Copyright Sociological Research Online, 1999

Katie MacMillan and Shelley McLachlan (1999) 'Theory-Building with Nud.Ist: Using Computer Assisted Qualitative Analysis in a Media Case Study'
Sociological Research Online, vol. 4, no. 2, <http://www.socresonline.org.uk/4/2/macmillan_mcLachlan.html>

Abstract

We examine Nud.ist software in terms of its ėtheory-buildingķ properties in order to access the extent to which Nud.ist can be used, not only to develop content categories, but also to develop a research method using two potentially incompatible approaches.
The methods, content analysis and discourse analysis, were used in a single case study on education news in the press. Our case study, on how news about education issues gets constructed and framed by the national press into generalized themes and narratives, was initially informed by an extensive content analysis of the news over a twelve month period. Having identified variations in press coverage, we then collected large quantities of media text on education issues, using Nud.ist to organize and to recode the subsequent data. Having categorized the news extracts our aim was to then explore whether Nud.ist could assist a discourse analysis of the text.

Keywords:

Content Analysis; Discourse Analysis; Education; Framing; Media; Metanarratives; Nudist; Theory-building

Introduction

1.1

Recent exchanges on 'QSR-Forum,'^[1] an e-mail discussion group on the use of qualitative software in research, have raised the question of whether the qualitative software package NUD*IST could be regarded as a research method in itself, with replies asserting that the medium (NUD*IST) had indeed become the method (the analysis). This exchange draws on the claim which inevitably accompanies descriptions of the software package - that NUD*IST has 'theory-building' properties. Descriptions of computer assisted qualitative data analysis software (CAQDAS), and in particular NUD*IST, frequently start from the premise that data analysis software can be used as a method of 'theory development' (Richards and Richards, 1991, 1994). According to the creators of NUD*IST, Thomas and Lyn Richards, the generation of categories in any 'code and retrieve' method always involves theoretical considerations, and by offering the opportunity for new ways of viewing the data clearly engages the researcher in theory-building.

'Decisions are being made about what is a category of significance to the study, what questions are being asked, what concepts developed, what ideas explored, and whether these categories should be altered, redefined, or deleted during analysis' (Richards and Richards, 1994: 447).

1.2

NUD*IST is described by its manufacturers as 'a multi-functional software system for the development, support and management of qualitative data analysis (QDA) projects ... used for handling Non numerical Unstructured Data by simple but powerful processes of Indexing Searching and Theory-building' (QSR home page http://www.qsr.com.au). We were already aware that this software programme could be used to hold large amounts of categorized and sub-categorized text. Our aim was to see if, by using NUD*IST (version 4)^[2], this kind of organization would be able to contribute to a detailed analysis of how news stories structured their accounts, or whether, as we suspected, we would end up discarding the categories and returning to the original text for a separate discourse analysis (DA).

1.3

Our study on education (see MacMillan and McLachlan, 1998) set out not only to examine how the media reported education news, but also to explore the extent to which content analysis and discourse analysis (methods based on very different approaches to data) could be used together in a single case study. The project^[3] within which the study was situated had used content analysis to give a broad overview of the routine content of the national newspapers in Britain, and had also carried out several DA studies examining news in terms of its contribution to information and democracy. The general audit of news information consisted of a daily monitoring of what are generically known as the broadsheets (the Financial Times; the Guardian; the Independent; the Telegraph; the Times; and the Independent on Sunday; the Observer; the Sunday Telegraph; the Sunday Times) and the tabloids (the Daily Mail; the Daily Star; the Express; the Mirror; the Sun; and the Express on Sunday; the Mail on Sunday; the News of the World; the People), and used content analysis as a quantitatively oriented method of measuring defined units of news (Manning and Cullum-Swan, 1994; see also Berelson, 1952). This method is well adapted to mass communications research, and for quantifying news content in general. However, while content analysis adequately defines news in terms of categories, it has little to say about the context of individual news stories. DA, on the other hand, studies talk and texts within the context of their occurrence 'as situated and occasioned constructions' (Edwards and Potter, 1992: 2). In examining newspaper texts, for example, our interests were in how descriptive categories are a concern and rhetorical resource for the newspapers themselves (MacMillan and Edwards, 1998; 1999).

1.4

For content analysis news items are unproblematically the stories which appear on the news pages of the newspaper, with what the story is about immediately definable in terms of 'types.' For DA categorization is not the method, but rather the phenomena to be studied. One of the aims of the project as a whole was to examine the extent to which these two potentially incompatible approaches could be used together in a single study. This required a method which would enable us to use the same data in different ways; to categorize news text, and to analyze its rhetoric. The proposition of a 'theory-building' method was instantly appealing, since, according to the originators of NUD*IST, it proposed a method of 'exploration and linking of theoretical and other organizing and explanatory concepts and statements' (Richards and Richards, 1994: 447).

