The uses of the term polarisation in Swedish newspapers, 2010–2021

Political polarisation has become one of the prime objects of academic scrutiny. Besides the increased scientific attention, the term polarisation has seen an uptick in use also in media and public discussions linked to recent phenomena as disparate as the Covid-19 pandemic, the spread of misinformation, and the rise of populism and antidemocratic movements within Western societies. In other words, polarisation is a term used by scientists, journalists, and citizens that has become central to conversations about politics.

Though many of the phenomena associated with political polarisation are global, American politics and media are known to be particularly polarised (Mason, 2018; Oscarsson et al., 2021), which correlates with the fact that much research has its empirical focus on the US (Kubin & von Sirkoski, 2021; Winkler, 2019). In addition, Western media cover American politics extensively, much more than any other country or region. This coverage is particularly high across the Nordic countries (Hammarlund & Riegert, 2011; Gross, 2011), even though the Nordic countries themselves are quite different from the US. The Nordics have relatively low levels of political polarisation, are characterised by multi-party systems, and have less separation within the media systems along partisan lines (Fletcher et al., 2020; Huddy et al., 2018; Oscarsson et al., 2021; Urman, 2020). We therefore argue that studying the rhetorical functions of, and the frames associated with, the term polarisation in Swedish print media is particularly interesting, since it enables us to observe the various uses of the concept across different political, social, and media contexts.

In this article, we study the meanings that Swedish newspapers assign to the term polarisation. Therefore, we do not depart from an a priori defined and ideal notion of political polarisation (i.e., how polarisation itself is framed) to uncover how it functions as a framing device. Instead, we study the term as it is used within these newspapers and then inductively assign the discovered categories to scholarly polarisation types. Of these, most relevant to the context of our study are ideological (distance between policy preferences) and affective (negative evaluation of the political outgroup) polarisation (Levin et al., 2021). In previous studies, ideological polarisation has been related to economic inequality, globalisation, and cultural change (Inglehart & Norris, 2017), and in Europe, to the rise of populist parties and the decline of traditional left–right divisions (Dalton & Berning, 2022). Affective polarisation has also been connected to the rise of populism (Harteveld et al., 2022) and to the role of identity and emotions in shaping attitudes towards immigration (Wojcieszak & Garrett, 2018). Wojcieszak and Garrett also find supporting evidence for the role of media in shaping perceptions of polarisation, which further merits the study of polarisation as a framing device.

In this article, we explore the frames constructed around polarisation to uncover how polarisation is conceptualised and used in Swedish news media discourse. We approach this goal through three complementary research questions.

To respond to these questions, we use a mixed-methods approach that combines an in-depth qualitative rhetorical analysis of polarisation as a framing device in 240 Swedish newspaper articles with a computerised analysis of 32,805 newspaper articles covering the period 2010–2021. The first research question aims to uncover how the term functions in everyday news texts, which discloses nascent ideas of its meanings and uses. The second and third question are posed to uncover the associative range of the term, its aggregate meaning, which may aid the interpretation of media content related to polarisation. Furthermore, these questions reduce potential bias through coding for manifested frame elements (Walter & Ophir, 2019). Taken together, this fundamentally intertextual approach was chosen to provide an understanding of how the term polarisation develops and functions as a media frame within our study period.

We begin by outlining the theoretical premises of our article, then move on to cover the state of the art and research design. The qualitative and quantitative studies are presented sequentially, each with their own empirical material, methodologies, and findings. These are followed by a summative discussion.

News media play a key role in “the production and reproduction of sociality, social relations, social structures, social systems, and society” (Fuchs, 2020: 377). The proliferation and reproduction of media framings contribute to a convergence of meaning around key symbolic terms, such as polarisation, within a community. This can constrain discourse in an ideological way, which is a main concern of framing theory (Entman, 2007). Conversely, the widespread use of a concept may have dispersive effects on its meaning. A notion such as polarisation may, in this sense, be confused, ambiguous, or change its range of meaning over time (Perelman & Olbrechts-Tyteca, 1971: 132–135). Studies have shown that in journalistic writing, entertainment and an exciting narrative can be just as crucial as factual or conceptual precision, which can increase the vagueness of scientific concepts (Secko et al., 2013; Madsen, 2003).

