Introduction
Depression in young people is of growing concern in Ireland, with an estimated 5% of young people experiencing depression and 20% experiencing some degree of emotional distress at any given time (Costello et al. Reference Costello, Mustillo, Keeler and Angold2004; Lynch et al. Reference Lynch, Mills, Daly and Fitzpatrick2006). Depression in young people is associated with poor mental health outcomes in adulthood and risk for suicide (Brent et al. Reference Brent, Perper, Moritz, Allman, Friend, Roth, Schweers, Balach and Baugher1993; Rao et al. Reference Rao, Ryan, Birmaher, Dahl, Williamson, Kaufman, Rao and Nelson1995; Carr, Reference Carr2003). A recent large-scale survey of 14 306 young people in Ireland (aged 12–25 years) conducted by Headstrong, the National Centre for Youth Mental Health, found that 30% of adolescents fell outside the normal range for depressive symptoms, with depression increasing with age (Headstrong, 2012). The adolescent respondents in the survey reported that school, family and friends were the three biggest stressors affecting their lives. For the young adults surveyed, university/studies, money, work and family were key stressors. Alcohol misuse was identified as a concern, increasing through the school years, with just 52% of school-leaving respondents falling within the normal drinking category. Bullying also formed a significant theme, with 41% of adolescents reporting to have been bullied at some point; those who reported being bullied were more likely to report depressive symptoms. Of the young adult respondents, 7% reported having made a suicide attempt at some point in their lives, and 51% reported suicidal ideation (these questions were not included in the survey for respondents under 18 years of age).
As Carr (Reference Carr2008) notes, children and adolescents are diagnosed using almost the same criteria that are used with adults, which does not take account of developmental aspects of depressive symptoms. Depression in young people is characterised by low mood or irritability as well as particular cognitive and behavioural features (Carr, Reference Carr2008). Low self worth and rejection sensitivity have been identified as cognitive vulnerabilities in adolescents (McCarty et al. Reference McCarty, Vander Stoep and McCauley2007). Behavioural features such as disturbances of psychomotor function, sleep and appetite are often reported; irritability may be more apparent than low mood in adolescents (Carr, Reference Carr2008). Fatigue and problems with executive function affecting attention, concentration and decision making are also common (Fava, Reference Fava2003; McCarty et al. Reference McCarty, Vander Stoep and McCauley2007).
These features are not a normal part of development, yet there is a tendency, even among health professionals, to think of depressed mood as in some sense ‘normal’ in adolescents and young adults (Iliffe et al. Reference Iliffe, Williams, Fernandez, Kramer, Gledhill and Miller2008). Lilienfeld et al. outline a number of surveys, suggesting that there is a widespread misconception of adolescence as a time of emotional turmoil, and that this misconception is shared by students, parents and health professionals (Lilienfeld et al. Reference Lilienfeld, Lynn, Ruscio and Beyerstein2010). For example, in one survey, 62% of medical students and 58% of nurses agreed that ‘the majority of adolescents show neurotic or antisocial behaviour sometime during adolescence’. As Lilienfeld et al. note, such misconceptions can prevent or delay help-seeking: ‘dismissing some adolescents’ genuine problems as merely a ‘passing phase’ or a manifestation of a normal period of turmoil may result in deeply troubled teens not receiving the psychological assistance they sorely need’ (p. 52). If mental health literacy is to improve, it is important that such misconceptions are identified and challenged; the mass media can play an important role in this process.
Mental health literacy and the media
The term ‘mental health literacy’ is used to describe people’s understanding of mental health issues and their management, and has been defined by Jorm et al. (Reference Jorm, Korten, Jacomb, Christensen, Rodgers and Pollitt1997) as ‘knowledge and beliefs about mental disorders which aid their recognition, management or prevention’ (p. 182). Jorm (Reference Jorm2000) notes several components of mental health literacy, including: the ability to recognise specific disorders; knowledge and beliefs about risk factors, causes and interventions; attitudes that promote help-seeking; and the ability to seek out mental health information.