1.5

Our case study began by identifying fluctuations in press coverage of education news over the first year of our audit of news information. We then examined the kinds of stories most predominant during times of increased media attention, and found that the majority of stories in the tabloid news were on events surrounding the Ridings school in Halifax, England. The events reported stemmed from a pupil's expulsion from the school for allegedly attacking a teacher. This news formed part of further reports on the threat of teacher strikes if a number of disruptive pupils were not expelled. Other news items surrounding the Ridings school covered rioting pupils, school inspections, the appointment of a new head teacher for the school, school closure following allegations of an attack on staff by pupils, government statements on discipline and order, and, based on school inspectors' reports, the pronouncement that the school was failing.

1.6

We collected a large quantity of textual materials, involving news reports, editorials, columns, and features from a sample of the national press. These items were not only specifically on the Ridings school, but also included accounts on disruptive pupils in other schools, with a frequently raised concern about issues of discipline and morality. There were news stories, for example, on government debates on corporal punishment, and press opinions on how the reported events reflected on the state of education and society in general.

1.7

We were interested in claims about NUD*IST's capabilities, and what this would actually mean to us in practice. NUD*IST is clearly regarded as a useful tool for handling large quantities of data, and assigning categories (Buston, 1997), but what would the other proposed qualities of the software mean to our research? Would a 'theory building' feature offer us more than an extensive or elaborate content analysis? Implicated in descriptions of NUD*IST is the notion that these 'theory building' properties are special, unique to particular software packages, as though it offers an approach in which theory is something which develops alongside the analysis, rather than analysis being guided by specific theoretical underpinnings. It is, however, difficult to imagine any research that is theory free - even SPSS methods are theory-implicative. We wanted, therefore to assess whether NUD*IST was offering anything more than a way of doing inductive descriptions of data; whether the method itself was theory neutral; and whether, in fact, it was able to do anything more than any filing cabinet, hard disk folder, or word processor.

An Overview of Software

2.1

As part of a more general development in information technology, as well as an increase in qualitative data collection in recent years (Coffey and Atkinson, 1996), since the 1980's the interest in CAQDAS as a useful research tool for qualitative research has gathered momentum. There are a growing number of software packages available, constantly being upgraded, such as Atlas/ti, NUD*IST, HyperResearch, and Ethnograph. There is also a large amount of literature within this field either praising the use of CAQDAS or discussing the dangers of the researcher becoming too reliant on such tools. Qualitative software packages are frequently described as useful for speeding up more mundane research tasks by providing an efficient system for categorizing and creating relationships in data (Barry, 1998; Richards and Richards, 1991; 1994. In contrast some of the main concerns involved in using these software packages are that these 'tools' may create distance between the researcher and the data through the need to impose categories, or that through using the same automated process there will be an unwanted conformity in qualitative data analysis (Agar, 1991; Kelle, 1997; Mangabeira, 1995; Seidel, 1991). The concern is that software packages will constrict the research process, with analysis guided by what the software can do, rather than by the data.

2.2

New software packages are frequently being developed, and the decision of which to chose, if any, is not a straightforward task (see Barry 1998; McLachlan 1996; Tesch, 1990; Weitzman and Miles, 1996; for an outline on the practical uses of a number of software packages). The packages available under the heading of 'qualitative software packages' are normally sub-divided into 'text retrievers', 'code and retrieve packages' and 'theory building software.' This is not a strict division of packages in relation to specific functions they perform, as the newer packages encompass elements from all three groups. Although most qualitative software packages perform text searches, some specialise in doing so. These powerful packages search for words or combinations of words at speed. 'Code and retrieve' packages not only search for words but also allow codes to be assigned to specific segments of text. More recently packages have become available which are described as 'theory building' (e.g. Richards and Richards 1994). These packages not only perform text searches and coding functions but also claim to assist 'theory building' with the data. Examples of these are AQUAD, Atlas/ti, HyperRESEARCH and NUD*IST. These packages normally allow the same facilities of the code-and-retrieve packages but try to go a little further. For example they may allow the researcher to code data which is not simply text based, such as videos and audio tapes (though in practice all this really means is that you are able to tell where and when your data has been stored in this form). Alongside this the packages may also enable you to build up hierarchical structures of codes and build elaborate search statements, allowing for the data to be investigated to a greater degree. This was an argument in favour of using the software in our own particular case study, with the initial promise that we could develop a theoretical basis not only for creating categories, organizing them hierarchically, and analyzing the resulting new categories, but also for forging a link between the methods of content analysis and the theory and methods of discourse analysis.