According to Entman (1993: 52), to frame something “is to select some aspects of a perceived reality and make them more salient in a communicating text, in such a way as to promote a particular problem definition, causal interpretation, moral evaluation, and/or treatment recommendation”. As such, to frame is a rhetorical action insofar as its role is to persuade an audience to adopt a certain belief or attitude toward an object through making notions present to the mind (Perelman & Olbrechts-Tyteca, 1971: 117), at the same time as rhetorical structure is considered to be an integral part of frame analysis (Van Gorp, 2007). While we take the term polarisation to generally be used as a framing device – that is, devices through which a news frame is manifested in media through “word choice, metaphors, exemplars, descriptions, arguments, and visual images” (Van Gorp, 2007: 64) – our inquiry aims to more precisely determine how it functions as such in Swedish news writing.

We understand rhetoric as the “theory of reasoned persuasion” that explains the properties of communication that “create or increase the adherence of minds to the theses presented for their assent” (Perelman & Olbrechts-Tyteca, 1971: 45). Consequently, “polarisation” mandates a study based on both the assumption that its connotative meaning is central to precisely what it makes salient, and that there may be patterns in the way that it is employed as a framing device. We approach the former through quantitative and qualitative analysis of newspaper contexts of polarisation, in a sense treating these contexts as “reasons” to adhere to a certain thesis of what polarisation is. For the latter, we inquire into the rhetorical functions of the word polarisation, that is, the abstract roles the word performs in persuading an audience to adopt the various theses of the newspaper articles. It concerns the direct manner of how aspects of a perceived reality are made salient.

A rhetorical function in this sense discloses the persuasive, expressive, poetic, and referential aims of utterances to analysis (Kinneavy, 1980). Insofar, the answers to our three research questions interact in several ways. For example, the actors, issues, and concepts associated with polarisation provide parts of its meaning and will connotatively interact with how reasonable and forceful a charge of being polarising is within a text’s persuasive framework. As our analyses show, patterns in the contextualisation and rhetorical functions of the term polarisation disclose both dynamics of the evolving frame of polarisation and how it is put to use by writers referencing that frame in persuasive discourse. This approach is somewhat novel to rhetorical scholarship, which has typically approached polarisation as a rhetorical strategy with corresponding expressions in style, argument, framing, or constitutive speech (Desilet & Appel, 2011; Fortuna, 2019; King & Anderson, 1971), rather than directly inquiring into its rhetorical uses as a notion in public discourse.

A few studies discuss polarisation as a news frame. Fiorina and Abrams (2008) traced the media description of American politics as polarised to the early 1990s, when electoral success was attributed to a “culture war”, driven by “angry white males”. They connected the narrative of polarisation to a change in the campaign tactics of the political parties: Their focus moved from swing voters to their partisan bases. Attracting swing voters requires ideological triangulation and centrist policies, whereas a base-turnout strategy emphasises ideological differences.

Robinson and Mullinix (2016) showed in a framing analysis of American newspapers that elite polarisation was portrayed as negative, and almost half of the articles called for more bipartisanship. Levendusky and Malhotra (2016) performed a content analysis of news articles in American newspapers for the election years 2000–2012 containing the word polarisation and found a sharp increase in the number of articles mentioning it. Polarisation was used in the context of an increasing number of issues and was intricately connected to partisan conflict.

This article builds on analyses such as the ones above done in the American context and aims to apply a similar approach to Swedish newspaper articles. In this way, we are breaking new ground, since we are not aware of any other study outside the US that charts how the concept of political polarisation has been articulated in the news. Moreover, to our knowledge, no computational frame analyses of polarisation have been published, as the two studies discussed above used content analysis only on a small subset of news items.

Our study used a mixed methods research design. In Study 1, we explore the first research question, about the meanings and rhetorical functions of the term polarisation in a qualitative analysis of 240 articles. This study also allows us to become acquainted with the data and the evolution of the context in which the term polarisation appears in newspapers. In Study 2, we examine the second and third research questions: which are the actors, and which are the issues apparent in the data.

