EU citizens and undocumented migrants in the news: quantitative patterns of representation in Swedish news media 2006-2016

ABSTRACT

Undocumented migrants and poor EU citizens have been frequent topics for Swedish political debate and media reports in recent years. However, there is a lack of representative, large-scale studies on media representation of these groups. This study aims to compare and analyse (1) the broad patterns of representation of EU citizens and undocumented migrants in Swedish national press and (2) how these patterns relate to relevant regulative events over the period 2006–2016. Theoretically, it draws on critical discourse analysis and social problems theory. The sample includes 10 022 referrals in 5411 news articles and the methodological strategy is inspired by corpus-driven discourse studies. Words that tend to occur in the near vicinity of the migrant categories (collocates) are a primary focus for the analysis. The study finds that media referred to EU citizens by employing poverty related discourses during the period 2013–2016 and that the governmental term ‘vulnerable’ EU citizens was adopted in media when a national coordinator was appointed. In contrast, undocumented migrants were associated with discourses on social rights over a time period (2008–2013) when rights to health care and education were debated and subsequently introduced.

KEYWORDS:

• EU citizens
• undocumented migrants
• quantitative
• media
• Sweden

Introduction

International migration and mobility raise many questions for national welfare societies and different forms of irregular migration are especially contested. In Sweden, the precarious position of migrants with no or unclear legal residence rights has raised intensive public and political debate on both inclusion/exclusion to welfare services and repressive strategies such as deportation and bans on begging. Undocumented migrants and EU citizens who do not meet the criteria for full residence rights represent crucial examples of contested migrant categories. The extensive Swedish media and political discourse on EU citizens sleeping rough and begging on the streets has made poverty visible in a way that implies a profound challenge of the notion of an inclusive welfare state. The visibility of these EU citizens can be contrasted against the, at least often presumed, hiddenness of undocumented migrants – ‘hidden refugees’ has in fact been one of the terms used to refer to this category in public debate (Nielsen 2016; Sigvardsdotter 2012). Both migrant categories are arguably positioned at the bottom of what Sainsbury terms the ‘hierarchical differentiation of immigrants’ social rights’ (2006, 230) and a profound uncertainty about how to understand and handle them has emerged. Thus, representation of these two categories in a country whose self-image is said to be that of a humanitarian superpower (Geddes & Scholten 2016) makes Sweden an intriguing case to explore.

There are several arenas in which perspectives on societal developments are brought up, interpreted and communicated, for example in politics or media. Although there is notable interaction between these different arenas (Hilgartner and Bosk 1988), mass media is likely to have a prominent position in promoting certain perspectives on issues that reach wide public attention. Media discourse is thus not an isolated product but both socially shaped and socially shaping (Fairclough 1995). By ascribing characteristics to and connecting migrant categories to specific social, administrative and legal conditions stressing possible solutions and designating responsible actors, media text can be influential in setting the political agenda. This study aims to compare and analyse (1) the broad patterns of representation of EU citizens and undocumented migrants in Swedish national press and (2) how these patterns relate to relevant regulative events over the period 2006–2016.

Previous studies have highlighted different aspects on how these categories have been represented and managed in the Swedish context; foci include social workers’ perspectives (Jönsson 2014; Björngren Cuadra and Staaf 2014; Nordling 2017; Ekendahl, Karlsson, and Minas 2018) social rights and access to services (Lundberg and Spång 2017; Nielsen 2016; Sigvardsdotter 2012) and other aspects (Ciulinaru 2017; Barker 2017). There is, however, a lack of systematic knowledge especially on a larger scale in how EU citizens and undocumented migrants have been represented in Swedish news media. The present study uses methods from the new field of corpus-driven discourse studies to analyse news media text related to these groups quantitatively. In particular, the analysis focuses on dominant patterns of words that tend to appear near EU citizens or undocumented migrants in order to generate broad and representative accounts of representation. It takes into account all articles where either of the migrant categories are mentioned in Swedish national press over time (2006–2016). This time frame includes important events and developments such as the EU enlargement in 2007 and legal changes in social rights for undocumented migrants and thus allows for an analysis of the relationship between these events and media representation.