2.3

In assessing the various strengths and weaknesses of qualitative software packages Christine Barry (1998) compares Atlas/ti with NUD*IST. Barry describes Atlas/ti as having a visually appealing user interface, the capability to view all the features at one time on the screen and an ability to make 'Hypertext' links (links between the textual data sets using memos or codes). A disadvantage is that the package lacks what are viewed as the more complex analysis options of NUD*IST. NUD*IST, according to Barry, is systematic, with a sophisticated searching system, and project management functions. Compared to Atlas/ti, NUD*IST has, however, a less user-friendly interface, and, because it lacks the Hypertext links facility, is limited in its ability to make links between data. In summary Barry views Atlas/ti as the best choice for 'simple' projects and NUD*IST as more suitable for 'complex' projects. Simple projects are defined by Barry as those involving straightforward, simple sample, one timepoint projects. Complex projects use different data types, larger samples, and mixed qualitative and quantitative studies. In conclusion Barry warns against placing too much emphasis on the analytical capabilities of software, since their effectiveness lies in the ability to 'manage' data rather than to analyze it.

2.4

After examining various evaluations of the qualitative software packages (e.g. Barry 1998; Buston 1997; Kelle 1997) we chose NUD*IST as most suitable for our project. It is important to note, however, that other studies we examined used NUD*IST to analyze interview transcripts, not newspaper texts. This affects to a certain degree the kind of evaluation we are able to make on the usefulness of this software package.

Education News Categories

3.1

For our education case study we were interested in examining a large number of newspaper headlines not only on the Ridings school itself, but also on more general issues of morality and discipline, and how these were made relevant to education news by the newspapers concerned. The broadest aims of the study were to examine the extent to which NUD*IST could be used to develop content categories, and, more ambitiously, to use both content analysis and DA in a complementary way. Whereas content analysis requires that segments of text are systematically analysed through the application of pre-determined, discrete categories ready for quantification (see Holsti, 1969; Krippendorf, 1980), the analytical claims of DA are tied to textual details, and refer to a wide range of grammatical, rhetorical and conversational devices (e.g. Billig, 1987; Edwards, 1997; Edwards and Potter, 1992; MacMillan and Edwards, 1998, 1999; Potter, 1996; Potter and Wetherell, 1987).^[4] For example, in a study of the British newspaper coverage of the death of Princess Diana, MacMillan and Edwards (1999) analyze descriptive categories used by the press themselves. We show how the behaviour of those involved in photographing Diana at the time of her death, is distinguished from the British press in general, through evocations of national identity. In converting a quote from Diana's brother, Earl Spencer, condemning the press, various newspapers involved:

...managed to reformulate and re-situate his statement as a condemnation of the paparazzi. Further, the emphasis on the nationality of the photographers constructed a division between them and the British press. Thus the newspapers were able to shift the focus and implications of Spencer's statement from the category 'press', which includes them, to the foreign press which does not, and in particular to the paparazzi. (MacMillan and Edwards, 1999, p. 160).

3.2

In the education study we were interested in looking at the kind of events which attract media attention and how these events are portrayed. Using the results of the overall news audit from the project to identify variations in education news cover in the press, we collected headlines and text from education news stories during what we identified as months of 'average' coverage, as well as from selected periods of variation. This included collecting all of the relevant newspaper items between October and December 1996 - the period indicated by the audit as showing a significant rise in tabloid coverage. From our reading of the news during periods of increased media attention, we began to build up a picture of recurring 'themes' in education news (including those on low standards and failing schools).

3.3

Collecting newspaper headlines and text on relevant topics required extensive newspaper searches. We treated the news headlines not only as a summary of the news report, but also as an evaluation, which gave the story an immediate impact and significance (see Hall et al, 1978; van Dijk, 1998). The headlines were to give us not only an initial overview of the quantity of stories on a particular topic, but also the newspapers' own particular gloss on the event. Our search was carried out manually and, where available, digitally. The latter search was done using CD-ROMs and newspaper web pages. It is worth mentioning that searching newspapers on CD-ROMs highlighted certain problems. The only titles available on CD-ROMs are (currently) the main national broadsheet newspapers, plus the Daily Mail. National tabloids were not available in this format and had to be searched manually. Relevant headlines and text were then both typed and scanned into the computer.^[5] Furthermore, there are differences, both in format and in content, between published hard copy and the digitized version. CD-ROMs provide text only, with occasional segments of the newspaper found to be missing. Digital word searches were likely to produce inflated results (Hansen, 1992; Soothill and Grover, 1997), finding items from unrelated topics, as well as relevant stories. Sifting through CD-ROMS for appropriate news stories, and discarding inappropriate ones, took a significant amount of time, and was an unexpected drawback in our use of computer software.

