25 Analysing text and speech: content and discourse analysis Fran Tonkiss in C.

Seale (ed) (2004) Researching Society and Culture. London: Sage (revised edition). pp. 367-92.

Contents Content analysis Coding content Content analysis and interpretation Discourse analysis What is discourse? Discourse in a social context Doing discourse analysis Selecting and approaching data Sorting, coding and analysing data Presenting the analysis Further reading

This chapter considers the use of written and spoken texts as the basis for social and cultural research. It focuses on two contrasting methods of textual analysis: content analysis and discourse analysis. These techniques provide a critical contrast in representing quantitative and qualitative approaches to the study of textual data. However, they share a common interest in the use of language in social contexts. Each approach provides insights into the way speech and texts help to shape and reproduce social meanings and forms of knowledge. Methods of textual analysis are relevant to a range of research subjects, in such fields as sociology, psychology, media and communications, history, politics and social policy, cultural studies, sociolinguistics, law, education, management and organization studies. The discussion begins by outlining content analysis in the study of textual data. I look at different ways of approaching textual content, how this method can be used in conjunction with other forms of analysis, and consider key problems of validity. The second part of the discussion is concerned with discourse analysis. I define the term discourse and explore discourse analysis in relation to three core stages of the research process: collecting data; coding and analysing data; and presenting the analysis. CONTENT ANALYSIS Content analysis is a quantitative method for studying textual data. It seeks to analyse texts in terms of the presence and frequency of specific terms, narratives or concepts. This can involve counting items (specific words or categories) or measuring the number of lines or amount of space given to different themes. Content analysis has a long pedigree in psychology and in communications research (see Berelson 1952). It is still frequently used in the analysis of media texts such as newspaper articles, radio and television reports and is closely associated with the study of visual content in photographic, film and television images (see chapter ?). In this discussion, however, I am concerned with the use of content analysis in relation to speech and written texts.

The principal strength of this approach lies in the clear and systematic study of textual content as a basis for analysis and interpretation. Content analysis is the primary method used for large-scale and comparative study of textual data. It potentially has a high degree of validity and reliability in terms of precise sampling, providing clear empirical evidence for research findings, and in allowing for replication and generalization. In grounding analysis on empirical content rather than on interpretive argument, furthermore, this can be seen as one of the most objective methods for the study of texts. Content analysis, however, while it shares many of the advantages of quantitative social research (including, it might be said, certain claims on academic legitimacy), has been subject to a number of criticisms. Chief amongst these is the objection that such analysis is concerned simply with crass content: with what is said rather than with how it is said; with the description of texts rather than their interpretation, meanings or effects. In this respect, debates over content analysis bear on a central methodological issue in social and cultural research. Content analysis can be placed within a broadly empiricist and positivist tradition of inquiry, concerned with the analysis of observable features or facts, rather than with less observable and often highly subjective questions of meaning. At the same time, in its focus on texts and speech content analysis is clearly interested in the production and reproduction of meaning. The method therefore raises quite sharply the question of whether a scientific model of social research can be appropriate to the study of social and cultural objects that are in large part defined by the meanings they hold for social actors. In what follows I examine how content analysis can help to address these questions of qualitative meaning and potential readings. Content analysis adopts a fairly standard model of research design. Having formulated a research topic, the researcher defines the relevant population of interest and then draws an appropriate sample from it. In this case, though, the research will be based on a sample of texts rather than a sample of people. This might come from a number of sources, including political debates and speeches, media texts, policy and legal documents, archival sources and other historical documents, tourist guides, publicity literature, press statements, and so on. In many instances the potential sample will be vast, given the scale on which texts are produced and circulated. Random sampling is a challenging task even for researchers working on a large scale, and often is not feasible for researchers involved in smaller studies. It is important, therefore, for researchers to delimit their sample very clearly for example, by setting time limits (television coverage over a one-week period; three issues of a magazine each year for twenty years), or defining the type of text as precisely as possible (a particular piece of legislation, presidential speeches or party political broadcasts, newspaper editorials or front pages). When working with purposive samples of these kinds, the researcher needs to make very clear the rationale for their selection of texts. Moreover, this selection should aim to produce a sample that is relevant to the research problem, representative of the field of interest, and manageable for the researcher to analyse in detail. Coding content Counting and analysis of textual data can proceed in various ways. A common starting-point is to define categories of analysis and to code the data using these categories. The categories may be pre-set by the researcher in advance of reading the data, or they may be based on an initial reading of the texts. In many cases, coding categories emerge from a combination of these two processes - some will be pre-set to reflect the aims and the theoretical framing of the research, further categories will arise from detailed reading and coding of textual content. This stage of the research requires intensive work to ensure that coding categories will capture the content of the texts in ways that are clear (reducing ambiguity and overlap) and