Previous corpus-driven discourse studies

In recent years, several studies have employed corpus linguistic techniques in order to analyse large samples of naturally occurring text. Studies have addressed for example media representation of migrant categories such as asylum seekers, refugees, immigrants and migrants (Baker and Tony 2005; Gabrielatos and Baker 2008; Abid, Manan, and Zuhair Abdul Amir 2017; Blinder and Allen 2016; Taylor 2014; Islentyeva 2020), undocumented migrants (Brouwer, van der Leun, and van der Woude 2017) or minority groups such as Muslims (Samaie and Malmir 2017; Baker, McEnery, and Gabrielatos 2013; Baker 2010). Other foci include how national borders are portrayed in policy documents and news articles (Vollmer 2017), elderly (Mautner 2007) or corruption (Orpin 2005) in large pre-existing corpora.

For instance, Gabrielatos and Baker (2008) find that there is conflation and confusion on migration terminology in British press and that negatively biased, nonsensical terms such as ‘illegal asylum seekers’ are used. Based on their findings, they argue that media is responsible for moral panic on migrants in the UK. Similar findings come from Blinder and Allen’s (2016) study in which a relation between public attitudes and media representations on migrant groups in the UK is supported. However, they also argue that media plausibly contributes to a mismatch between public perception of and actual migration statistics. Brouwer, van der Leun, and van der Woude (2017) test the hypothesis that Dutch media increasingly connected undocumented migrants with crime leading up to a proposal to criminalize illegal stay in the Netherlands. Their findings, however, do not lend support to that hypothesis. Typically, these studies employ corpus linguistic techniques such as collocation (i.e. the relationship between a search term and words that occur in its near vicinity) and keyword analysis on large samples of data. The idea is basically that the number of times certain messages are communicated increases the likelihood of changes in social practice and discourse.

Two recent manually coded content analyses have studied how immigration was framed in Swedish national news media between 2010–2015. Strömbäck, Andersson, and Nedlund (2017) analysed the occurrence of different forms of immigration (refugee or labour among others) and positive and negative frames of immigration in news sections. For instance, negative frames (such as immigration weakens social cohesion) were more frequent. The authors argued that this was not supported by empirical evidence on the effects of immigration and could thus contribute to readers overestimating costs and problems. Bolin, Hinnfors and Strömbäck (2016) categorized editorials on immigration into positive, negative or neutral and found the latter to be the most prominent (79%) followed by negative frames.

A theoretical perspective on media discourse

As a framework for analysing the relation between media text and other aspects of society, the present study takes theoretical input from critical discourse analysis (CDA). Discourse analysis is often associated with close readings of relatively small samples of text, but Baker et al. (2008) have argued and demonstrated that using corpus linguistic methods for analysis of large samples of texts can be successfully combined with theoretical perspectives from the field of CDA. Baker et al. stress that they ‘do not view CDA as a method nor are specific methods solely associated with it’ (2008, 273) which stands in contrast to the often prevailing view on CDA as a theoretical and methodological package (Winther Jörgensen and Phillips 2000). Agreeing with Baker et al.’s position, some theoretical notions from CDA are employed and complemented with a theoretical perspective on competition between social problems.

Norman Fairclough, the key theorist in CDA, defines discourse as ‘ways of representing aspects of the world’ (Fairclough 2003, 124); that is, a discourse is a representation of a certain aspect such as a practice or structure (e.g. migrant categories) from a certain perspective. Different discourses on migration, for instance, can represent the phenomenon as a threat to national cohesion and identity, a welcomed importation of labour force, or an ethically just feature of an open society. Furthermore, repetition, commonality and stability over time are all implied in the term discourse (albeit in different degrees depending on the scope of the particular discourse) (Fairclough 2003). This means discourses cannot be mere individual expressions or thoughts but rather shared persistent views and should arguably leave traces in the form of repetition of certain words or phrases. Consequently, changes over time in these patterned traces imply discursive change.

One of the key aspects of Fairclough’s theoretical perspective on discourse is that it is dialectically related to the social world and thus both socially shaped and shaping (Fairclough 1995). Fairclough distinguishes between three analytical dimensions of communicative events such as articles in newspapers – text, discursive practices and other (non-discursive) social practices. The text dimension is represented by the actual text and its qualities, which in the present case is articles from news media on undocumented migrants and EU citizens. The text dimension is embedded in discursive practice, which includes the processes of text production and reception in which pre-existing discourses are used to both create and interpret the text. This can be related to professional practices of journalists and the chain of communicative events media articles go through before publication (Fairclough 1995). The discursive practices, in turn, are embedded in (non-discursive) social practices; thus, discursive practices mediate the relationship between text and the social world (Winther Jörgensen and Phillips 2000). Social practices relate to the broader societal and cultural context and include the political system and changes in the political landscape, administrative and legal settings and practices, the presence of migrant groups etc.

