.56, p < 0.001). Interestingly, those who talked about history were more likely to blame NATO, but only as long as stances were not being considered, i.e., in Model 1 (b = 0.94, p < 0.001). Finally, in the English Twittersphere, being pro-Russian (b = 2.19, p < 0.01) and talking about the economic consequences of the war (b = 0.93, p < 0.05) was positively associated with blaming NATO. Furthermore, two interesting associations emerged, although they were not statistically significant. First, using moral arguments to evaluate the conflict parties, such as arguments about justice and humanity (p = 0.055), was positively related to blaming NATO. Second, tweeting about interactions that make some kind of demands was negatively associated with blaming NATO (p = 0.056).

In our third research question we wanted to determine if NATO was held accountable for finding a solution, taking into account causal attributions, such as blaming NATO, Russia, or Ukraine for the invasion (incorporated in Model 3). To do this, we used logistic regression models again (Table 6).

First, we were able to find that in the Russian Twittersphere, there was a positive relationship between holding NATO responsible for finding a solution and talking about interactions between actors that are based on desires (b = 1.45, p < 0.05), demands (b = 0.96, p < 0.05), and material exchanges (b = 1.73, p < 0.01). In addition, evaluating conflict parties based on their truthfulness was negatively linked to holding NATO responsible (b = −2.08, p < 0.05). Finally, tweets that blamed Russia to have started the full-scale invasion against Ukraine were more likely to state that NATO could provide a solution to the conflict, but this relationship was not significant (p = 0.095).

Moving to the German Twittersphere, we saw that discussions about diplomatic negotiations (b = 0.71, p < 0.05) and the violence of war (b = 1.10, p < 0.05) often went hand in hand with expecting NATO to find a solution. In a similar vain, responsibility attributions toward NATO were associated with actors being asked or demanded to act (logic of action: ought, b = 0.91, p < 0.01) or to interact with a structure (logic of action: technology, b = 0.93, p < 0.05). At the same time, mentioning historical events (b = −1.61, p < 0.05) and evaluating the conflict parties based on truthfulness (b = −2.05, p < 0.01) were negatively linked to perceiving NATO as a possible peacemaker. Additionally, tweets that were supportive of Russia were less likely to believe that NATO could provide a solution to the conflict, but this relationship was not significant (p = 0.076).

Finally, the English-language Twittersphere was making NATO responsible when talking about military action (b = 0.61, p < 0.05) and in the context of asking or demanding actors to act (b = 1.12, p < 0.001) or to interact with a structure (b = 0.62, p < 0.05), but also during negotiations (b = 0.94, p < 0.001). In addition, tweets that voiced their support for Ukraine were also more likely to hold NATO accountable for finding a solution (b = 2.44, p < 0.05), as were tweets that blamed Russia for the invasion (b = 0.56, p < 0.05). Talking about human interest topics (b = −2.01, p < 0.05), on the other hand, and blaming NATO for the invasion (b = −0.90, p < 0.05) were both negatively linked to believing that NATO could provide a solution.

Our study comes with limitations that have to be addressed. To begin with, we investigated tweets spanning from 1 February to 31 May 2022. As the invasion continued and more information about the Russian military attack and the atrocities with which the Russian army operates is released, the public opinion reflected in the Twittersphere might have changed. Future research could include data covering a greater time span to enrich the findings and provide a broader picture of the discourse.

Furthermore, it is crucial to take into account the origin of the tweets before making any conclusions. Most of the tweets included in the open-access dataset used in this study lack geographic tags, meaning that while we can determine the language used in the tweet, we cannot ascertain its origin. It is worth noting that individuals from various countries, including Ukraine, may have written tweets in Russian intending to influence the opinions of Russian citizens. Similarly, some tweets written in German may have come from German-speaking countries such as Austria or Switzerland. English tweets likely have the most diverse origins. As a result, we have opted to refer to language communities instead of countries. It is conceivable that factors beyond language use, such as political culture and history, are responsible for the group differences in perceiving NATO as an instigator or peacemaker. This uncertainty about the origin of the tweets makes our results more difficult to interpret because they cannot be clearly and exclusively attributed to cultural and historical idiosyncrasies. However, it is important to remember that the tweets we examined represent actual content that Twitter users who speak Russian, German, and English encounter when participating in discussions in their native language, irrespective of their country of origin. Therefore, analyzing the discourse that transcends country barriers is intrinsically valuable because it indicates the frames that citizens are exposed to when engaging in these language-specific conversations.

Speaking about the origin of the tweets, we also need to discuss their creators. It is possible that at least some tweets that mention NATO are coming from bots or trolls and are not reflective of public opinion. They can show us what content Twitter users are exposed to but might also bias the findings about how often NATO is blamed for escalating the war or appealed to for finding a solution. Finally, it is difficult to generalize public opinion based solely on tweets since the manually-coded sample size is quite small, and the tweets may not be representative of the population as a whole.

More critically, our observational framework does not enable us to determine the direction of the association between frame elements and blame or responsibility attribution. While we can identify which frame elements correspond to these stances, we do not know which comes first when composing the tweet. Do users first consider a problem thoroughly and then interpret the situation before deciding whether NATO is responsible or to blame for finding a solution? Or do they already have a strong preconceived stance on these matters, leading them to concentrate on topics and evaluative standards that align with their viewpoint? Future experimental research should seek to untangle these relationships.

Finally, while the interpretative repertoire framework as proposed by Baden and Springer (2014, 2017) empirically poses challenges, especially when analyzing short text forms such as tweets, we firmly believe that the interpretative repertoire framework makes a strong contribution to the theoretical development of framing theory. We encourage future research to explore whether a combination of framing theory and the interpretative repertoire framework can be applied to analyzing the media portrayal of armed conflicts in longer text forms, such as Facebook postings or traditional media coverage.