exhaustive (including all relevant content). A key aim in constructing and applying codes is to limit the margin for interpretation on the part of individual researchers. The reliability of the coding process is an important consideration in content analysis: will different researchers, that is, code the data in the same way? Content analysts frequently use tests of inter-coder reliability or inter-rater agreement to ensure that codes are matched to content in a consistent manner. At the pilot stage of the research, and as a further check in the course of data analysis, a number of researchers will code sample texts using the same set of coding categories and guidelines for their use. The degree of agreement between researchers acts as a test of the reliability of the content analysis as a whole (see Neuendorf 2002 for a more detailed discussion of validity and reliability in content analysis). An alternative starting-point for content analysis is for the researcher to compile a simple key-word count. Rather than beginning with a coding frame, the researcher begins their analysis with a frequency count of the main items in the text. This often is done using computer searching and coding of texts. There is a variety of software available to assist in content analysis, which will analyse texts in a number of ways: including key-word lists and word frequencies, identifying main ideas, analysing patterns of word use, comparing vocabulary between texts, and full concordances that list and count all words that appear in a text. More advanced software uses theoretical frameworks derived from linguistics and psychology to code and analyse textual data. These computer-assisted approaches are extremely useful when researchers are dealing with a large amount of material the initial reading of the texts is done by computer, measuring the frequency of specific words and providing a broad picture of the texts as a basis for analysis. In this way they organize a large body of data for the researcher in a manageable form. Moreover this approach can be seen to minimize researcher bias by grounding the initial analysis in the manifest content of the texts that is, in the content as it has been written or spoken by the producer, rather than as it has been read by the researcher. Such techniques are not limited to researchers working on large-scale studies or with extensive resources. More and more textual material is available in electronic form over the internet and in electronic archives. Newspaper articles, policy documents and political speeches are three key resources that researchers can retrieve at little or no cost, and internet material in general is a growing field for analysis. Even where researchers do not have access to software packages, they can replicate basic searching and counting exercises as a starting-point for their analysis of printed or electronic texts. Key word counts are commonly used, for example, to give a broad-brush account of politicians speeches. In the first presidential debate in the run-up to the 2000 election, for instance, George Bush used the words abortion eight times, American seventeen times, and taxes thirtythree times, whereas Al Gore said abortion once, America twice, and taxes twenty-five times (Beard and Payack 2000). While these counts might look suggestive, they are of course only an analysis in the simplest sense in terms of sorting and enumerating the data. The context, the meaning and the effect of the speeches remain open to interpretation. A basic counting exercise looking for the term abortion in this debate, for instance, would not capture either the presence or the larger meaning of Bushs use of pro-life as against Gores use of anti-choice. Content analysis and interpretation This example indicates how the value of content analysis to social and cultural research tends to go beyond simple processes of counting to take in issues of meaning and context. This can work by combining content analysis with other methods, or by using it as a framework for the more interpretive analysis of texts. Two examples might illustrate this analytic potential, as well as providing versions of the alternative approaches to counting and coding data described above.

In their study of media coverage of the Israel/Palestinian conflict, Philo et al. (2003; see also Philo 2001) combined content analysis with focus group and survey methods. They were interested not only in how the conflict was covered in the news, but also in how the news was read and understood. The content analysis centred on television news reports of the Intifada that began in September 2000, based on a clearly defined sample. This was drawn from major news programmes (lunchtime, early evening and late night bulletins) on two British television channels with the largest news audiences (BBC1 and ITN) over an eighteen-day period from the start of the Intifada. The sample yielded eighty-nine reports, which were recorded and transcribed in full. Philo et al. then set up a number of themes to code the content of the news stories: including fighting/violence, origins of the crisis, and peace negotiations. News content was then analysed in terms of how much coverage was given to these different themes. Here the researchers used the measurement of space as the basis for their content analysis. They found that, of 3536 lines of text in the reports, only 17 lines included any account of the history or causes of the conflict. This is in itself a striking finding, based on a fairly straightforward coding and counting exercise: less than 0.5% of the mainstream news coverage in this sample contained information on the background to the conflict. Philos research is particularly interesting, however, in taking the analysis beyond the counting of manifest content. The content analysis formed part of a larger study that involved audience focus groups with eighty-five people, and survey questionnaires conducted with three hundred 17-22 year-olds in Scotland. The aim of these other parts of the study was to examine where people got their information about the Israel/Palestinian conflict, and to explore their understanding and perceptions of what was going on. The survey of three hundred young people, involving respondents who said their main source of information was television news, found that many of those questioned knew very little about the origins or character of the conflict. For example, 71% of those surveyed did not know who were settlers and who was occupying the occupied territories; 11% thought that the settlers were Palestinian and the occupation a Palestinian occupation, while only 9% knew that it was an Israeli occupation. This multi-method approach combined quantitative (content and survey) analysis with qualitative (focus group) analysis so as to examine the news texts themselves and to explore questions of audience interpretation and understanding. While this does not provide a basis for asserting that the media content causes individuals to perceive these issues or events in certain ways, the research does show that a low level of understanding on the part of sample audience groups is paralleled by the limited amount of background information being provided by mainstream media content. In this context, the quantitative analysis of textual data supplements (and, it can be argued, supports) the analysis of audience members reading of news media. Content analysis, then, forms part of a research design that seeks to address the complex problem of linking media messages and audience effects. A different approach to content analysis is illustrated by Seales (2002) work on news reports of people with cancer. In this research study, Seale was concerned with how news coverage might differ in its treatment of men and women with cancer. His analysis was based on a sample of 358 articles published in the English language press during a one-week period. In contrast to Philo - who pre-selected categories or themes to code his data - Seale began his content analysis by running a word concordance on the media texts. This is an operation, using computer software, which lists all the words that appear in the texts in order of the frequency with which they appear. The text is reorganized so that it reads not as a narrative but as a count of individual words. Such computer searching sorts the text in terms of its quantitative content without the researcher imposing any categories on the initial analysis. Seale used this basic analysis to identify the different themes and uses of language that were evident in the texts. For example, Seale noted the frequency with which the