For both studies, the media database Retriever was queried for the search term polarise* (i.e., encompassing the Swedish noun polarisation, the adjective/adverb polarised, and the verb polarise). The database was accessed through a special agreement between Lund University and Retriever Media Group, using a researcher API. Notably, Retriever does not consistently distinguish between news and opinion pieces or between newspaper sections. As a result, we provide a holistic analysis of the uses of the term polarisation and its word family, without considering the various journalistic subgenres.

The first part of our research used a combination of manual coding and interpretation of the uses of the word polarisation as a framing device in 240 articles from the ten largest national daily newspapers [Storstadspress] in Sweden. Six articles from every quarter of each year from 2011–2020 were randomly selected, excluding results not concerning people or ideas (i.e., optics, physics, climatology). Four coders analysed the use of the term, and each was assigned one quarter from every year in the period. Four benchmark analyses were used as a guideline for coders. The resulting codes were checked against the benchmarks, with only an insignificant number of interpretations deviating from the benchmark, thus ensuring intercoder consistency (O’Connor & Joffe, 2020). Detailed information about the articles analysed and quoted in this section can be found in Table 1 in our online data repository (OSF, 2023).

To establish which issues and actors are connected to polarisation and how polarisation is employed as a framing device, coding was performed for contextual information (article genre, author) and qualitative aspects of the word use. Several categories were identified through an inductive initial reading of the material, followed by a feedback loop where smaller categories were merged or dismissed. At the end of this category development process, the following codes were found to be consistent throughout the material: Site (i.e., Sweden, US, Europe, the Internet, the media, etc.); Target (who or what is described as polarised, i.e., actors, debate or issue, public opinion); Quality (ideological, social, material, or affective); Specificity (the degree of specific vs. general in context); Issues (political topics related to polarisation); Strategy (whether polarisation was implied to be a strategic action or not); and Expert quote (whether polarisation appeared in an expert quote). An overview of the coding results is provided in the Appendix.

Considering the results yielded by the category identification, the sample was revisited with a qualitative analysis of rhetorical functions performed using the notion of polarisation. The main rhetorical functions identified were as problem definition or description of a situation or actor, as motivation for a position, as a moral evaluation of a position or actor (negative, as in an indictment, or positive, as in supportive), or to contrast a position or actor with others. Our analytical scheme thus follows the traditional frame analysis (Entman, 1993), in which problem definition, moral evaluation, causal attribution, and treatment recommendation are seen as core functions of a frame following.

For Study 2, a total of 32,805 articles were obtained by querying the Retriever database for Swedish articles that included the same phrase used in Study 1, anywhere in its text (i.e., title, captions, and main body). In contrast to Study 1, we did not just include the category “Storstadspress” (top-10 national newspapers) in our analysis, but also included “Prioritised Provincial Press” (regional press), and “Landsortspress” (local press), and we refer to these categories as media types throughout the rest of the paper. Since data for regional and local press on Retriever is consistently recorded only after 2009, we restricted our query to the period of 1 January 2010–30 June 2021, which coincides with the end of the second Löfven government in Sweden.

Since publishing practices varied during the observed period, we designed a baseline query against which the first query was compared. That baseline query identified all articles that included the terms politics, economy, society, or any combination thereof, with the rationale that the usage of these terms should be relatively constant over time. This was the most general query we could design given the practical limitations of the Retriever interface, leading to almost two million identified articles.

Drawing on previous developments within computational frame analysis (cf. Walter & Ophir, 2019), we rely on computational methods to analyse the large amount of text that we collected. Using such natural language processing methods can be a way to augment human coding and “offer insight into what frames might exist in a dataset, which theory has not already anticipated” (Nicholls & Culpepper, 2021: 162). Using the entire data collected and analogue to Study 1, we proceeded to identify the context in which references to polarisation were made and how such references changed over time. For this purpose, we analysed all text with respect to named entities (i.e., words referring to places and individuals). In line with the rules of Swedish spelling, we operationalise named entities as those words starting with a capital letter (not appearing at the beginning of a sentence). We analysed which named entities are mentioned most often in the context of polarisation and how the relevance of these entities changed over time.