Within different domains such as the media, certain social developments are interpreted as problematic and thus get the status of a social problem. Hilgartner and Bosk use the following definition of social problems: ‘putative condition or situation that is labeled a problem in the arenas of public discourse and action’ (1988, 55). They stress that there is competition over attention between different social problems in a limited space in different institutional arenas. Among the factors involved in what problems gain attention, a sense of drama and novelty can increase the level of attention a social problem receives while repetition can cause ‘saturation’ and leave a space open for other social problems to take; and problems gain from relating to shared cultural themes or political interests (Hilgartner and Bosk 1988). The aspect of competition within a limited space stresses that not all social problems can be in the spotlight at all times and that the selection processes are not random, albeit complex. The factors of drama and novelty on the one hand and saturation on the other highlight that there is a need for renewal of social problems in order to maintain their level of received attention. That is, saturation happens when the same discourses are used repeatedly to frame the problem to the extent that it will no longer capture the attention of the audience. In the case of media, this is likely a concern for editors and journalists because it could impact readership and thus economic profit. Consequently, saturated problems will either decline in terms of news reporting or be renewed by framing the problem differently as to maintain readers’ interest.

Fairclough’s text dimension is reflected in the present study’s empirical focus on word patterns and thus on certain lexical choices in media. Importantly, these patterns do not reflect specific discourses by themselves as that would require a closer analysis of each article in context. The rationale here is instead that these patterns broadly reflect traces of discourses which reflect social practice. In this study, word patterns on the text level are put in relation to regulative events more closely related to non-discursive social practices. Following this theoretical backdrop, the study includes regulative events on the national level such as legal changes and governmental investigations that are reflections of social practice.

As a point of departure for the analysis, an inductive-style strategy is employed starting from the quantitative patterns in the text to identify what words are related to undocumented migrants and EU citizens in media text. Secondly, a subset of these patterns is used to analyse problems and solutions associated with these groups over time in relation to a selection of important regulative events. The analyses are primarily descriptive.

Attention in public policy

The political public debate on EU citizens in Sweden has in recent years largely focused on Romanian and Bulgarian Roma who allegedly beg for money on the streets and sleep rough. Citizens from these countries gained access to the freedom of movement within the EU after the expansion in 2007. In 2012, begging EU citizens began appearing in larger numbers and increased sharply to be approximated at around 5000 in 2015 (Swedish Government 2016). The phenomenon quickly gained public attention and in 2015 about half of the Swedish population were positive to a ban on begging (Zelano 2015). In January 2015, the Swedish government appointed a national coordinator for ‘vulnerable EU citizens’ whose role was to support and coordinate agencies’, non-governmental organizations’ and local governments’ work with EU citizens ‘who do not have a right of residence in Sweden … [and] … far less chance of gaining access to welfare’ (Swedish Government 2016, 13). That is, EU citizens who do not meet the criteria for residence past three months and thus, in line with EU regulations, are not entitled to subsidized health care or social assistance. A year later the final report was released, which suggested no legal changes but that County Administrative Boards should continue coordination of municipalities and non-governmental organizations. The report itself, however, received critique in the public debate for some of its positions, for instance that children of vulnerable EU citizens should not be granted access to education in Sweden. Parallel to the coordinator’s report, a legal change in July 2017 aimed to make evictions easier for landowners and was initiated partly due to the presence of unauthorized settlements of vulnerable EU citizens.

In contrast to the free movement of EU citizens, undocumented migrants per definition lack the right to reside on a state’s territory (De Vito et al. 2015). This makes the category subject to what de Genova calls ‘deportability’, defined as ‘the possibility of being removed from the space of the nation-state’ (de Genova 2002, 439). Undocumented migrant can be seen as a residual category in the sense that it refers to not belonging to a legal migrant category (Sigvardsdotter 2012). According to a governmental report from 2010, between 10 000 and 50 000 undocumented migrants were presumed to reside in Sweden (Envall et al. 2010). Two legal changes on health care and education respectively that increased undocumented migrants’ formal social rights in Sweden took place in 2013. Following a debate caused by UN critique of Sweden’s restrictive approach 2007, a public investigation suggested that undocumented migrants should be able to access health care on the same terms as Swedish citizens (Swedish Government 2011). However, the subsequent legal change meant undocumented migrant children gained full health care rights and restricted to care ‘that cannot be deferred’ for adults (Lundberg and Spång 2017). Furthermore, a right to education for undocumented children was established in 2013 after the suggestions of two public investigations published in 2007 and 2010. In Table 1, relevant regulative events for both migrant categories are summarized.