language of emotions occurred in the news content. He went on to code the texts in terms of the presence of a range of items indicating different states of emotion. Such items as joking or being humorous, feeling supported by others, accepting ones illness or showing a fighting spirit were coded as positive or good emotions. References to fear or anxiety, anger, isolation, misery or depression were coded as negative or bad emotions. The content analysis suggested that the emotions of women with cancer were discussed in the news texts more frequently than those of men; this was particularly the case for bad emotions and especially for feelings of fear and anxiety (Seale 2002: 110). It is important to note at this point the degree of interpretive work being done by the researcher in assigning content to different codes. How is one to assess, for example, the difference between showing a fighting spirit (a good emotion) and anger (a bad emotion)? These categories are not necessarily clear-cut, and their different meanings will in part come from the context in which they are used. A quantitative content analysis in this way is used together with a more qualitative analysis of textual meaning. By looking not only at the number of times different emotional states are mentioned, but also the ways in which they are talked about, Seale develops an interpretive analysis of how gender inflected the representation of cancer in the press. He argues that the texts tend to reproduce ideas of gender difference in stressing womens skills in the emotional labour of self-transformation, developing personal resources to deal with their situation, while for men cancer is more commonly portrayed as a test of pre-existing character (2002: 107). The analysis of media content is informed by theoretical perspectives on gender and selfhood, and by the interpretation of textual meanings. This raises critical questions about the objectivity of content analysis. As discussed earlier, the validity of this method tends to rest on claims about the neutrality of the researchers role and the objectivity of the results. There are a number of points to be made here. Firstly, the objectivity of content analysis can be questioned in that qualitative judgements often underlie the definition of coding categories. Secondly, even where researchers use more neutral key word counts, content analysis will tend to reproduce a repertoire of dominant themes or narratives (for example, in political and media discourse), and therefore can help to reinforce the power of these categories. Thirdly, content analysis can be seen to assume a shared world of meaning that is accessible in the content of texts. It is not at all clear, however, that statements about content tell us very much about either producers intentions or consumers interpretations. It needs to be asked how far the categories used in content analysis reflect the intentions of the author, the understanding of the reader, or merely the perceptions of the analyst. One cannot simply read from intentions via content to audience readings: simply, we do not necessarily all read the news in the same way. In these respects, content analysis encounters the problem of how meaning operates that runs through the analysis of social texts.

DISCOURSE ANALYSIS The organization of textual meaning is central to discourse analysis. This qualitative approach to textual analysis can sometimes seem a difficult method to pin down because it is used in different ways within different fields (see Wetherell, et al. 2001a, Hammersley 2002). While its origins lie most firmly in the disciplines of linguistics and social psychology, the method has been widely taken up within sociology, media and communications, cultural studies, socio-legal studies, education, management studies, history, politics and social policy. In this discussion I am chiefly concerned with these broadly social approaches to textual analysis, focusing on how social categories, knowledges and relations are shaped by discourse. Approaches to discourse analysis within psychology and linguistics, in contrast, tend to focus more