3.4

For a more detailed examination we collected news stories, articles, and letters from the national press between October and December 1996. After an initial reading of the stories we created a set of rough categories, and then searched CD-ROMS for stories containing the words 'Ridings school,' 'violence,' 'moral education,' and 'corporal punishment.' Stories during this time related to events surrounding what became dubbed as 'the worst school in Britain' (see Clark, 1998), the 'failing' Ridings school in Halifax. This major news item provided the backdrop for various other stories that were to follow - on morality, the need for discipline and on social problems. We categorized the headlines according to themes which reoccurred throughout the news items, columns, and editorials, having first imported the text into NUD*IST.

3.5

Here we encountered problems that a reading of the on-line manual (NUD*IST's 'Help' facility) did not prepare us for. Getting information into NUD*IST requires that text be saved as 'text only,' with 'headers' appropriately formatted with asterisks, spaces and carriage returns. Each file was also given its own 'header' allowing data to be labelled with information including details on, for example, the name and date of the newspaper. Descriptions in the manual on formatting are not clear, and our resulting problems - including spacing and carriage returns - meant that we needed to re-import the data a number of times. Specifying the size of text units was more straightforward. Each headline was treated as a text unit, and assigned a code on its own.^[6] Each headline was also given its own additional 'header' which allowed the data to be labelled with information, including details on, for example, name, date and page numbers of the source. Once the data had been successfully imported into NUD*IST we then began the coding process.

Working with NUD*IST

4.1

NUD*IST's Index System is a 'nodal' system for organising data, visually represented by a tree structure. Although not limited to this structure, it encourages using hierarchical categories from the outset. Categories are linked in a 'parent' (main category) and 'child' (sub-category) relationship. Each branch ending on the visual tree represents a different coding category or 'node' (see figure 1), with the nodal categories containing information connected by a common category definition. We began by isolating the issues we wanted to focus on, and creating some general 'nodes' with which to organise our data. The basic story themes from the headlines became 'parent' nodes, from which other themes were created. However, despite the nodal system being a key feature, and claimed advantage of NUD*IST, this wasn't as simple as anticipated. For example, figure 1 shows that the data in this project has been categorised into various nodes. From the parent node of 'Ridings' the data has been split into various child nodes representing different 'culprits' from the Ridings news story such as 'pupils', 'government officials', and 'unions.' As the analysis progressed we needed to rename and recode the various categories. Although the renaming of codes is straightforward in this system, moving data between categories is more problematic and it took time to reorganise our data into a new coding system. It was at this point, when our attention was drawn away from the data, and we became increasingly preoccupied with operating NUD*IST, and assigning, and reassigning categories, that we began to doubt whether this kind of coding process would actually be able to contribute significantly to our final analysis. We spent more time re-evaluating our choice of nodes, questioning why we picked particular categories, and how coding the newspaper headlines and text like this could help us, than we did on examining what the press were saying about education. Coding data in this way, whether as flexible categories, or as a pre-formed coding system, inevitably decontextualizes the data, and the more the text got segmented into categories, the more it lost the specific context from where, in DA terms, it make sense. A possible solution might have involved putting more text into each node, and using NUD*IST's memo facilities to signal what prompted each particular categorization. This, however, is no more helpful to the analysis than a well-organized set of index cards and box files/computer files containing transcripts.

4.2

Selecting headlines, searching CD-Roms, formatting text for NUD*IST, and creating categories as part of a nodal system were all time consuming. However, once the data was coded and available for key word text searches, then NUD*IST operated efficiently. Setting aside our reservations about coding and context we began to play around with NUD*IST, to see what it was capable of. Having entered and categorized the data, in terms of newspaper, month, and then story type, we were able to look at the kind of stories which occurred in the various newspapers, and from there to examine how the reports were worded, and how frequently certain word combination and 'actors' (i.e. Government; teachers; pupils; etc.) occurred. Based on initial readings of the broadsheet and tabloid news around the time of the Ridings events, we hypothesized a difference in descriptions between the newspapers. For example, pupils were frequently described in the tabloids as out of control, as 'yobs,' 'thugs,' 'tearaways,' and 'bullies,' but not so in the broadsheets. We were interested how these various 'actors' were portrayed, and how, in the newspapers' description of them, they were constructed as contributing to, culpable, or answerable for a variety of problems in education, and within society in general.