Table 1. Regulative events for both migrant categories.

Eu citizens	Undocumented migrants
2007 – Romania and bulgaria become eu member states 2015 – national coordinator appointed 2016 – coordinator’s public investigation 2017 – legal change on evictions	2007 – UN critique2010 – Public investigation on right to education 2011 – Public investigation on right to health care2013 – Legal changes for health care and education

Methods

The corpus

The examined corpus in this article consists of articles published in the online versions of four daily Swedish national newspapers over the period 2006–2016. Inclusion criteria for the articles were that at least one of the search words for EU citizens or undocumented migrants should be mentioned at least once. Two of the newspapers (Dagens Nyheter (DN) and Svenska Dagbladet (SVD)) can be considered broadsheets in terms journalistic content. DN is positioned as independently liberal and SVD as independently conservative which in Sweden both translate to the right wing of party politics. The two other newspapers (Expressen and Aftonbladet) can be classified as tabloid papers. Expressen is independently liberal while Aftonbladet is independently social democratic.

The sampling strategy aimed at creating a corpus of widespread Swedish daily news media. The four included newspapers are daily and have physical copies. Additionally, they are among the most well-read and influential Swedish newspapers. Online newspaper reading generally increased in Sweden during the studied period while reading of paper copies decreased (Andersson, 2018). In an average week during March 2017, Aftonbladet’s website was visited by 3.8 million viewers, Expressen’s by 3.3 million, DN’s by 1.1 million and SvD’s by 1 million (Kantar Sifo 2017). The sampled newspapers can thus be presumed to represent and influence majority or mainstream discourses on contemporary social issues. Other sources, such as social media, could generate valuable insights into the dynamics of public discourse at large but mainstream media was chosen due to its vast reader base and power to influence public discourse.

The corpus was sampled using an online Swedish media database called Retriever Research which gives restricted access to news media from Sweden and other countries. The database allows Boolean search queries as well as selecting specific sources and time periods. Retriever automatically erases duplicates (80% agreement or more) when articles are selected for download and approximately 2–3% of articles were removed by that function. Specific terms denoting for example nationality or ethnicity were avoided in the search strings in order to create a corpus that matches the study’s focus on the two migrant categories. For instance, ‘Roma’ or ‘beggar’ were not included although they have frequently occurred in the EU citizen debate. The sampling was made separately for each migrant category, newspaper and year to enable separate analyses.

Initial scoping searches were made prior to the final sampling in order to test what terminology was adequate, to ensure that relevant terms were used and that truncation did not generate too many irrelevant hits. For EU citizens, two of the most common general terms were used (EU citizens and EU migrants) for the final sample. For undocumented migrants only one term was used (‘without papers’, Sw. papperslös). The final sample of 5411 articles (see Table 2) included all articles in which these terms were mentioned. Overlap between the two corpora was low (86 articles in each corpus included both migrant categories).

Table 2. Sample overview. Distribution of articles between the four newspapers.

	DN	SvD	Expressen	Aftonbladet	Total
EU citizens	810	971	429	985	3195
Undocumented	596	661	259	146	2216
Total	1406	1632	688	1131	5411

Terminology in Swedish press

Although the main focus in this study in terms of EU citizens is on the small but widely discussed fraction of these that have limited residence rights, the plain term ‘EU citizen’ was used as a point of departure methodologically. In line with the inductive study design, the advantage of this is that it enables an examination of the way EU citizens are described without predetermining a specific terminology. When EU citizens are referred to in news media, the term tends to focus on the specific group of interest here (but not exclusively) and this is also expected to be reflected empirically. ‘EU migrants’ tend to focus exclusively on the group of interest for this study. ‘EU citizens’ and ‘EU migrants’ were combined into one search term for the analysis. Thus, when EU citizens are referred to in the results section, both terms are included unless otherwise stated.

Undocumented migrants are most commonly referred to as ‘without papers’ in mainstream media. Sometimes other terms are used, although rather rarely. The most notable exception is ‘illegal immigrants’ which was used in an earlier sample for this study. However, a preliminary analysis showed that ‘illegal immigrants’ was predominantly used in reports about the United States. This was presumably a result of an adoption of terminology used in American politics and media. Thus, it seems that ‘without papers’ dominates media content in Sweden.