Information about mental health is now often accessed outside the patient–clinician setting, with most of the information provided by the media. Grilli et al. (Reference Grilli, Ramsay and Minozzi2002) cite the mass media as ‘the leading source of information about important health issues’ (p. 2) and suggest that health services ‘should consider mass media as one of the tools that may encourage the use of effective services and discourage those of unproved effectiveness’ (p. 7). Francis et al. (Reference Francis, Pirkis, Blood, Dunt, Burgess, Morley, Stewart and Putnis2004) found that media reporting of mental health in Australia was generally of good quality, with a focus on policy/programme initiatives and issues related to causes, symptoms and treatment. Although much useful information on mental health is provided by the media, with increasing collaboration between the media and mental health organisations (Anderson, Reference Anderson2003), misinformation and bias are also evident. Exaggerated associations between violence and mental ill health have sometimes (Foster, Reference Foster2006), though not always (Kalucy, Reference Kalucy2011), been noted in the media, for example, and reductionistic portrayals of depression have been documented (Clarke & Gawley, Reference Clarke and Gawley2009; Clarke, Reference Clarke2011). Age biases are also evident. Henson et al. (Reference Henson, Chapman, McLeod, Johnson, McGeechan and Hickie2009) found that the television news portrayed young people with mental illness in a more negative light than was the case for adults, for example. Aside from the way mental health issues are framed, language choices are telling. Use of labels such as ‘patients’ and ‘cases’ or descriptors such as ‘afflicted by’, ‘suffering from’, and ‘victim of’ infer a lack of agency and reduce individual responsibility (Olstead, Reference Olstead2002; Galvin, Reference Galvin2003).
The accuracy of media representations of depression is particularly important, given that mental health literacy among members of the public is sub-optimal. An Australian study by Jorm et al. (Reference Jorm, Korten, Jacomb, Christensen, Rodgers and Pollitt1997) found that many members of the public could not identify conditions such as depression when given unambiguous descriptions. Jorm et al. presented participants with vignettes of a person with symptoms of major depression or schizophrenia and found that 39% of participants correctly identified depression and 27% correctly identified schizophrenia (Jorm et al. Reference Jorm, Korten, Jacomb, Christensen, Rodgers and Pollitt1997). More recently, Reavley and Jorm (Reference Reavley and Jorm2012), using a similar methodology, found evidence of improved mental health literacy, with 74% of participants accurately identifying depression, and fewer people labelling depression as a ‘nervous breakdown’, a ‘psychological problem’ or ‘stress’. However, a quarter of participants could not correctly label the description of depression. An Irish study produced similar findings [Health Service Executive (HSE), 2007], with 74% of respondents accurately identifying depression from a vignette (see also Lawlor et al. Reference Lawlor, Breslin, Renwick, Foley, Mulkerrin, Kinsella, Turner and O’Callaghan2008). However, 13% labelled the same description a ‘nervous breakdown’, and 30% incorrectly labelled a vignette describing schizophrenia as depression. Furthermore, only 5% of respondents accurately estimated the lifetime prevalence of mental health problems.
Despite great accessibility to information through the media, many people report difficulties finding relevant information on mental health. Data from the European Health Literacy Survey showed that 31% of respondents from Ireland reported difficulties in finding information on mental health issues such as depression, with lower income and education level associated with poorer mental health literacy (Doyle et al. Reference Doyle, Cafferkey and Fullam2012). There is significant public investment in research on mental health issues, but biases in reporting by the media means that potentially informative research does not always come to the public’s attention. For example, a study in the UK by Lewison et al. (Reference Lewison, Roe, Wentworth and Szmukler2012) found that media reporting of research on mental health tended to under-represent research on depression and alcohol-related issues, and focus on biological research rather than psychological interventions. Therefore, it would seem that there is still progress to be made in improving access to information about symptoms, treatments and others’ experiences of depression, as well as information about ongoing research. Being able to identify and label the problem has been shown to predict effective help-seeking behaviours (Wright et al. Reference Wright, Jorm and Mackinnon2012), and it is important that the symptoms of depression are recognised and that appropriate supports are identified at an early stage. Wilson et al. (Reference Wilson, Rickwood and Deane2007) found that intention to seek help decreased as depressive symptoms increased, in a sample of children and adolescents. It would seem that the media have a key role to play in improving mental health literacy and it is important to study how mental health issues are portrayed and how relevant information is presented to the public.