We also trained word2vec models that arranged all words in all texts in a common vector space, where proximity between words corresponds to a relatively high co-occurrence and semantic relatedness of words (Goldberg & Levy, 2014). Pre-processing included the deletion of stopwords and lemmatisation of the remaining text which, given the relatively low amount of text available, was necessary to support word2vec by reducing the dimensionality of text and make the words associated with polarisation more interpretable (i.e., avoiding the possibility of finding many inflections of the same lemma in the words surrounding polarisation). Then, documents were divided into two batches, based on whether they were published before or after 1 January 2016. This splitting point was chosen for several reasons. First, this presents the half-way point in the observed time period and therefore allows us to observe change in the use of the term polarisation. We abstained from analysing the data on a year-to-year basis because the amount of data per year was too small to allow per-year word2vec modelling. Second, 2016 is also the year during which the American presidential campaign of Donald Trump started, which we expected to be an important event in the context of polarisation. We thus trained two separate word2vec models on each of those batches and compared commonalities and differences between the associations that the models learned between polarise* and other words. Note that although the amount of text is at the lower end of what is recommended for reliable word2vec models, we only used our models to estimate distances between pairs of words that include polarise*. All our documents are guaranteed to include polarise* at least once (through the query), and the model only considered terms that appeared at least ten times.

Study 2 result: Time trends

We start our analysis by presenting the general time trends of the usage of the expression polarise* in Swedish newspaper articles. Figure 1 (left panel) shows the number of entries from the query per quarter and media type. Across media types, we see an increase in the production of articles that include references to polarisation. Although the 12-year observed period might be deemed relatively short for studying language changes, Figure 1 illustrates that incorporating entries from before 2010 would yield only minimal observations. Consequently, it would not support a quantitative analysis. As a result, we pinpoint the early 2010s as the period when the concept of polarisation began to gain traction. To normalise the trend against the total article production during that time, the right panel in Figure 1 shows the quarterly distribution of articles relative to the number of articles in the baseline query. In other words, the right panel shows the relative frequency of references to polarisation in each media type. Again, we see that across time, references to polarisation increased, were most common in national press and quadrupled during the observed period.

We also tested for differences with respect to this time trend between the top-ten national newspapers by circulation but found that all outlets followed a similar trend. Although polarisation as a term was increasingly used across different media types, it was less common in local newspapers. This is likely due to their lesser focus on American politics.

While Figure 1 establishes the increase of articles including the term polarisation, we find that this can mainly be attributed to a term in the main body of an article. A similar time trend could not be observed for headlines or headers, where the term hardly appeared at all. On average, the term polarisation was used 1.15 times per article in our sample, and more than 99 per cent of the articles with mentions of polarisation used it less than three times per text. In other words, polarisation is almost never the main topic of an article or central in its vocabulary.

FIGURE 1

Study 2 result: Named entities

As a first analytical step, we identified the named entities that were most common in sentences including the term polarisation. We find that US-related entities were (by far) the most common among the 50 most mentioned entities. Table 1 shows the distribution of the geographic area from which the entities come. Furthermore, Table 1 also lists the most common entities by region, showing that entities from the US are more focused on politicians (primarily Donald Trump and Barrack Obama), while entities in Sweden and other regions are far more focused on geographic locations (in particular, cities). For more details, the Appendix presents the full list of the 50 most common entities.

TABLE 1

Most commonly named entities found in sentences that included polarise*

Entities from	Same sentence mentions	Most common entities (by number of mentions)
USA	5,223	USA (2,417), Trump (833), Obama (323), Washington (172)
Sweden	3,070	Sweden (1,680), Stockholm (302), Gothenburg (218), Malmö (151)
EU	2,176	Europ* (700), EU (409), Russia (162), Germany (158)
Other	697	Israel (193), Turkey (174), Iraq (133), Erdogan (71)

Figure 2 provides more detail on the politicians mentioned in the context of polarisation. The left panel shows that, whenever the term was used, Swedish media generally reported more about American politicians, with the notable exceptions of 2014–2015: During this period, Swedish general national and European elections were taking place, and the country was transitioning from a centre-right to a Social Democratic government. During this election, the right-wing populist party Sweden Democrats made significant gains, establishing itself as the third largest party in Sweden. In other words, the word polarisation only began to be used in a Swedish context when the populist right-wing party rose to power. Interestingly, as can be seen in Figure 2, this usage of the term polarisation was preceded by the usage of the term in the context of American politicians and national elections, speaking in favour of the concept of political polarisation being imported from the US to Sweden.