Analytical approach

The statistical analyses were done using WordSmith Tools 7 (Scott 2016), which is a software for statistical analysis of text corpora. In order to examine how the two migrant categories were described and what kind of immediate context they were mentioned in, collocation analysis was used. Collocations are based on concordances, which are pieces of text where a word (EU citizen, for instance) occurs in a corpus. When performing collocation analysis, the words in these concordances are counted and words that regularly appear near a target word in a statistically significant way are referred to as collocates (Baker 2006). What counts as ‘near’ another word is determined by the researcher by selecting how many words to the left and right respectively will be included in the analysis. These positions are referred to by direction and a number signalling how many steps in that direction the position is (e.g. L4 is the fourth word to the left). Choosing a span that is too wide will include words that are less likely to refer to the target term and a span that is too narrow will exclude potentially relevant words. In line with previous research, a span of L5-R5 was considered adequate for this analysis. The list of co-occurrences signals which words are used in the context of the search term. However, words that tend to occur often in the whole corpus (like ‘the’ or ‘of’ in English) will also tend to occur often near the specified word. Aside from pure frequency, there are a number of tests that ‘take into account the frequency of words in a corpus and their relative number of occurrences both next to and away from each other’ (Baker 2006, 101). That is, they signal whether the co-occurrences are statistically associated with the search word. Since the different tests ‘favour’ different kinds of words, Baker (2006) suggests that a combination of tests can be used. The two tests used in this study are Mutual Information (MI), which favours low-frequent lexical words (as opposed to grammatical), and log-likelihood (LL), which favours more high-frequency words (Baker 2006). After removing words that fell under the test values chosen as thresholds for considering words collocates (MI≥5.0, LL≥6.63), the collocations were ranked according to frequency which ultimately favours the more common words in the context of the migrant categories. Additionally, a self-made list of words was used to automatically remove any remaining function words that lacked analytical value at the study’s level of analysis (e.g. prepositions, pronouns and some verbs). Both span width and threshold values are the same as in previous similar studies (Brouwer, van der Leun, and van der Woude 2017; Blinder and Allen 2016).

In the first step, the collocates were calculated in WordSmith and the 20 most frequent collocates for each migrant category were included. As a second step, the total of 40 collocates were coded and ordered into content-based themes independently of migrant category. These themes were used to broadly classify collocates for further analysis. Two original categories (‘numbers’ and ‘place and residence’) were merged into ‘other’ because they lacked direct analytical value. In the last step collocates categorized as ‘problems’ and ‘solutions’ were examined over time by calculating collocations for each year and group separately. The results from the last step are presented in relative frequency per year in order to minimize the influence of uneven distribution of occurrences over time. Interpretations of the empirical material are tentative and mainly focus on possible implications of discursive patterns and the relationship between the patterns and regulative events.

Results

Media attention 2006-2016

The frequency of published articles on undocumented migrants and EU citizens between 2006 and 2016 is shown in Figure 1. Several notable peaks could be identified; in the case of undocumented migrants, media attention varied between as low as 40 articles in 2007 and reached its maximum at 527 in the last year of the span. The first peak in 2008, the year after the UN report critique, was over two times the number of articles compared to the previous year. In 2013, almost 400 articles mentioning undocumented migrants were published, which coincides with the health care and education legal changes. Based on a quick examination of concordances, the highest peak in 2016 could partially be explained by discussions on undocumented migration in the ongoing presidential election in the US.

Figure 1. Published articles 2006–2016.

For EU citizens, the significant rise in media attention 2014–2016 peaking in 2015 with 1075 articles shows the sudden intensity of the recent debate. Prior to this, the annual frequency of articles varied between 85 in 2008 and 190 in 2013. The annual occurrences of the two terms ‘EU migrants’ and ‘EU citizens’ were also examined (not in table/figure). The term EU migrant did not exist in media terminology until 2012 and occurred rarely in 2012–2013. In the following years, it rose to a significant peak in 2015 of about twice the popularity of ‘EU citizens’ but dropped significantly in the following year. The second term, EU citizen, also increased during the same time period but to a lesser extent and did not drop as much in 2016. That is, the popularity of the first term does not seem to have been long lasting and was surpassed by the second term in 2016. This is possibly the result of an adaptation in the media to the terminology used in governmental contexts.

Collocates for both migrant categories

The first analytical step was collocation analysis for both migrant categories separately across all newspapers and over the whole time period. The 20 most frequent collocates are found in Table 3. Overall, numbers (represented by ‘#’) were the most frequent collocates for both categories. One of the general differences between collocates for the two categories was that EU citizens had several adjectives that modify the target term (e.g. vulnerable, homeless). These were also almost exclusively appearing in the L1 position. In contrast, undocumented migrants had several collocates that seem to be modified by ‘without papers’. This was most evident for the different migrant categories appearing as collocates (e.g. refugees, immigrants) where the R1 position was most common. It is likely that this difference is a result of the grammatical difference between undocumented and EU citizens – the latter can only be used as a noun while the former can also be an adjective. Furthermore, EU citizens is a more ‘neutral’ term by itself while the term undocumented has an embedded problematization (being without documentation).