In the current study, we explored the portrayal of depression in young people and youth mental health in the Irish newspaper media. We used a quantitative approach to examine recurring words and word patterns used in newspaper reports on young people’s mental health, applying a corpus-based approach to a large set of media reports.
Corpus-based approaches to media texts
In the current study, we used a corpus-based approach to examine word frequency and word patterns across 269 texts. The ‘texts’, here, are machine-readable files, in this case newspaper reports concerned with youth mental health. These make up the corpus, or collection of texts, which is then analysed with a computer using the tools of corpus linguistics. Corpus linguistics is a heterogenous field involving a number of distinct approaches and methods, which have in common the use of computerised tools to analyse large data sets. Such methods have been applied to many topics, including health-related issues, and are often used to examine large sets of media data because, as noted by McEnery and Hardie (Reference McEnery and Hardie2011), they can deal with a data set ‘of a size which defies analysis by hand and eye alone within any reasonable timeframe’ (p. 2). A computer program is used to search through large amounts of text and to identify frequent words, word patterns and other textual features, including keyword frequencies, clusters and concordances of keywords (see McEnery & Hardie, Reference McEnery and Hardie2011 for an overview). Frequency data provide a quantitative overview of word use in the data set – what words are used, how often and how many distinct words are used, for example. Keywords are statistical abstractions from frequency lists (McEnery & Hardie, Reference McEnery and Hardie2011); they are identified by comparing the frequency of a word in the data set with its frequency in a reference corpus, which allows us to determine whether the word occurs with statistically higher or lower frequency than might be expected. Clusters allow us to see what other words co-occur with the high frequency words. Concordances provide a more qualitative analysis; using concordancer software, we can search a set of texts for a target term (word or phrase) and these will be extracted as lines of text so that the context before and after the appearance of the target word can be examined (McEnery & Hardie, Reference McEnery and Hardie2011).
One advantage of the corpus linguistics approach is its objectivity, and it may be particularly useful when approaching a topic such as media coverage of mental health, where the researcher will inevitably have preconceived ideas about biases in media portrayal. As Bondi and Scott (Reference Bondi and Scott2010) note, the researcher ‘cannot switch off his or her intelligence, we cannot see a constellation of stars merely as blobs of light… we cannot see text without seeing meaning’ (p. 45). The computerised keyword analyses will identify words that are statistically more frequent in the newspaper reports compared with a reference corpus, whereas the patterns of clusters of words and concordances can be examined to see what lexical choices are made and how language is used (Gabrielatos & Baker, Reference Gabrielatos and Baker2008; Jaworska & Krishnamurthy, Reference Jaworska and Krishnamurthy2012). The analysis can produce words and word combinations that are expected, but it will also find words and combinations of words that may not have been anticipated. The broad frequency-based analysis provides an overview of recurring words in the newspaper reports, which may reflect recurring themes; a detailed analysis can be used to examine the dominant representations in the reports and the language used to shape them (Tognini-Bonelli, Reference Tognini-Bonelli2001). Of course, the compilation of the corpus is key here; the data set must comprise texts that are representative and that have been selected using appropriate search terms. A biased data set will not generate an objective analysis.
The current study
The current study examined Irish newspaper accounts of depression in young people, focusing on broadsheet publications between the years 2007 and 2011. The aims of the study were to identify, via frequencies and keyword analysis, the recurring themes in Irish newspaper reports on depression in young people, and to examine the textual patterns that repeat in reporting on these themes. We focused here on a quantitative approach to the data set. We also considered whether media reporting around depression and young people mirrors concerns reported in surveys (such as from the My World survey (Headstrong, 2012) by young people in Ireland.
Method
Newspaper articles concerning depression in young people were located online using the LexisNexis News & Business database, ProQuest and the Irish News Archives databases. The search query was limited to the three traditional broadsheet newspapers that have the largest circulations nationally (based on data from the Audit Bureau of Circulations): The Irish Times, The Irish Independent and The Examiner. The search was limited to a 5-year period from January 2007 to December 2011. The search terms were ‘depression’ or ‘depressed’ or ‘mental health’ and ‘youth’ or ‘young people’. The term ‘mental health’ was included in the search because in an initial inspection of newspaper articles we noted that sometimes the word depression was not used; an article might refer instead to mental health and low mood, for example. We limited our search to a 5-year period for two reasons. First, we wanted to compile texts that were representative of current reporting on youth depression and mental health. Second, we aimed to limit texts to several hundred reports, as all would be read to screen out non-relevant reports; for example, a small number of reports referring to ‘depression’ in the economic sense were excluded. All the articles were read and any article found not to be primarily about depression, specifically in young people, was excluded.