After 2015, we again see a clear dominance of American politicians, particularly with the presidential campaign of Donald Trump. We find that, by orders of magnitude, Trump is the most mentioned politician, mentioned more often than other American and Swedish politicians combined. As shown in the right panel of Figure 2, articles using the term polarisation mentioned Trump most frequently overall; the upward pointing line signals that the mentions of Trump were on the rise throughout 2020. Of all the politicians identified in our corpus, he was by far the one described as most polarising. Looking across the three panels, the uses by Swedish newspapers of the term polarisation was most intrinsically connected to the figure of Trump, who may single-handedly be the driver of the of the polarisation-related coverage of American politics.

FIGURE 2

As a second analytical dimension, we qualitatively categorised those same words regarding their affective connotations (relating to the concept of affective polarisation) or relationship with specific issues (ideological polarisation). We also considered including a category for actors, institutions, and other named entities; however, none of these entities appeared to be uniquely enough associated with the term polarise*. Approximately half of the words associated with polarise* did not fit with our categorisation and are therefore not included in our analysis. These words included primarily synonyms and antonyms (the full list of words is presented in the Appendix).

TABLE 2

Words associated with polarise* in the word2vec models

Type of polarisation	2010–2015 & 2016–2021	Unique 2010–2015	Unique 2016–2021
issues and ideology	Segregering [segregation] Progressiv [progressive] Främlingsfientligheten [xenophobia] Inkomstklyfta [income gap] Tolerans [tolerance]	Rasismen [racism] Islamofobin [islamophobia] Transparensen [transparency] Kunskapsklyfta [knowledge gap]	Politisering [politicisation] Politikerförakt [politician contempt] Populismen [populism] Faktaresistens [fact resistance] Otryggheten [lack of security]
affective	Misstro [distrust/suspicion] Oro [worry] Hätsk [virulent] Osäkerhet [uncertainty] Stämning [mood] Infekterad [infected] Fientlighet [animosity]	Tryck [pressure] Förfäran [dismay] Förhoppningsvis [hopefully]	Missnöje [discontent] Uppskruvad [wound up] Hårdna [lit. hardening] Oförsonlig [irreconcilable] Råare [cruder/beastlier] Förgiftad [poisoned] Pessimism Optimism Misstänksamhet [suspicion]

Comments: Within each cell, words are ordered by their frequency of occurrence.

Comparing the cosine similarity between the words in Table 2, we used the standard t-test to determine whether the mean cosine values were statistically different from each other and found that the term polarisation was generally closer to words in the “issues and ideology” category than those in the “affective” category (p < 0.05). Strikingly, the words under the “issues and ideology” category unique to the 2016–2021 period describe more general social trends such as politicisation, populism, and contempt towards politicians, while the words unique to and common with the earlier time period primarily deal with concrete policy issues such as segregation, the income gap, and racism. Thereby, the term polarisation seems to have moved from a term that is used to discuss concrete issues to a term that is describing more general societal developments and, likely, an issue in itself, which is a confirmation of the qualitative analysis performed in Study 1.

Not surprisingly, whenever used to denote an attribute or affective term, the word polarisation is associated with negative moral evaluation. This negativity appears to grow stronger over time, with “förfäran” [dismay] as the only strong affective modality in the early period, contrasting with “oförsonlig” [irreconcilable/spiteful], “råare” [beastlier, cruder] and “förgiftad” [poisoned] in the latter. In conjunction with the qualitative study, this indicates that the negativity frame intensified throughout the period – not only in scope, but also in terms of level.