closely on the rhetorical and technical use of language (see Billig 1987; van Dijk 1997; Fairclough 2003). Discourse analysis takes its place within a larger body of social and cultural research that is concerned with the production of meaning through talk and texts. As such, it has affinities with semiotics, which is primarily concerned with visual texts (see Chapter ?), and with conversation analysis (see Chapter ?). While approaches to discourse analysis vary, they share a common understanding of language as an object of inquiry. To the discourse analyst, language is not simply a neutral medium for communicating information or reporting on events, but a domain in which peoples knowledge of the social world is actively shaped. Anyone who has been in an argument with a skilled or slippery debater will be aware of the way that language can be used to compel certain conclusions, to establish certain claims and to deny others. Discourse analysis involves a perspective on language that sees this not as reflecting reality in a transparent or straightforward way, but as constructing and organizing the terms in which we understand that social reality. Discourse analysts are interested in language and texts as sites in which social meanings are formed and reproduced, social identities are shaped, and social facts are secured. WHAT IS DISCOURSE? Discourse can refer to a single utterance or speech act (from a fragment of talk to a private conversation to a political speech) or to a systematic ordering of language involving certain rules, terminology and conventions (as in legal discourse). This more systematic approach allows researchers to analyse how discourses inscribe specific ways of speaking and understanding. Viewed in this way, a discourse is a group of statements which provide a language for talking about i.e. a way of representing a particular kind of knowledge about a topic (Hall 1992: 290). Such an approach is often associated with the work of the French thinker Michel Foucault, and his interest in how discourses help to produce the very categories, facts and objects that they claim to describe (Foucault 1972: 49; see also Chapter ?). Discourse, in this sense, does not refer simply to language or speech, but to the way language works to organize fields of knowledge and practice. Following Foucault one might ask, for instance, how our understanding of sexuality is shaped by various moral, medical, legal and psychological discourses. How is the concept of deviance (for example, `mad' or `delinquent' behaviour) defined and talked about within discourses of psychiatry or criminology? Moreover we can go on to ask: how are these discursive constructions linked to social practices, to social institutions, and to the operation of social power? A good example of this kind of approach is Bells (1993) use of discourse analysis to examine how the crime of incest is constituted under English and Scots law. Incest was criminalized in English law in 1908; while it has been criminal in Scotland since 1567, the law was modernized in 1986. Bell bases her analysis on the parliamentary debates surrounding both pieces of legislation (Bell 1993: 126-7). Her interest is in how incest is defined as a criminal act in ways that draw on particular forms of expertise and evidence, securing certain arguments and excluding others. Bell identifies three key knowledges that shape the legal discourse about incest. The first of these concerns issues of health - articulated in terms of the dangers of inbreeding in the 1908 debates and genetic risks in the 1980s (Bell 1993: 130-31). While medical or scientific arguments appear in both debates, Bell points out that they are in themselves insufficient to define the offence of incest. For example they do not explain why incest would be wrong if there was no chance of conception, they focus on the possible consequence of the act rather than the act itself. At the same time they effectively define incest as a problem in rather limited ways, as referring only to sexual relations between men and fertile women, and to blood relatives rather than adoptive or step-family. The victim of incest, furthermore, is understood to be

the potential offspring, rather than either of the parties directly concerned. In these terms Bell examines how medical knowledges shape the discourse on incest to produce particular definitions of the problem itself and the subjects it involves. The scientific discourse of genetic harm is supplemented by a second body of knowledges that construct the offence of incest in terms of sexual, psychological or child abuse. Here the speakers in the parliamentary debates are seen to draw on discourses of child protection, social welfare and psychology. Incest is constructed as wrong on the basis of mental harm, coercion and violence, defined in terms of power relations within the family. The victim of incest is represented within these discourses as children or young women who are vulnerable to (especially male) adults. While such a conception of incest might be seen as more in keeping with current understandings, Bell does not claim to assess the relative truth of these competing accounts (1993: 129). Rather she is concerned with the differing ways in which they produce incest as a legal fact, defining the problem and the victim in various terms. The third key frame within which incest is constructed in these debates is as a threat to the family as a social institution. This is particularly important for the inclusion of adoptive and step-relations in the definition of incest under Scots law, in contrast to the scientific arguments seen earlier. Here, the offence of incest is construed in terms of a breach of trust within the family, and as violating the family as a social bond rather than simply a genetic one. Such an understanding involves an extended notion of the family unit, as well as its importance to a wider social and moral order. Bells analysis is interesting in showing how the category of incest, while often naturalized as a primary human taboo, can be understood as a legal artifact molded by various discourses. English and Scots law define incest differently; moreover the parliamentary debates that inform these laws draw on contrasting and sometimes conflicting knowledges which go beyond the legal sphere. Incest is constructed as a legal fact via discourses of medical science, psychology, child protection, social welfare, the family and moral order. It follows that legal discourse is neither pure nor disinterested, but is shaped by wider networks of language, knowledge and power. This point goes beyond semantics: discourse ways of speaking about and understanding an issue is important here because it helps to shape the practical ways that people and institutions define and respond to given problems. DISCOURSE IN A SOCIAL CONTEXT Perhaps the easiest way to think about discourses as linking language, knowledge and power is to take the model of `expert' languages. Doctors, for example, do not simply draw on their practical training when doing their job; they also draw on a medical language that allows them to identify symptoms, make diagnoses and prescribe remedies. This language is not readily available to people who are not medically trained. Such an expert language has a number of important effects: it marks out a field of knowledge, it confers membership, and it bestows authority. Firstly, medical discourse establishes a distinct sphere of expertise, setting out the domain of medical knowledge and the issues with which it is concerned. In recent years, for example, there has been a debate as to whether chronic fatigue syndrome or ME should be considered as primarily a physical or a psychological problem, and, to an extent, whether such a condition can be said to exist at all. One way in which this debate plays out is in the language used to describe the condition. The term `myalgic encephalomyelitis' clearly medicalizes the condition, while a term such as `yuppie flu' does not. The use of language plays a notable part in arguments for recognizing a condition as a `proper' illness: as a valid object of medical expertise and a suitable case for medical treatment. Medical discourse in this sense helps to delimit a distinct field of knowledge, and to exclude certain facts or claims from this field.