Figure 1: NUD*IST working interface

4.3

Figure 1 shows the NUD*IST working interface. Here we can see the 'Node Explorer,' which contains information on each node - including the name, position and contents of that node, as well as the results of text searches and index searches. The 'Document Explorer' contains all of the raw data files originally entered, while the 'Tree Display' shows the root nodes as well as a figure of the whole tree. One of NUD*IST's limitations is that it does not allow a full Tree Display, including named nodes, to be printed. In order to look at our own hierarchical structure it was necessary to write out all of the nodal categories manually.

4.4

At this stage we were able to examine the contents of the nodes, the original data files and the nodal tree structure, to re-name nodes and to add additional nodes as research progressed. However, while categories were easily renamed, revising the data within the nodes was more problematic, requiring the manual removal of data, item by item, in order to create a new sub-category under a new name. The ability to quickly and easily re-structure data relies on not wanting to change the categories too much - like re-locating computer files into new folders, the content of each file remains untouched. If anything this acts as a disincentive to re-categorizing data at the raw level, and a further restriction on what constitutes analysis.

Figure 2: Node Explorer

4.5

The above figure gives a more detailed breakdown of the nodes within a tree system. As we can see in the 'October' node, each 'parent' category is subdivided into five 'children' nodes, and subdivided a further five times. This root subdivision is repeated with each of the other parent nodes, with some minor variations ('children' were added, or detracted, according to story 'themes' during each of the chosen months). In Figure 2. the 'Ridings' node for 'October' is highlighted, revealing 13 documents (named newspapers) and 96 text unit (96 headlines from these newspapers). Viewing each document is a simple matter of activating NUD*IST's 'Browse' button.

Figure 3: Text Searches

4.6

This figure once again shows the 'Node Explorer,' but this time also displays 'Text Searches.' This display shows the list of text searches carried out on the education data, including the node address, the coding status (how many documents contained the word/phrase searched for), and any restrictions placed on the search (such as the exclusion of certain nodes). We were interested in certain lexical items - descriptive words used in the newspaper headlines - and how these conveyed a particular version of events. Our search highlighted the frequency with which certain words were used by tabloids, in comparison to the broadsheets, and how often these stories became front page news. Although NUD*IST is not necessary for this kind of news count, the searches were swift and efficient. We searched, for example, for emotive terms such as 'yob' (see text search no. 2), 'hell' (11), 'tearaway' (19), and 'crisis' (4), and how frequently such words formed part of the descriptions of pupils' behaviour, or reported conditions at the Ridings school. This initial exploration formed the basis of a more detailed textual analysis on how such descriptions worked as part of a larger, more persistent meta-story on society in general.

Figure 4: Text Search Node Browser

4.7

Figure 4 shows the Node Browser displaying the results of one of the Text Searches (see figure 3) carried out for the key word 'yob.' This is only a small section of the node browser, which in this case contained a large amount of data. However from this segment we can see how the search process has accessed the documents (newspapers) containing the word 'yob' or similar (e.g. 'yobs,' 'yob kids,' 'yobbos'). The screen shot shows the headline, the name of the newspaper, and the month. The date of the newspaper is shown at the end of the headline (in brackets).

Figure 5: Index Searches

4.8

Figure 5 shows the Index searches carried out on the data. For example, by using the 'intersect' tool to search both the 'Ridings' category and the 'violence' category in the 'October' parent node, we were able to collect items coded under both nodes. This usefully showed the extent to which, when the Ridings school was in the news, it was reported in conjunction with descriptions of violence. These Boolean/Set searches are more efficient in NUD*IST than in trying to do the same thing via a word-processor, and searching file by file.

Theory Building

5.1

The data which finally formed the basis of our conclusions, of course, reflected an inevitable process of selection, as well as the subsequent categorization which was imposed on the newspaper text. All analysis is to some extent selective. In our case, for example, it began with the decision to study education in the media, and then in choosing which key words to retrieve news stories with, which stories to include, and from there which categories to assign segments of data to. It is this process of selection, division and subdivision of text, which the creators of NUD*IST refer to as 'theory building.' However, for us, this method of 'theory building' involved an unavoidable decontextualization. Lexical terms were coded and counted in NUD*IST, and this, rather than focusing our interest, got in the way of the subsequent discourse analysis. It was only later, when NUD*IST was abandoned (both as a tool and as a method), did the analysis really begin to take shape.