Table 3. List of collocates for EU citizens (n = 5478) and undocumented migrants (n = 4544). Top 20 ranked by frequency. Years 2006–2016.

	Eu Citizens			Undocumented Migrants
Rank	Collocate	Frequency	%	Collocate	Frequency	%
1	#	814	14,9%	#	688	15,1%
2	Vulnerable(utsatta)	410	7,5%	Care(vård)	501	11,0%
3	Sweden (Sverige)	345	6,3%	Right(rätt)	381	8,4%
4	Homeless(hemlösa)	240	4,4%	Refugees(flyktingar)	304	6,7%
5	Poor(fattiga)	231	4,2%	Child/-Ren(barn)	277	6,1%
6	Right(rätt)	229	4,2%	Immigrant/-S(invandrare)	198	4,4%
7	Child/-Ren(barn)	153	2,8%	Millions(miljoner)	173	3,8%
8	Many(många)	140	2,6%	Asylum Seeker/-S(asylsökande)	165	3,6%
9	Great Britain(Storbritannien)	135	2,5%	Hidden(gömda)	157	3,5%
10	Malmoe(Malmö)	122	2,2%	Migrants(migranter)	140	3,1%
11	The Number(antalet)	118	2,2%	Stockholm	106	2,3%
12	Reside/-S(bor)	115	2,1%	Health Care(sjukvård)	104	2,3%
13	Reside/-S(vistas)	108	2,0%	Hunt(jakt)	82	1,8%
14	Begging (tiggande)	101	1,8%	LIVE/-S(lever)	79	1,7%
15	Stockholm	100	1,8%	DEPORT(deportera)	69	1,5%
16	Millions(miljoner)	95	1,7%	THE POLICE(polisen)	66	1,5%
17	Camp/-S(läger)	91	1,7%	CALLED(kallade)	62	1,4%
18	Coordinator(samordnare)	84	1,5%	Worker/-S(arbetare)	61	1,3%
19	APPLIES(gäller)	79	1,4%	Humans(människor)	60	1,3%
20	The Country(landet)	75	1,4%	Women(kvinnor)	56	1,2%

As a second step, the collocates from both analyses were crudely categorized under four themes which emanated from content-based coding (problems, solutions, auxiliary categories and other). The distribution of collocates under these themes (see Table 4) sheds light on apparent differences in how the two migrant categories were represented in media. For EU citizens this meant a significant focus on various poverty related problems (e.g. homeless, begging, vulnerable) rather than solutions. Thus, it seems that EU citizens were rather actively problematized in media in a way that arguably constructed the migrant category as a problem in itself. In contrast, collocates in the solution theme were more common for undocumented migrants. These solutions predominantly referred to on the one hand rights and (health) care and thus welfare solutions but on the other hand collocates denoting repressive action or order issues (deport, police) were present. Furthermore, among the collocates for undocumented migrants a more diversified way of referring to them using migrant or other person categories appeared in the auxiliary categories theme. In comparison to undocumented migrants, this made EU citizens appear as an anonymous crowd rather than real life, human beings. This is quite contrary as to what one might expect given that the kind of problems discursively associated with EU citizens are publicly visible. It also stands in contrast to the notion of undocumented migrants as a hidden mass.

Table 4. Collocates coded under thematic headings.

	EU citizens	Undocumented migrants
Problems	Vulnerable Homeless Poor Begging Camp/-s	Hidden
Solutions	Right Coordinator	Right Care Health care Hunt Deport The police
Auxiliary categories	Child/-ren	Child/-ren Women Humans Worker/-s Migrants Asylum seekers Immigrant/-s Refugees
Other	#Millions Manythe number Stockholm The countryreside/-s (x2) Sweden Great britain Malmoe Applies	#millions Live/-s Stockholm Called

Problems and solutions over time

In a last step, the collocates that were coded as problems and solutions were examined over time. The relative frequency of each of these collocates per year is displayed in Figure 2 for EU citizens. A general observation is that the collocates that describe different problems began receiving attention in around 2012 and certainly in 2013. In fact, this can be said about all examined collocates except ‘right’. The pattern largely followed the number of articles published over time in the sense that the rise of different problem-collocates happened at around the same time as the rise in media attention, as presented in Figure 1. This could arguably be seen as an effect of the uneven distribution of articles over time. That is, since a larger share of analysed articles originated from the last years in the time span, they had a bigger impact on the collocation analysis for the whole time period and consequently on which collocates were examined over time. However, when collocates were calculated separately for two equally comprehensive time periods (not in table/figure), no further problem/solution collocates were found among the highest-ranking ones.