The resulting 269 newspaper articles, containing 176 223 words, were analysed using WordSmithTM Tools (Scott, Reference Scott2008), a program used for counting word frequencies, finding word patterns and identifying keywords (words that occur with greater or less frequency than would be expected in comparison with a reference corpus). WordSmithTM was initially used to compile a word list on the basis of frequency. This stage is useful to examine the vocabulary used in the texts and the extent to which word choices recur. The frequency lists generated can be used to examine the proportion of texts using the word, and well as the word’s overall frequency in the corpus. The type–token ratio is also computed here. This provides an estimate of the proportion of distinct words in a sample of text and can be thought of as an approximate measure of richness of vocabulary, or lexical variation; a higher type–token ratio indicates a greater variety of words in the texts. However, smaller text samples will give rise to larger type–token ratios, and thus to compare type–token ratios across different sized text samples, a standardised type–token ratio is computed. Here, WordSmithTM calculates the average type–token ratio based on iterations across 1000 word text samples; the figure returned is a percentage of new types for every n (here 1000) tokens.
In the second stage of the analysis, keywords were identified by analysing the content in comparison with a reference corpus. The significance of the difference in frequency between the word in the wordlist and in the reference corpus is calculated statistically using log-likelihood tests (Dunning, Reference Dunning1993). The log-likelihood score allows a comparison to be made across corpora that will typically be of very different sizes and, according to Leech et al. (Reference Leech, Rayson and Wilson2001, p. 16), shows ‘how significantly characteristic or distinctive of a given variety of language a word is, when its usage in that variety is compared with its usage in another’. Keywords can then be grouped according to specific topics. Baker (Reference Baker2006) suggests using a large reference corpus, which is representative of language in general and can act as ‘a good benchmark of what is ‘normal’ in language’ (p. 43). Here, we used the British National Corpus (BNC), a 100 million word collection of samples of mainly (90%) written language taken from a wide variety of sources, which is representative of British English from the later 20th century. The written BNC sources include regional and national newspapers, periodicals and journals, school and university essays, academic books and popular fiction.
Finally, textual patterns were explored in the context of word clusters (such as patterns of repeated phraseology) and collocates (i.e. other words that tend to occur in the neighbourhood of the search word).
Results
Table 1 presents summary statistics for each year sampled. The average word length of the reports was 644.2 words (s.d.=431.4). There were 11 267 distinct words (types) in the corpus, with a standardised type–token ratio of 42.3. This type–token ratio is around the average for written texts (Baker, Reference Baker2006) and is consistent with higher type–token ratios found for informational styles of press reporting (Biber, Reference Biber1988) and is similar to other specialist reporting (Reference MusacchioMusacchio, 2006). A lower ratio might have been expected if, for example, similar reports were simply repeated again and again in the media.
Table 1 Summary statistics for the newspaper reports of depression in young people over the 5-year period
a Computed across the complete data set (2007–2011).
Word frequency
Table 2 shows the 25 most frequent words used in the corpus, omitting function words (e.g. the, a, to) and gives a sense of the topic under discussion. We use the term ‘content word’ here to refer to any noun, verb, adjective or adverb; content words, in the linguistic sense, are a more restricted category of word, excluding some verbs, for example.
Table 2 The 25 most frequent content words
While most of the frequent words shown in Table 2 simply reflect the topic that was searched (e.g. young people, mental health), the inclusion of the word ‘suicide’ among the most frequent items suggests that reporting of depression in young people and reporting of suicide are associated. In fact, the word suicide occurred more often than the search term depression, at 662 mentions versus 589 (see Table 2), and was mentioned in almost half of the newspaper reports (45%), whereas the word depression appeared less often overall, but occurred in 86% of reports. The word ‘children’ also occurs with high frequency. Table 3 shows the content words that appeared in the majority of the reports sampled (i.e. above 50% of texts).