5.2

One of the most striking features of descriptions of NUD*IST is in the way that notions of theory are glossed over and misrepresented. A theory, according to dictionary definitions, is an explanatory model, a body of principles used to explain a phenomenon.^[7] Richards and Richards, however, discuss how using this software encourages the assembly of 'little ideas hardly worth calling theories' which then link with 'other theories and make the story, the understanding of the text.' (1994: 448). This either presupposes that a single coherent understanding of the data can be assembled by using a random collection of thoughts, ideas, and hunches, to weld together different theoretical approaches ('other theories'), or, and this seems more likely, that the authors are using the term 'theory' as interchangeable with that of 'method.' What NUD*IST does not do is build theory, but rather it encourages hierarchical links between categories, using memos and interpretations.

'Consider, for example, how we work when developing theory from the text. We often get going by finding little things that relate in some meaningful way - perhaps, if our interest is in stress, that certain topics get discussed in anxious ways (and that is something that good coding and retrieval can find for us). So then we start looking for components in those topics that might cause anxiety, often by studying the text, finding or guessing the components and coding for them, recalling situation facts not in the text, and looking for suggestive co-currences of codes. We might on a hunch start looking at text passages on people's personal security and how they arrange it ... to see if there is some possible connection between components occurring in the anxiety topics and security arrangements' (Richards and Richards, 1994: 448. Emphasis added).

5.3

What seems to be required, for NUD*IST users to treat a process of categorization as 'theory-building' is an uncritical acceptance that this is how theory works. For DA 'little ideas,' guessing components, and recalling situational details are part of the way that the analyst imposes an interpretation of events on the data (see MacMillan, 1995). In order to accept that NUD*IST is 'theory-building' we need to redefine theory as subjective rather than empirical - that is, that the analysis is in how we organize our categories, rather than what we have worked out from the text. For example, Richards and Richards (1994) treat it as unproblematic that certain topics can be seen to be discussed in 'anxious ways,' as if we all share a knowledge of what 'anxious ways' means, and that this kind of talk can somehow be spotted by anyone doing the analysis. A study by Buston (1997) also reflects this misapprehension about analysis. Buston uses NUD*IST to categorize talk in terms of 'non-compliance,' 'embarrassment,' being 'in denial,' or 'forgetting.' It is not, however, notable from her extracts that these are topics of importance for the speakers, but are rather issues for the analyst. How, for example, does one assign a data extract to the 'in denial' category (when it is not addressed directly by the speaker), without first having to guess what kind of talk displays 'denial'? Like Richards and Richards, Buston's interpretations are guided by the categories she attaches to her data.

5.4

While these are the kinds of interpretative judgements that DA does not make in principle, our own experience with NUD*IST leads us to conclude that NUD*IST's method will inevitably involve dividing data into coded 'chunks' (Richards and Richards, 1994: 448), and assigning categories. For our own study this meant that any expectations of theory-building had to be abandoned. Our theory was discourse analysis. Since NUD*IST's method had become for us yet another form of content analysis, it had little to offer towards a discursive study of news text.

A Discourse Analysis of Newspaper Themes

6.1

What NUD*IST could, and does offer is a method for organizing large quantities of text. We found that, although disproportionately time-consuming, NUD*IST was useful for storing data and assigning codes as a way of highlighting how frequently certain terms were used, and by what newspapers. We also examined how often certain kinds of descriptions were used, and who they were linked to. In examining 'blame,' for example, we were able to see how often, who, and what this was most frequently ascribed to. However, coding should be distinct from the analysis itself (Potter and Wetherell, 1987), and in order to analyze how such descriptions work in the newspaper text, we needed to go back to the original articles. This is where NUD*IST became superfluous to our study. Because DA cannot be done on content categories, we could only be alerted to a general trend in the news. We then needed to return to the raw materials, to find and collect a relevant data set for DA, and to devise analytical themes from them.

Conclusion

7.1

Our case study examined how the media frame news reports (Iyengar, 1991; Stahl, 1989) into a more generalized theme on morality, modern youth, and the state of society. Although we were able to use NUD*IST to show how frequently blame was assigned to particular 'actors,' the software could not, of course, tell us how such blame was assigned. In a detailed DA, focusing on specific words, we showed how the press could build opinionated viewpoints into the phenomena described, rather than maintaining a strict separation between opinions and facts.

7.2

We had set out to examine the extent to which we could develop a link between two inherently oppositional approaches. Content analysis was used as a method of coding news on a large scale, and DA provided the theoretical basis for showing how news, and what is treated as newsworthy, is rhetorically managed. NUD*IST was explored in terms of its claim to 'theory-building', as a way of developing an approach which could use both methods compatibly.