Secondly, medical discourse confers membership in allowing health professionals to communicate with each other in coherent and consistent ways. Language in this sense represents a form of expert knowledge that professionals draw on in their everyday practice and reproduce in their interactions. The internal conventions and rules of medical discourse act as a way of socializing individuals into the medical professions, and enabling them to operate competently within them. In this respect, discourse has a role to play in the institutional organization of medical knowledge and its professional culture. Thirdly, medical discourse authorizes certain speakers and statements. Doctors' authority is perhaps most obviously expressed by their access to an expert language from which most of their patients are excluded. On an everyday level, while we may at times be frustrated by the use of medical language to describe our symptoms, we may also be reassured that our doctor is an authority on these matters. More generally, medical authority is asserted in the use of expert discourses to dismiss competing accounts, such as those associated with homeopathic and alternative remedies. Expert languages provide an obvious and a very fruitful area for research, however discourse analysis is by no means confined to this domain. Discourse analysts might study formal policy or parliamentary discourse (as we have seen with Bell 1993), but also the popular discourses used in politicians speeches and manifestos (Fairclough 2000). A researcher interested in the discursive construction of race and racism might analyse political speech and debates (for example Smith 1994), the news media (van Dijk 2000; see also van Dijk 2002), or the everyday language of racism in talk and texts (Wetherell and Potter 1993). In all cases, the analyst is concerned with examining the way that specific forms of text and speech produce their versions of a social issue, problem or context. DOING DISCOURSE ANALYSIS It is difficult to formalize any standard approach to discourse analysis. This is partly because of the variety of frameworks adopted by different researchers, partly because the process tends to be data-driven. However, while there are no strict rules of method for analysing discourse, it is possible to isolate certain core themes and useful techniques which may be adapted to different research contexts. In the discussion that follows I consider some of these in terms of three key stages of the research process: selecting and approaching data; sorting, coding and analysing data; and presenting the analysis.

Selecting and approaching data I have stressed the `special' character of discourse analysis as a method of research its distinctive approach to language and its resistance to formulaic rules of method. However, the discourse analyst is faced with a common set of questions that arise within any research process. What is the research about? What are my data? How will I select and gather the data? How will I handle and analyse the data? How will I present my findings? Formulating a research problem can be one of the most difficult moments in social research. Sometimes it can seem like a very artificial exercise - qualitative research frequently is data-led and the researcher cannot be certain quite what the research will be about until they have begun their analysis (see also Chapter ?). Discourse analysis often adopts a wait-and-see attitude to what the data throw up. This is underlined by the fact that this form of research is not so much looking for answers to specific problems (`What are the causes of juvenile crime?'), as looking at the way

both the problem, and possible solutions, are constructed (`How is juvenile crime explained and understood within current political discourse?'). Explanations of juvenile crime might drawn on accounts of moral decline, poor parenting, the absence of positive role models, inadequate schooling, poverty, lack of prospects, adolescent rebelliousness, and so on. This is not to say that the issue juvenile crime does not exist or has no meaning, but asserts that social actors make sense of this reality in various, often conflicting ways. If a dominant understanding of juvenile crime rests on discourses of poor parenting, for example, it is likely that the problem will be tacked in a different way than if it was commonly understood in terms of a discourse of material deprivation. As with other forms of social and cultural research, discourse analysis often begins with a broad even vague interest in a certain area of social life. The way this broad interest becomes a feasible research topic is strongly linked to the choice of research methods. You may, for example, be interested in researching the topic of immigration. There are different ways of approaching such a topic, and these will influence how the research problem is defined. You might analyse statistical data relating to the number of people entering a country in each year, their countries of origin, and patterns of change over time. Or you could select a sample of people who have settled in a place, and use interviews to research aspects of their experiences of immigration. Using discourse analysis, you might choose to examine political debates surrounding immigration legislation, or analyse press reports on immigration issues, or anti-immigration literature published by right-wing organizations. In selecting these different methods, the research problem that you are setting up will be rather different. A statistical analysis will be asking about the number and origins of people entering a country in any one year, and will be seeking to trace whether patterns of immigration are changing in these respects. A researcher might use interviews, in contrast, to explore the subjective experience of immigration, focusing on such issues as people's experience of immigration bureaucracy, or the process of integration, questions of cultural difference, the notion of `home', and so on. A discourse analysis, meanwhile, might be concerned with how immigration is constructed as a political issue, the ways in which immigrants are represented within public discourses, the manner in which certain conceptions of immigration are warranted in opposition to alternative ways of thinking for example, dominant representations of immigration in terms of illegality or threat (see Van der Valk 2003; see also Philo and Beattie 1999). A starting point for such a study could be as simple as: `How is immigration constructed as a problem within political discourse?' The analytic process will tend to feed back into this guiding question, helping to refine the research problem as you go along. Having set up the problem that you wish to investigate, the next step is to collect data for analysis. This will in part be determined by how you are defining the issue. Do you want to look at immigration policy? Or do you want to look at aspects of immigrant identity? Are you interested in media representations of immigration issues? Or do you want to explore attitudes towards immigration within sections of the public? Depending on how you are conceptualizing the research problem, you could collect data from a number of sources. These include parliamentary debates, political speeches, party manifestos, policy documents, personal accounts (including interviews), press or television reports, and campaigning literature. As with textual analysis more generally, a discourse analyst potentially can draw on a very wide range of data. However the primary consideration in selecting textual material is its relevance to the research problem, rather than simply the number of texts analysed. It is therefore especially important to make clear the rationale for your selection, and how it might provide insights into a topic. When doing discourse analysis it is not necessary to provide an account of every line of the text under study, as can be the case in conversation analysis. It is usually more appropriate and more informative to be selective in relation to the data,