Table 3 The content words featured in most of the texts
The high frequency of words such as ‘depression’, ‘young’, ‘people’, ‘mental’ and ‘health’ reflect the search terms used to compile the data set. Other words also occur in a majority of texts. The words ‘year’ and ‘years’, for example, tended to occur, describing ages and school years (e.g., ‘in 5th year’, ‘16 year old’, etc.), and also time markers such as ‘earlier this year’, ‘previous year’, ‘in the past year’, phrases often used when reporting statistics. The word ‘time’ tended to occur in phrases such as ‘at the time’, ‘the first time’, ‘all the time’ and ‘at a time when…’. The word ‘children’ occurred most frequently in the cluster ‘children and adolescents’, with ‘children and young people’ and ‘children and teenagers’ also used.
Words relating to symptoms and emotional terms were less frequent in the data set. Of these, ‘anxiety’ was the most frequent term used, with 125 mentions across the corpus. Terms related to sadness, loneliness and unhappiness were less frequently found, and references to other characteristics of depression (e.g. anhedonia, problems with concentration, sleeping, appetite, etc.) were infrequent. For example, sadness (as ‘sad’ or ‘sadness’) occurred 41 times, with ‘unhappy’ and ‘unhappiness’ mentioned 23 times. Sleep (including ‘insomnia’) is mentioned 37 times. The extent to which words occurred with greater than expected frequency is explored in the next section, by analysing keywords.
Keywords
Measuring keyness allows a more focused analysis than a simple frequency count. Keyness compares the frequency of a search word in the corpus with its frequency in a large reference corpus and involves computing the statistical probability that the frequency of the word deviates from the estimated frequency. Therefore, a high keyness value acts as a good thematic indicator and gives a sense of ‘aboutness’ (Scott, Reference Scott2000) Keywords, in this context, are words that occur with greater or less frequency than would be expected on the basis of a large reference corpus. The significance of the difference in frequency between the word in the wordlist and in the reference corpus is calculated statistically using log-likelihood tests (Dunning, Reference Dunning1993).
Using the BNC as the reference corpus, we found that a total of 890 keywords (or 8% of the tokens) occurred more (791), or less (99), frequently than would be expected on the basis of the reference corpus, at p<0.00001. Table 4 shows the 25 most salient (i.e. highest keyness scores) keywords. The keywords identified include many words that might be expected, given our search terms: depression, mental health and young people/youth. Also included are some predictably unusual words when comparing our data set with a reference corpus: HSE (the authority that delivers health and social services in Ireland), Dublin and Ireland, for example, as well as ‘psychiatric’ and ‘disorder’.
Table 4 Most salient keywords in the youth depression corpus
HSE, Health Service Executive.
All are statistically significant at p<0.00001.
The term ‘psychiatric’ occurred most often along with ‘unit’, ‘services’ or ‘hospitals’. ‘Disorder’ appeared in a range of contexts, but the term ‘illness’ was used twice as often. The appearance of the term ‘psychiatric’ in the list also reflects the prominence of the profession in reports on depression in young people. The term ‘psychiatrist(s)’ (n=78) was used more often than as ‘psychologist/(s)’ (n=43), ‘counsellor/(s)’ (n=36), and was also more frequent than the terms ‘doctor/(s)’ (n=66) and ‘GP/(s)’ (n=42).
The frequency of the word ‘children’ (572 mentions, in 53% of articles) is noteworthy, considering that the search was more generally targeted at reports concerning ‘young people’. Reference to ‘parents’ was made in 38% of the articles and often occurred in the context of parents as potential supports.
Three key terms may be useful thematic indicators, in so far as they emerge as key, even though they are not directly related to the original search terms used to collect the text samples: suicide/suicidal, alcohol and bullying. The word ‘suicide’ appeared 662 times and was found in 45% of the newspaper reports over the 5-year period (see Tables 2 and 3), suggesting a strong association between reporting on depression in young people and mention of suicide. ‘Alcohol’ was also strongly associated with the discussion on depression in young people, with 203 mentions across 23% of articles sampled, referring to the misuse of alcohol. The term ‘bullying’ also emerged as key, although less frequent, appearing 95 times in 14% of the newspaper articles. Analysis of the word clusters showed that bullying was generally portrayed as having a causal role in the onset of depression in a young person.