7.3

The main advantage of NUD*IST is its ability to handle large data sets and to perform coding functions and text searches. It can be usefully employed in the early stages of content analysis, to help define what categories are to be used. It can also be used to collect relevant data for discourse analysis, and put it into some kind of organized set, ready for analysis. But it must be stressed that NUD*IST does not do the analysis, nor even play much part in it. While no qualitative software package can in practise build theory, nor do analysis for the researcher, the notion of 'theory-building' does create an immediate and unrealistic expectation about NUD*IST's capabilities (Mangabeira, 1995). NUD*IST was no more helpful for us in connecting the data and methods used in our case study than a well-organized set of files would have been. While our various technical problems with NUD*IST are inevitably linked with our inexperience as first time users (see Barry, 1998), this in itself highlights the need for NUD*IST's on-line 'help' manual to provide clearer instructions on how to set up and to use the software. Our expectations for using NUD*IST as something other than a computerized coding system were fostered by the repeated descriptions of it as a theory-building programme. The promoting of such expectations must, in part at least, be attributed to NUD*IST's creators, and the way that, in their descriptions of NUD*IST, 'theory' is indistinguishable from 'method,' and method from analytical tool.

7.4

This vague notion of what counts as theory is echoed in Kelle's work on 'Theory building in qualitative analysis.' Kelle (1997) argues that the dangers of methodological bias in using computer-aided analysis have been overemphasized, and describes links between NUD*IST's theory building properties with the rules and methods of grounded theory, as a way of attempting to allay such fears. The author, however, defines 'theory' as something which 'can be regarded as a network of categories' (1997: ¶3.5). This self-serving definition, although difficult to contradict because of its lack of clarity, is hardly informative. It misleadingly implies that anything which contains a set of related categories is a theory (as if Ikea's furniture catalogue were, of itself, a theory of furnishing).

7.5

Such vague definitions of theory used in conjunction with descriptions of NUD*IST's capabilities go some way towards explaining how the participants of the 'QSR-Forum' (see Introduction, above) could discuss whether NUD*IST had indeed become the method of analysis, without questioning if, on the same grounds, a pencil, typewriter, or Word for Windows, could also be considered as methods of analysis. NUD*IST may or may not be useful in contributing to theory development. What is at issue is the way that descriptions of it create unrealistic expectations of NUD*IST as something more than a methodological tool. In the light of this NUD*IST users should be encouraged to critically re-examine its capabilities, particularly concerning claims about 'theory-building' and NUD*IST's role as a method of analysis.

Notes

¹ To subscribe to QSR mailing list send your message to: mailing-list-request@qsr.com.au. Include the following in the message: SUBSCRIBE qsr-forum+your-first-name+your-last-name.

² The makers of NUD*IST are releasing a new version in May 1999 called NVivo. Although we have not yet investigated the capabilities of this new package it claims to have improved many of its features. These include being able to deal with a variety of input data formats, new data linking and coding options and improved reporting and exporting facilities.

³ The ESRC funded project 'Information and Democracy' (L126251016) was part of ESRC's Media Economics and Media Culture programme. The project carried out an extensive audit of the national news media over a two year period from September 1996 to September 1998. Data for the education case study was restricted to the first year's results.

⁴ For different applications of DA, including analysis using schematic models of discourse, see, for example, van Dijk (1993). For 'critical discourse analysis' see writers such as Fairclough (1995), Fowler (1991), and Kress and Hodge (1979).

⁵ Although the technology for scanning various textual formats is constantly improving, with hardware getting faster, better, and cheaper all the time, we did encounter various problems with our own OCR scanning software. This included the disproportionate amount of time required to scan in newspaper print for the resulting quality of scanned text.

⁶ When dealing with interview transcripts or long pieces of text NUD*IST requires the researcher to specify the size of each text unit which will have codes assigned to it. This may vary in size from a sentence to a whole page depending on the requirements of research.

⁷ See, for example, the OED definition, including 'a system of ideas or statements held as an explanation or account of a group of facts or phenomena; a hypothesis that has been confirmed or established by observation or experiment, and is propounded or accepted as accounting for the known facts; a statement of what are held to be general laws, principles, or causes of something known or observed.'

Acknowledgements

We gratefully acknowledge helpful comments from Derek Edwards, Peter Golding, and the anonymous reviewers on an earlier draft of this paper.

References

AGAR, M. (1991) 'The right brain strikes back' in R. Lee and N. Fielding (editors) Using computers in qualitative research. London: Sage.

BARRY, C. (1998) 'Choosing qualitative data analysis software: Atlas/ti and Nudist compared', Sociological Research Online. (3), 3, <http://www.socresonline.org.uk/3/3/4.html>.

BERELSON, B. (1952) Content analysis in communication research. Glencoe,IL:Free Press.

BILLIG, M. (1987) Arguing and thinking: A rhetorical approach to social psychology. Cambridge: Cambridge University Press.