extracting those sections which provide the richest source of analytic material. This does not mean that one simply `selects out' the data extracts that support the argument, while ignoring more troubling or ill-fitting sections of the text. Contradictions within a text (including and perhaps especially those parts that contradict the researcher's own assumptions) can often provide a productive point of an analysis.

Sorting, coding and analysing data Discourse analysis has been called a `craft skill' (see Potter and Wetherell, 1994: 55), and compared to riding a bike - a process that one picks up by doing, perfects by practising, and which is difficult to describe in a formal way. Doing effective discourse analysis has much to do with getting a real feel for one's data, working closely with them, trying out alternatives, and being ready to reject analytic schemes that do not work. While it has been argued that discourse analysis is not centrally concerned with `some general idea that seems to be intended' by a text (Potter and Wetherell, 1987: 168), the overall rhetorical effect of a text provides a framework in which to consider its inconsistencies, internal workings and small strategies of meaning. Potter and Wetherell refer to these as the interpretive repertoires at work within a discourse, the ways of speaking about and understanding a topic that organize the meanings of a text. If there is one rule of method that we might apply to discourse analysis, it would be Durkheim's first principle: abandon all preconceptions! At times it can be tempting to impose an interpretation on a sample of discourse, but if this is not supported by the data then it will not yield an adequate analysis. We cannot make the data `say' what is simply not there. Most discourse analysts would reject the idea that texts are open to any number of different, and equally plausible, readings. Rather, analytical assertions are to be grounded on evidence and detailed argument. In this respect discourse analysis entails a commitment to challenging common-sense knowledge and disrupting easy assumptions about the organization of social meanings. Discourse analysis is an interpretive process that relies on close study of specific texts, and therefore does not lend itself to hard-and-fast `rules' of method. Even so we might take a cue from Foucault (1984: 103), who suggested that one might analyse a text in terms of its structure, its architecture, its intrinsic form and the play of its internal relationships. Put simply, this directs our attention to the organization and the interpretive detail of given texts. Here we can identify some useful pointers for analysis: (i) identifying key themes and arguments; (ii) looking for variation in the text; and (iii) paying attention to silences. Note that these represent devices or tools for opening up a text, rather than a fixed set of analytic strategies. The tactics that you adopt as an analyst come from engagement with the data themselves, rather than from any textbook approach. Identifying key themes and arguments A common starting-point for analysis is to locate key categories, themes and terms. Identifying recurrent or significant themes can help you to organize the data and bring a more systematic order to the analytic process. In this way, discourse analysis draws on more general approaches to handling and coding qualitative data (see Chapters ? and ?). The analytic process involves sifting, comparing and contrasting the different ways in which these themes emerge within the data. On a simple level, the repetition or emphasis of key words, phrases and images reveals most clearly what the speaker or writing is trying to put across in the text. This can provide the basis for a critical interrogation of the data. What ideas and representations cluster around key themes? What associations are being established between different