We also tested the cosine similarity between the aforementioned named entities and the term polarise* and found that all American politicians from Figure 2 are closer in terms of cosine similarity to polarise* than the closest Swedish politicians (Jimmie Åkesson and Björn Söder, the party leader and the leader of the parliamentary group – and later second deputy speaker of the Riksdag – of the Sweden Democrats, a radical right-wing party). These results point in the same direction as the analysis from Study 1, where we also saw the term polarisation being mainly used as a framing device in the context of American politics, and more seldomly in the Swedish context (only during national elections).

This particular distinction between the mentions of American versus Swedish politicians led us to visualise (see Figure 3) the polarisation conceptual space along a spatial dimension that separated the American and the Swedish context (on the vertical axis) and a time dimension, before and after 2016 (on the horizontal axis).

FIGURE 3

The main takeaway from Figure 3 is that affective polarisation terms are most closely associated with the American context, whereas ideological polarisation terms are more common in the Swedish context. The affective terms cover sentiments such as pessimism and optimism, as well as a feeling of increased negativity (harder, more intense, infected, worry, suspicion, incompatibility, cruder). Ideological terms associated with the American context are populism and knowledge resistance; in the Swedish context, ideology is denoted by words such as insecurity, segregation, income cleavage, islamophobia, and xenophobia. Another interesting tendency is that, after 2016, the incidence of affective polarisation words increased relative to the period before 2016. In contrast, ideological terms did not change much before or after our benchmark year.

Our study of the use of the term polarisation as a framing device in Swedish newspaper articles during the last decade found that polarisation is increasingly used in the media coverage of politics and society. However, most of the time, the word did not appear in the title or header of the article but was used instead in the main body of the texts. When measured as a proportion of the overall coverage of economic, social, and political news, the frequency of polarisation quadrupled in the period 2010–2021, with the term’s use taking off especially from 2016 onwards. This could be due to the rise to power of Donald Trump, an American anti-immigration populist leader who won the American presidential election in 2016. Just before the election of Trump, the Syrian refugee “crisis” unfolded in Europe, which led to a large influx of asylum-seekers on the European continent. The dramatic increase in the use of the term in the Swedish print media may be due to these two concomitant events and their long-term implications, since populism and immigration are the two factors that the academic literature often connects to polarisation (see, e.g., Grande et al., 2019).

In response to our first research question, we find that the newspaper articles did not contain coherent news framing about polarisation, and that the functions of the word polarisation changed gradually throughout the period. Initially, the term was used for descriptive purposes in problem definitions, to describe a concrete situation or debate, as a simple adjective in highly defined circumstances. In more recent newspaper articles, the term seemed to be primarily used for the purpose of rhetorical amplification – infusing the term with a moral evaluation. We also found strategic uses of the term in debate, either to justify one’s own position, or to undermine one’s adversaries. Over time, the media discourse about polarisation moved from a focus on concrete issues towards a more general discussion of social and political tensions. The overall trend points to an increase in negative uses of the term over time.

Two notable observations from the qualitative analysis are the diffusion of the bipolar meaning of the word polarisation and, conversely, the seeming reification of the phenomenon in some texts. The bipolar implications of polarisation is certainly reductive in many cases reporting on both American and Swedish politics. It does, however, make some sense applied to the bipartisan American system that Study 2 showed to be dominating reporting using the concept. We find that these implications get blurred over time in the uses in media, which raises some critical considerations. The shift detected in our material is seemingly metonymic (cf. Laclau 2008), in the sense that Swedish and American polarisation are contingently associated, rather than analogous, through the conflictual orientation implied by the uses in Swedish media discourse. If the term infuses frames with a sense of bipolarity, as would be the case in a metonymic association of American and Swedish politics, it could influence the way citizens perceive multi-party politics, constraining it to bipolar oppositions (Levendusky & Malhotra, 2016). On the other hand, if polarisation is reified into a state or social force, or understood as a trait of contemporary society, it can divert attention from the realities of conflict. Used in this manner, the term can shape the view of conflicts as expressions of polarisation, rather than polarisation as the expression of material or political disparities. While our studies found examples of polarisation in this reified sense, the comparable lack of polarisation as the main focus of articles would suggest that it is generally understood as an epiphenomenon. Both these tendencies relate to Swedish media framing of politics and should be considered in analysis of conflict reporting.