BUSTON, K. (1997) 'NUD*IST in action: Its use and its usefulness in a study of chronic illness in young people', Sociological Research Online. (2), 3, <http://www.socresonline.org.uk/2/3/6.html>.

CLARK, P. (1998) Back from the brink: Transforming the Ridings School - and our children's education. London: Metro Books.

COFFEY, A. AND ATKINSON, P. (1996) Making sense of qualitative data: Complementary research strategies. London: Sage.

EDWARDS, D. (1997) Discourse and cognition. London: Sage.

EDWARDS, D. and POTTER, J. (1992) Discursive psychology. London: Sage.

FAIRCLOUGH, N. (1995). 'Media discourse.' London: Edward Arnold.

FOWLER, R. (1991). 'Language in the news: Discourse and ideology in the press.' London: Routledge.

HALL, S., CRITCHER, C., JEFFERSON, T., CLARKE, J., and ROBERTS, B. (1978) Policing the Crisis: Mugging, the state, and law and order. London: Macmillan Press.

HANSEN, A. (1992) Computer-assisted analysis of newspapers: An introduction. Discussion Papers in Mass Communications. Leicester: Centre for Mass Communication Research.

HOLSTI, O. R. (1969) Content analysis for the social sciences and humanities. Reading, MA: Addison-Wesley.

IYENGAR, S. (1991) Is anyone responsible? How television frames political issues. Chicago: University of Chicago Press.

KELLE, U. (1997) 'Theory building in qualitative research and computer programs for the management of textual data.', Sociological Research Online. (2), 2, <http://www.socresonline.org.uk/2/2/1.html>.

KRESS, G., & HODGE, B. (1979). 'Language as ideology.' London: Routledge & Kegan Paul.

KRIPPENDORF, K. (1980) Content analysis: An introduction to its methodology. London: Sage.

MacMILLAN, K. (1995) 'Giving voice: The participant takes issue', Feminism and Psychology, 5, (4), 547-552.

MACMILLAN, K., and EDWARDS, D. (1998) 'Designer families: A discourse analysis of fact and accountability', Journal of Sociolinguistics, 2/3, 323-345.

MACMILLAN, K., and EDWARDS, D. (1999) 'Who killed the princess? Description and blame in the British press', Discourse Studies, 1 (2), 191 - 214.

MacMILLAN, K and McLACHLAN, S. (1998) 'A sign of the times: Education news in the tabloid press', CRC working paper series #5. Loughborough University: Communication Research Centre.

MANGABEIRA, W (1995) 'Computer Assistance, qualitative analysis and model building', in R. Lee (editor) Information technology for the social scientist. London: UCL Press.

MANNING, P. K and CULLUM-SWAN, B (1994) 'Narrative, Content, and Semiotic Analysis' in N. D. Denzin and Y. S. Lincoln (editors) Handbook of qualitative research. London: Sage.

McLACHLAN, S (1996) 'Qualitative software packages: A review', CRC working paper series #1. Loughborough University: Communication Research Centre.

POTTER, J. (1996) Representing reality: Discourse, rhetoric and social construction. London: Sage.

POTTER, J. and WETHERELL, M. (1987) Discourse and social psychology: Beyond attitudes and behaviour. London: Sage.

RICHARDS, T. J. and RICHARDS, L. (1991) 'The transformation of qualitative method: Computational paradigms and research processes', in N.G. Fielding and R.M. Lee (editors.). Using computers in qualitative research. London: Sage.

RICHARDS, T. J. and RICHARDS, L. (1994). 'Using computers in qualitative research' in N. D. Denzin and Y. S. Lincoln (editors) Handbook of qualitative research. London: Sage.

SEIDEL, J. (1991). 'Method and madness in the application of computer technology to qualitative data analysis', in N. Fielding and R.M. Lee (editors) Using Computers in Qualitative Research. London: Sage.

SOOTHILL, K. and GROVER, C. (1997) 'A note on computer searches of newspapers', Sociology, 31 (3), pp. 591-596.

STAHL, S. D. (1989) Literary folklorists and the personal narrative. Bloomington: Indiana University Press.

TESCH, R. (1990) Qualitative research: Analysis types and software tools. London: Falmer Press.

Van DIJK, T. A. (1998) 'Opinions and ideologies in the press' in A. Bell and P. Garrett (editors) Approaches to media discourse. Oxford: Blackwell.

VAN Dijk, T.A. (1993). 'Principles of critical discourse analysis.' Discourse and Society, 4 (2), 249-283.

WEITZMAN, E. and MILES, M. (1996) Computer programs for qualitative data analysis: An expanded sourcebook. Thousand Oaks, California: Sage.

Copyright Sociological Research Online, 1999