10

actors or problems? Are particular meanings and images being mobilized? How are different subjects spoken about and positioned within the text? Bells analysis, for example, was organized around the three core arguments she identified as shaping the political debates on incest and thereby defining this category as a social and legal problem. Within this frame, she considered the different ways these arguments positioned the victims of incest: as potential offspring, as vulnerable children and young people, as the family as a whole. These analytic strategies helped to open up the interpretive repertoires the ways of speaking and modes of understanding at work in the texts she studied. Reading for key categories and themes can highlight how meanings are attached to abstract or vague formulations that then become difficult to challenge. In her analysis of right-wing discourses on immigration, for instance, Van der Valk (2003) examines how a notion of the people is mobilized within political debates in France so as to exclude certain groups. She argues that not only do these discourses represent immigrant others in a negative light (especially in terms of criminality), but these negative associations are transferred to those seen as allies of immigrants specifically, the political left. The invocation of the people works not only to legitimize the anti-immigration discourses of the right, but also to question the legitimacy and the loyalties of the left. References to such abstract notions as the people, `the community' or `family life' in political discourses are hard to rebut, because they seem to embody values which no one would want to dispute, but at the same time are often imprecise. For these reasons, they can become powerful sites of meaning. Looking for variation in the text Another useful tactic for opening up a piece of discourse is to look for patterns of variation within the text. Differences within an account point us to the work that is being done to reconcile conflicting ideas, to cope with contradiction or uncertainty, or to counter alternatives. By paying attention to such variations the analyst disrupts the appearance of a `smooth' piece of discourse, allowing insights into the text's internal hesitations or inconsistencies, and the way that the discourse aims to combat alternative accounts. Huckins (2002) study of media discourse on homelessness draws an example from detailed analysis of a newspaper editorial. He argues that the text emphasizes substance abuse and mental illness as two of the chief causes of homelessness, but does not include strategies for addressing these problems in its discussion of public responses to homelessness. Rather the text concentrates on charity and voluntary action, on jobs, and on policing and criminalization. There is a mismatch, then, between the textual account of the causes of homelessness, and the account of possible solutions. Huckin reads this mismatch in terms of a conservative political agenda that stresses the role of charity, opportunity and private enterprise over public welfare programmes. Indeed, the public responses highlighted in the text are associated with law and order, which the analysis suggests may be inappropriate to problems of mental illness and substance abuse. In reading for variation in the text, Huckin develops an argument about the way preferred explanations of homelessness are advanced, while alternative arguments are countered. Looking for associations, and reading for variations or contrast, represent two tactics for analysing what Foucault called the play of internal relationships within a text. In her account of right-wing political discourse, Van der Valk analysed how associations are created between the people and the parties of the right, and between immigrants and the left. In the study of homelessness, Huckin explores the variation or inconsistency between the diagnosis of a problem (mental illness and substance abuse), and the account of practical solutions (charity and policing).

11

Attending to silences Huckins study also points us to the silences that run through media accounts of homelessness. He argues that textual silences on issues of racism, domestic abuse, and lack of affordable housing support a reading of homelessness in terms of individual pathologies and criminal behaviour. Similarly, van Dijk (2000) refers to the silence of ethnic minorities in media coverage of race. Minority voices are seldom heard in mainstream media, he argues, and when they do appear often are marginal or treated with scepticism. These kinds of analysis require the researcher to adopt a rather `split' approach to the text. That is, it is necessary to read along with the meanings that are being created, to look to the way the text is organized and to pay attention to how things are being said. At the same time, discourse analysis can require the researcher to read against the grain of the text, to look to silences or gaps, to make conjectures about alternative accounts which are excluded by omission, as well as those which are countered by rhetoric. While I have argued that we cannot force our data to say things that are not there, we can as critical researchers point out those places where the text is silent, to think about what remains unsaid in the organization of a discourse. Such a move can help to place the discourse in a wider interpretive context.

Presenting the analysis The final stage of the research process involves developing and presenting an argument on the basis of your discourse analysis. It is at this point that the researcher is concerned with using language to construct and warrant their own account of the data. This aspect of the process provides a useful context in which to consider the relation of discourse analysis to issues of validity, writing and reflexivity (see Chapter ? for a discussion of the reflexive writing of research reports). Social researchers might think about research validity in terms of both internal and external validity (see also Chapter ?). Internal validity refers to the coherence and consistency of a piece of research, and in particular how well the data presented support the researcher's conclusions. External validity, on the other hand, refers to whether the findings are generalizable to other research or social settings. Discourse analysis has a particular concern with issues of internal validity. Its reliance on close textual work means that it develops arguments on the basis of detailed interpretation of data. How coherent is the interpretive argument? Is it soundly based in a reading of the textual evidence? Does it pay attention to textual detail? How plausible is the movement from data to analysis? Does the researcher bring in arguments from outside the text, and if so how well supported are these claims? Discourse analysis is concerned with the examination of meaning, and the often complex processes through which social meanings are produced. In evaluating discourse research we should therefore be looking for interpretive rigour and internal consistency in argument. Analytic claims need always to be supported by a sound reading of data. In this sense, good discourse analyses stand up well to the demands of internal validity. However this is not to say that discourse research aims to offer a `true' or objective account of a given text. The discourse analyst, like other social actors, aims to provide a persuasive and well-supported account, offering an insightful, useful and critical interpretation of a research problem. This makes it difficult to advance claims to external validity. The discourse analyst seeks to open up statements to challenge, interrogate taken-for-granted meanings, and disturb easy claims to objectivity in the texts they are reading. It would therefore be inconsistent to contend that the analyst's own discourse was itself wholly objective, factual or generally true. This is especially the case where discourse analysts deal with relatively small data sets emerging from specific social settings.