To respond to our second research question about the actors most often named in connection with the term polarisation, both qualitative and quantitative approaches point out that polarisation was most often used to describe political developments in the US throughout our study period, constituting almost half (46.7%) of same sentence mentions. The term was also associated with Swedish politics (in 27.5% of all sentences), which was equal to the proportion of mentions in connection to all other countries combined (26%). An analysis of named entities showed that certain political figures were seen as more polarising than others. In the American context, candidates for the American presidency, particularly Donald Trump, were frequently mentioned in the context of polarisation. In Sweden, the top figures associated with polarisation were the leader of the radical right-wing party Sweden Democrats, Jimmie Åkesson, Prime Minister Stefan Löfven, and to a lesser extent, the leader of the Centre Party, Annie Lööf. This confirms the findings of previous studies about the “Trump effect” (e.g., Inglehart & Norris, 2017) as well as the role of populist anti-immigration parties as drivers of polarisation (e.g., Dalton & Berning, 2022; Harteveld et al., 2022).

Our most interesting finding comes as a response to our third research question, about the issues with which media coverage associated polarisation. Here we find a clear distinction between the American and Swedish context. Terms denoting affective polarisation were clearly more used as framing devices in stories on American politics and society, whereas those connected to political issues appeared in the Swedish political and social context. We can conclude that Swedish media coverage most prominently depicts the US as a locus of polarisation, and that it describes American polarisation as primarily affective. In contrast, Swedish politics is not framed as being as polarised as in the US; when occurring, polarisation in Sweden happens along issue lines such as the income cleavage or immigration. Thus, both value-based distinctions (along the GAL-TAN scale) and left–right contestation are present in the politics and society coverage in Sweden, supporting recent evidence about party politics produced by Hagevi and colleagues (2022).

Looking at the trends in the use of the term polarisation over time, the distinction between the US and Sweden remains meaningful for the entire period covered in our study. While ideological polarisation is rather constant throughout, the affectively laden reporting about polarisation increases, but only in texts related to the American context.

Based on the above distinctions, we find that it may be fruitful to separate two salient ways in which polarisation acted as a framing device in the Swedish newspapers during our research period. The first frame that became crystalised, which we might call “US politics as polarised”, appears as the most common, stable, and well-defined through its strong association to actors and its articulation in connection to events such as elections. The US is also clearly associated with an affective dimension of polarisation, which takes on a more pronounced negativity over time.

A decidedly more diffuse and less referenced frame of “polarisation in Sweden” seems to be present in parallel to the American frame. The Swedish frame is more strongly associated with ideological or issue polarisation, in particular migration, but overall, it has a weaker attachment to particular actors and policies. This frame does have an affective aspect, however, it is less salient than is the case of the American frame. This diffusion is also evident in how the named entities analysis showed it to be typically referencing locations (cities) rather than persons in Sweden.

The rhetorical functions of description and amplification are not restricted to any single frame. In particular, amplification is used across contexts, with the rhetorical force hinging on connotations of conflict, spite, irrationality, and morality. Our study constitutes a novel contribution to the study of polarisation in rhetorical research and may therefore lay the groundwork for further analyses that test the generalisability of these two functions in other media systems.

Our study has several limitations, among which we can count the small number of articles analysed in the qualitative sample. For the quantitative analysis, the major differences in frequency use of the term from year to year made tracing a more fine-grained chronological change impossible. In addition, the database used did not provide consistent information about the author names and the newspaper section where articles were published, and thus we could not detect if the term was used by particular authors or in a specific journalistic genre (editorial, reportage, book review, etc.).

To address these limitations, future research should endeavour to produce more fine-grained analyses of the media production of polarisation frames. This could be achieved in qualitative ways, by performing interviews with journalists who authored texts that included the term polarisation, to find out which meanings and functions they assigned to this particular framing device. Quantitatively, a computational model could be trained to detect the valence of texts where the word polarisation has a central position, to compare with the valence of articles where the term is used only in passing. Moreover, the model could potentially extract the genre of the article to check whether the notion of polarisation has different functions depending on the type of article it appears in.