Figure 2. EU citizens collocates. Relative frequency per year.

‘Homeless’ was one of the more prominent collocates, especially for 2013 when it was present in 23% of occurrences and both surrounding years were high at around 8–9%. It continued to drop in following years, to being barely existent in 2016. In the 2013 peak, a number of issues were raised such as the number of ‘homeless’ EU citizens, a governmental census on the group and access to shelters. ‘Poor’ was also prominent in 2013 and varied between 5–7% until 2015. ‘Camp/-s’ appeared 2014–2015 at comparatively low rates and ‘begging’ as well as ‘coordinator’ in 2015–2016. In the last years, the rise of ‘vulnerable’ was apparent from 4% in 2014 when a governmental report on human trafficking using the phrase ‘vulnerable EU citizens’ was released – to 15% in 2016. This rise coincides, to a high degree, with the rise in media attention – making ‘vulnerable’ very frequent among collocates. ‘Right’ was the only collocate that was present across the whole timeline. A sharp increase of ‘right’ in 2010 may be explained by Swedish police’s deportation of Romanian beggars the same year and in this context the kind of rights being referred to was residential – that is, the right to move attached to the EU citizenship.

Focusing on the main events indicated in Figure 2, the accession of Bulgaria and Romania into the EU in 2007 seems to have had little effect on media representation, since article frequency was low during that period as well (see Figure 1). The years before a national coordinator was appointed in 2015, several problematizing collocates were prominent but they were somewhat gradually superseded by ‘vulnerable’ – the preferred governmental term. The main conclusion to draw from this is that Swedish news media adapted to governmental discourse.

In Figure 3, the corresponding collocates for undocumented migrants can be found. The period between 2008 and 2012, which includes the debate on health care rights, stands out with two collocates that were quite prominent. In between 14–21% of occurrences over the years 2008–2012 ‘care’ was within the collocation span. This alone shows how central this question was in relation to undocumented migrants. ‘Right’ followed a similar trend but with slightly lower annual shares of occurrences. It seems that ‘right’ was quite closely connected to the health care debate, in phrases such as ‘undocumented migrants do not/should have health care rights’.

Figure 3. Undocumented migrants collocates. Relative frequency per year.

The peak of ‘hunt’ and ‘police’ in 2013 was likely due to police methods for finding and deporting undocumented migrants that caused debate the same year. In the last year of the timeline, none of the collocates except ‘deport’ stand out. This, however, largely referred to reports on the 2016 presidential election in the USA.

Interestingly, the main events that marked the start (2007 UN report) and end (2013 legal changes) of the debate on health care for undocumented migrants was in large reflected empirically in the sense ‘care’ and ‘right’ collocates were relatively more prominent during that time period. Although these collocates were low in 2007, the following years marked the start of the debate. The same goes for 2013, which however was a peak in article frequency. The issue of right to health care was thus brought up in media before any extensive governmental action and partially dropped after the legal changes.

Discussion

This article has addressed the representation of two migrant categories in Swedish news media and how these migrant categories have been associated with problems and solutions in relation to regulative events over an 11-year time period. One of the main conclusions to draw from the analysis is that there were central differences in the media representation of EU citizens and undocumented migrants. In comparison to each other, EU citizens were tied to several poverty related problems while remaining quite anonymous whereas for undocumented migrants, solutions and ways of diversifying them into person categories were more prominent. Furthermore, media representation could be related to main regulative events. In the case of undocumented migrants, representation focused on rights and care during the period captured in between the UN report and the subsequent legal changes after which these issues gradually lost attention (cf. Lundberg and Spång 2017). The connection to governmental events is most striking for EU citizens in terms of a terminological shift to ‘vulnerable’. Additionally, there is not much to suggest there was any conflation or entangling of the migrant categories because shared collocates were few and the overlap of articles was low (2,6–2,9%).