Word patterns
Tables 2 and 3 highlight the frequent use of the words ‘depression’ and ‘suicide’ in the corpus, but to examine how these words are used we must consider their use in context. To examine patterns beyond the single word, frequency lists of word clusters were derived using WordSmithTM for words found in the corpus related to ‘depression’ and ‘suicide’. For example, ‘depression’ was grouped with ‘depressed’, ‘depressive’, ‘depressant’, and ‘suicide’ with ‘suicidal’ and ‘suicidology’. Collocates (that is sequences of words that occur in proximity together) were computed by searching five words to the left and five words to the right of each target word, with a cluster size of three word sequences. Clusters that occurred at least 15 times are shown in Table 5. The most common collocates of depression and related terms (e.g. depressed) were variations of the verb ‘suffer’ (‘suffering with’, ‘suffer from’, etc.), with 68 mentions of depression terms in the immediate context of ‘suffering’, and 174 uses of the stem ‘suffer’ in total. (The term ‘sufferer’ was rare, however, with nine mentions in total.)
Table 5 Clusters of the search terms ‘depression’ and ‘suicide’ with collocates for clusters that reoccurred 15 times or more
The collocates for ‘suicide’ showed few distinct patterns (see Table 5), but concordances were informative. A concordance is a list of occurrences for the search term presented within its context, typically presented as ‘key words in context’ by concordancing software. Concordances introduce a qualitative element to the analysis and can form the basis of a critical discourse analysis (Gabrielatos & Baker, Reference Gabrielatos and Baker2008). The search terms were examined in their context, showing eight words before and after the search term. The concordances for terms related to ‘suffering from/with depression’ showed that this terminology was often used when presenting statistics, particularly prevalence rates, with 38% of the 68 mentions providing numerical information about mental health. The concordances for ‘suicide’ showed that the word often occurred in the context of ‘attempt’, with 44 mentions of ‘suicide attempts’ or ‘attempted suicides’ in the data set. The term ‘commit suicide’ was also evident, although less frequent, with 15 mentions across its various types (‘committed suicide’, ‘commit suicide’, etc.), showing that this term continues to be used in newspaper reporting, albeit infrequently, 20 years after the decriminalisation of suicide in Ireland.
Discussion
The themes of suicide, alcohol use and, to a lesser extent, bullying noted in the analysis of newspaper reports mirror some of the concerns noted in surveys of young people in Ireland (Headstrong, 2012). The inclusion of the term ‘suicide’ among the most frequent items is noteworthy and shows how reporting on depression in young people and suicide often co-occur. The word ‘suicide’ occurred more often than the search term ‘depression’, with just under half of the newspaper reports on youth depression explicitly mentioning the term ‘suicide’ or a related term (e.g. suicidal). The regular discussion of suicide in the context of depression in young people, although important, could reinforce the misconception that only people who are depressed will try to take their own lives (Cullen, Reference Cullen2006) or that all individuals with depression are at risk for suicide. Further investigation of the ways in which suicide and depression are reported would seem to be warranted.
The frequency of use of the word ‘children’ is also noteworthy, considering that the search was more generally targeted at reports concerning ‘young people’. This suggests that children are increasingly being considered in discourse on depression and young people. Reference to ‘parents’ was made in a substantial proportion of the articles (38%), and generally occurred in the context of parents as potential supports. Many reports seemed to be aimed at parents in particular, consistent with the demographic of the newspapers sampled.
Although other studies internationally have noted a focus on biological and medical themes in media reporting on depression in young people (Clarke, Reference Clarke2011), this focus is less apparent here. Results suggest less emphasis in these Irish newspapers on the various symptoms (e.g. mentions of sadness, sleep problems, appetite issues) and treatment options available, which might make it difficult for parents to identify whether their child is depressed. Instead, the Irish newspapers that we sampled frequently mention factors associated with depression for the young person (e.g. bullying and alcohol use) and the potential consequences of untreated depression (e.g. suicide). Perhaps more emphasis on symptoms and treatment might encourage help-seeking and improve mental health literacy in young people and the parents of young people. The relative lack of emphasis, based on frequency, on symptomology contrasts with the analyses of North American media coverage (Clarke, Reference Clarke2011), but may have emerged using a qualitative approach. Our results must be interpreted in the context of the frequency-based methodology, a limitation that is discussed in more detail below.