12

These are unlikely to support claims of being more widely representative, and raise critical problems regarding generalization. The response to such a problem goes back to the argument that discourse analysis requires a distinct way of looking at social research. This involves an interpretive commitment to processes of meaning in social life, a certain modesty in our analytic claims, and an approach to knowledge which sees this as open rather than closed. By adopting such an approach to knowledge, the analyst and the reader can be confident of the internal validity and wider relevance of a particular account while remaining open to other critical insights and arguments. Discourse analysis fits into a broad range of social research methods which between them seek to analyse general social patterns but also to examine the devil in the detail. A critical and open stance towards data and analysis may also be understood as part of a reflexive approach to social research. In aiming to be reflexive in their research practice, social researchers question their own assumptions, critically examine their processes of inquiry, and consider their effect on the research setting and research findings whether in terms of their presence in a fieldwork situation, the way they select their data, or how their theoretical framework shapes the process of data collection and analysis. Reflexivity also involves attention to the writing strategies that researchers employ to construct a research account (see Chapter ?), and here the insights of discourse analysis are very useful. In writing this chapter, for example, I have drawn on various discursive strategies in an effort to make my account fit into a methods textbook. While stressing that discourse analysis does not sit easily with hard-and-fast rules of method, I have at the same time drawn on a particular language (data, evidence, analysis, validity) and on particular forms of textual organization (moving from theory to empirical examples, using subheadings and lists) so as to explain discourse analysis in the form of a fairly orderly research process. An attention to the way that language is put to work is a useful tool for any reader or researcher who wants to think critically about social research processes and to evaluate research findings.

Further reading Neuendorf (2002) provides a comprehensive guide to content analysis, while Berger (2000) offers a concise introduction to the method in communication research. Wetherell, et al. (2001a) collects together pieces by key writers on different approaches to discourse theory, while (2001b) offers a practical guide to discourse analysis; Fairclough (2003) is another extremely useful textbook. For shorter introductions, Gill (1996) and Potter and Wetherell (1994) are clear and very helpful. Both provide detailed examples of the use of discourse analysis in media research Web resources The Content Analysis Guidebook Online is a useful web resource accompanying Kimberley A. Neuendorfs The Content Analysis Guidebook (Sage 2002). It is accessible at http://academic.csuohio.edu/kneuendorf/content/ Martin Hammersleys Discourse analysis: a bibliographical guide provides a clear introduction to methodological debates about discourse analysis, and a useful guide to reading. This is available at http://www.cf.ac.uk/socsi/capacity/Activities/Themes/In-depth/guide.pdf Additional references

13

Beard, R. and Payack, P.J.J. (2000) Presidential debates mirror long-term school decline. http://www.yourdictionary.com/library/presart1/html. Bell, V. (1993) Whats the problem? The construction and criminalisation of incest, in Interrogating Incest: Foucault, Feminism and the Law. London: Routledge: 126-49. Berelson, B. (1952) Content Analysis in Communication Research. Glencoe: Free Press. Berger, A.A. (2000) Content analysis in Media and Communication Research Methods: An Introduction to Qualitative and Quantitative Approaches. London: Sage. Van Dijk, T. (ed.) (1997) Discourse Studies: A Multidisciplinary Introduction. London: Sage. van Dijk, T. (2000) New(s) racism: a discourse analytical approach, in S. Cottle (ed.) Ethnic Minorities and the Media. Buckingham: Open University Press: 33-49. van Dijk, T. (2002) Discourse and racism, in D. Goldberg and J. Solomos (eds) The Blackwell Companion to Racial and Ethnic Studies. Oxford: Blackwell: 145-59. Hall, S. (1992) The West and the Rest, in S. Hall and B. Gieben (eds) Formations of Modernity. Cambridge: Polity. Hammersley, M. (2002) Discourse analysis: a bibliographical guide. http://www.cf.ac.uk/socsi/capacity/Activities/Themes/In-depth/guide.pdf. Huckin, T. (2002) Textual silence and the discourse of homelessness, Discourse and Society 13/3: 347-72. Fairclough, N. (2000) New Labour, New Language? London: Routledge. Fairclough, N. (2003) Analysing Discourse: Text Analysis for Social Research. London: Routledge. Foucault, M. (1984) What is an author? in P. Rabinow (ed.) The Foucault Reader. Harmondsworth: Penguin. Neuendorf, K.A. (2002) The Content Analysis Guidebook. London: Sage. Philo, G. (2001) Bad news from Israel: media coverage of the Israeli/Palestinian conflict. Glasgow University Media Group. http://www.gla.ac.uk/departments/sociology/Israel.pdf. Philo, G. and Beattie, L. (1999) Race, migration and media, in G. Philo (ed.) Message Received. London: Longman. Philo, G., Gilmer, A., Rust, S., Gaskell, E. and West, L. (2003) Television coverage of the Israel/Palestinian conflict, in D.K. Thussu and D. Freedman (eds) War and The Media: Reporting Conflict 24/7. London: Sage. Seale, C. (2002) Cancer heroics: a study of news reports with particular reference to gender, Sociology 36/1: 107-26. Smith, A-M. (1994) New Right Discourse on Race and Sexuality. Cambridge: Cambridge University Press. Van der Valk, I. (2003) Right-wing parliamentary discourse on immigration in France, Discourse and Society 14/3: 309-48. Wetherell, M. and Potter, J. (1993) Mapping the Language of Racism. New York: Columbia University Press. Wetherell, M., Taylor, S. and Yates, S.J. (eds) (2001a) Discourse Theory and Practice: A Reader. London: Sage. Wetherell, M., Taylor, S. and Yates, S.J. (eds) (2001b) Discourse as Data: A Guide for Analysis. London: Sage.

14