Different forms of ‘old’ poverty problems were brought up in relation to EU citizens and when a governmental terminology started to form (vulnerable EU citizens), it was adopted in media. This shows a preference for governmental discourse but it also reflects uncertainties in media about what kind of terminology should form the problem representation on vulnerable, poor, homeless or begging EU citizens. In a more general sense these collocates indicate a discursive pauperization of EU citizens that construct them as target groups for various interventions. Barker (2017) uses the term ‘benevolent violence’ to capture the duality and mixed responses from the Swedish government (e.g. simplifying evictions but not banning begging) closely linked to the benevolent ambitions of the welfare state. Tentatively, the choice of ‘vulnerable’ over more specific terms can be interpreted as a kind of discursive adjustment to encompass these diverging responses in social practice.

For undocumented migrants, the weight instead lied on solutions – or more precisely the right to health care or the lack thereof. One could argue that the lack of rights was one way of defining the problem with undocumented migrants. However, in the political debate that preceded the legal changes, opposing arguments were raised that such legislation would send inconsistent signals about Swedish policy towards undocumented migrants (Nielsen 2016; Sigvardsdotter 2012). The rights-oriented patterns of discourse on undocumented migrants can also be interpreted as more abstract and reflective of a notion of invisibility than their visible EU citizen counterparts, for which concrete social problems were apparent. In both cases the encompassing welfare state set the stage on which the textual patterns could be interpreted as anomalies – one of rightlessness (undocumented migrants) and one of destitution (EU citizens). The textual output produced by media discursive practices have shown to repetitively reinforce these positions.

In terms of amount of media attention, the case of EU citizens was more sudden and concentrated to the last years of the time span. Theoretically, this could indicate that (the problems of) EU citizens were able to compete with other contemporary social problems like the refugee crisis in 2015. It is likely that EU citizens gained attention by the expressions of visible poverty people inevitably see in their everyday lives and this was addressed in the media. Its ‘success’ as a social problem can be related to factors addressed in the competitive perspective on social problems as outlined by Hilgartner and Bosk (1988). The representation has aspects that connect to drama (visible poverty), novelty (new problem connected to mobility in an enlarged EU) and shared cultural themes (‘old’ poverty expressions seen as remnants of a period predating the modern welfare state). Arguably it is the combination of being a new and old, visible problem that has instigated attention in media and other public arenas (cf. Swärd 2014).

In both cases, media discourse followed events such as legal changes and governmental reports. Arguably, news media was also involved in setting the political agenda and thus promoting certain perspectives or actions, by selecting topics to report on and perspectives to highlight. In the case of EU citizens, the distribution of articles over time speaks in favour of this since the subject began receiving substantial attention before any clear governmental action. The accession of two new countries to the EU in 2007 barely received any attention at all, seen in articles mentioning EU citizens. In other words, news media did not predict the development in the following years. Undocumented migrants and the lack of right to health care was also brought to attention in the media several years before the legal change was carried out and right/care collocates actually dropped the same year.

Methodologically comparable studies in other national contexts have found media to be misleading and media discourse on various migrant categories to be rather overtly negative in terms of their depictions (Brouwer, van der Leun, and van der Woude 2017; Blinder and Allen 2016; Gabrielatos and Baker 2008). Apart from the problem-oriented collocates (e.g. poor/homeless/begging), this was not as evident in the present study. This could be the result of an ambition to keep a neutral or balanced tone in relation to contested migrant categories (Bolin, Hinnfors, and Jesper 2016). In using a different methodology and sample, the predominantly negative framing of immigration in Swedish media found by Strömbäck, Andersson and Nedlund (2017) is not necessarily incommensurable with this. For instance, the level of analysis (article) is different from the present study’s (word) and the overall framing might not be reflected as clearly in collocational patterns as in manual coding.

Methodological limitations

The methodology of the present study has both benefits and shortcomings. One of the benefits of a corpus-driven method is being able to analyse large and representative samples of text. On the other hand, it decontextualizes the text in the sense that words are taken out of their immediate grammatical/semantical context irrespective of their meaning in actual sentences but also in whole articles. Additionally, as with all methods, methodological choices affect the validity and scope of the study. For instance, an important part of the recent debate on EU citizens in Sweden has been on a possible begging ban and the issue of begging was not as prominent in the results as one might expect. A possible explanation for this is the choice to focus predominantly on collocates drawn from a restricted frame before and after the target word. It is possible that this and other issues might have been reflected better using other strategies that, for instance, would take larger sections of the articles into account. But what was lost in focusing on these smaller sections was arguably gained in precision and increased certainty in that collocates were in fact connected to the migrant categories.

Disclosure statement

No potential conflict of interest was reported by the author.