The most common collocates of depression and related terms (e.g. depressed) were variations of the verb ‘suffer’ (‘suffering with’, ‘suffer from’, etc.). The recurrent use of the verb ‘suffer’ (‘suffering with’, ‘suffer from’, etc.) in reporting on depression suggests a lack of agency in recovery from depression and reinforces the notion of the passive role of a young person experiencing depression (Olstead, Reference Olstead2002; Galvin, Reference Galvin2003). The ready association of ‘suffer’ with depression also conveys a medical condition, reinforcing discourse patterns associated with illness. The relative infrequency of terms relating to signs of depression, or emotional distress, seems at odds with this focus. The prominence of the psychiatrist among other professions (e.g. GPs, psychologists) is also relevant here and could be taken as suggesting that depression always requires intervention from a psychiatrist; future research might examine the context in which expert voices are reported, and the textual patterns that occur in such cases.
One advantage of the corpus linguistics approach lies in its objectivity. Word frequencies and keywords are computed by the software programs and researcher biases are reduced; this is particularly useful when approaching a topic such as media coverage of mental health, where the researcher will inevitably bring preconceived ideas to the analysis. The analysis is data driven and largely quantitative; it is inductive and directs the researcher’s attention towards recurring textual patterns that might otherwise have gone unnoticed (Gabrielatos & Baker, Reference Gabrielatos and Baker2008). However, the method used in the current study has limitations that must be considered.
First, because we used a purely quantitative approach here, we were focused solely on the numbers of mentions. The extent to which the frequency of word use and salience (or focus in an article) is linked is debatable, however, and our method of analysis will have been better suited to constructs with few synonyms. There are many ways to describe low mood, for example, and these various descriptions will not be captured within a single count of frequency. Perhaps this is why emotion-based terms appeared with lower frequency than might be expected; there may be greater variety used in describing emotions, which would lead to their being underestimated in a frequency count.
Second, the choice of reference corpus must be considered. Keywords emerge in comparison with a large-scale reference corpus and therefore many health-related words, which will occur across a wider variety of written texts, will not surface among the most salient keywords in the analysis of depression. The choice of the reference corpus is important; ideally, a health-related corpus would have been used as the norm corpus, but as none were available, the more general BNC was used here, and the data must be interpreted with this in mind. It is therefore important to consider frequency data alongside keywords.
Third, frequencies do not take into account the social, cultural or discursive context of the data (Wilson et al. Reference Wilson, Rickwood and Deane2007). Sometimes words that are not used more, or less, than might be expected are also of interest, as are the reasons for their use or omission. Frequencies do not tell us why certain textual patterns are, or are not, found (Gabrielatos & Baker, Reference Gabrielatos and Baker2008). The method, as used here, does not tell us anything about the directionality of associations between words. For example, in the press coverage of depression in young people, alcohol is among the factors mentioned with greater average frequency than expected. But what kind of discussion around alcohol is presented? Is alcohol presented as a cause or as a consequence of depression? Similarly, parenting or issues related to parents are mentioned, often in the context of supports but sometimes in other contexts. These differences in use need to be examined.
Finally, a report might be written about a topic without using a particular word. ‘Suicide’, for example, might be written about without use of that term; a purely quantitative analysis may exclude such reports. While collocation profiles can be used to examine discourse patterns and salient ideas associated with the search term or topic of interest, an approach focusing on the use of language would allow a more contextualised analysis. Discourse analytic approaches have been used effectively to address a range of issues in health care and it has been argued that a corpus-based approach can be used alongside a critical discourse analysis to identify linguistic patterns and also address why they might have occurred in the way they did (Gabrielatos & Baker, Reference Gabrielatos and Baker2008).
Notwithstanding these limitations, the method used here would seem to have potential as the basis for a quantitative audit of reporting around youth depression and the potential impact of reporting styles on readers. This quantitative approach could be supplemented with a more qualitative method, allowing the data to be examined in a broader context, an approach that might be used in future research.
Acknowledgements
This study was conducted while the first author was participating in an undergraduate summer research programme funded by the National University of Ireland Maynooth and supervised by the second author. The authors gratefully acknowledge this funding and the constructive comments of three anonymous reviewers.
