Copyright Sociological Research Online, 1997

 

Buston, K. (1997) 'NUD*IST in Action: Its use and its Usefulness in a Study of Chronic Illness in Young People'
Sociological Research Online, vol. 2, no. 3, <http://www.socresonline.org.uk/2/3/6.html>

To cite articles published in Sociological Research Online, please reference the above information and include paragraph numbers if necessary

Received: 14/2/97      Accepted: 19/9/97      Published: 30/9/97

Abstract

Much has been written in recent years about using computers packages to assist qualitative data analysis. The focus has been on the implications of this development for the analytical process. This paper describes the use of NUD*IST in a recently completed study of the experiences of chronically ill young people, and assesses the epistemological effects of its usage. As well as providing basic practical information on some of NUD*IST's functions, whilst highlighting one particular way of utilising the package, the paper addresses some of the methodological issues raised in the CAQDAS debate by drawing on this 'real-life' experience.

Keywords:
CAQDAS; Chronic Illness; Computer Software; NUDIST; Qualitative Research; Young People

Introduction

1.1
Using computers to aid qualitative research was first publicly debated in the early 1980s. Patton (1980) described the way in which he used punch-cards to store indexing references. Drass (1980) went one step further and presented a software programme designed for this purpose. By 1984, the use of computers in qualitative research was topical enough to warrant a double issue of Qualitative Sociology and, by 1990, the proliferation of textual-analysis programmes encouraged Tesch (1990) to write the first book of its kind, reviewing their capabilities. Since then, packages have proliferated further. Of those available today, many are very sophisticated in their scope and functions. Several texts now exist which review individual programmes (Weaver & Atkinson, 1994; Weitzman & Miles, 1995; Burgess, 1995).

1.2
These developments have been accompanied by much debate about the implications this new technology has for the way in which qualitative analysis is carried out (Fielding & Lee, 1991; Richards & Richards, 1991; Weaver & Atkinson, 1994; Kelle, 1995; Burgess, 1995; Coffey et al, 1996; Lee & Fielding, 1996; Kelle, 1997). Reservations have been expressed, centering round the issue that the computer comes to control research - encouraging or even forcing the researcher into adopting certain procedures. The alienation of the researcher from his/her data has been another oft-expressed worry. What has been emphasised throughout, though, is the ability of textual analysis packages to ease the researcher's workload, save time and generally enhance the power of qualitative analysis. Indeed, many would agree with Richards' assertion that:

The debate about whether to compute in qualitative research seems to be over. ... All researchers working in the qualitative mode will be clearly helped by some computer software (Richards, 1995: p. 105).

1.3
Nevertheless, discussions relating to specific methodological issues continue, with users continually urged to be aware of the epistemological effects of using particular software. From the danger that Computer Assisted Qualitative Data Analysis Software (CAQDAS) may be blurring the line between qualitative and quantitative analysis (Hesse- Biber, 1995), to, more recently, assertions that CAQDAS may be fostering an undesirable convergence towards a unitary ideal-type of data collection, storage and analysis (Coffey et al, 1996; Lee & Fielding, 1996), methodological issues surrounding the use of CAQDAS continue to be explored and debated.

1.4
Less common in the CAQDAS literature, however, have been descriptions of precisely how individual packages have been used in 'real-life' research projects. Although there are some such examples (notably the chapters contained within Burgess et al, 1995), ongoing discussions and pleas for help on the Qualitative Software mailing list as well as needs expressed amongst colleagues evidence a demand for more, particularly from those considering embarking on computer-aided qualitative analysis for the first time. Such descriptions are useful in highlighting, in concrete terms, the kinds of issues which have to be considered and the kinds of strategies which may be adopted. They also add meat to the, often abstract, theoretical debate on CAQDAS, allowing potential users to assess the merits of automating aspects of qualitative data analysis through being able to consider the experiences of others, across a variety of packages and data-sets. This paper aims to describe how NUD*IST was used in one particular project, using this experience to address some of the most salient methodological issues that have been raised in the literature.

The Project

2.1
The study described here aimed to investigate what life is like for young people with persistent health problems. It grew out of a perceived need by a group of clinicians to find out more about commonalities in experiences across illness groups amongst teenagers and young adults. The aim was to discover whether more generic (as opposed to specifically illness-based) clinics and health-care initiatives may be a useful way of proceeding in an era when funding is limited. As such, it was to be one of many projects which investigated chronic illness in young people, with its design intended to allow a group of 'sufferers' to tell their stories and talk freely about what it is like for them to have these chronic health problems. The central research question was: 'what is it like to be young and to have persistent health problems?'.

2.2
The main data collection tool was in- depth interviews, conducted with 112 young people with one of three types of health problems: asthma, mental disorder or recurring symptoms following on from a head injury. They were invited to talk about the health-care they had received, and about the impact of their health problems on their day-to-day lives. An interview schedule, organised under eleven broad headings, was used which provided a structure for the interviewer, whilst allowing the respondent to raise, and talk at length, about particular issues and experiences of importance to him or her. The average length of the interviews was just under one hour. Consequently, once transcribed, there was a great deal of text to handle and analyse.

2.3
Respondents were also asked to complete the Offer Self-image Questionnaire (OSIQ) (Offer et al, 1992), a 129-item personality test used to measure the self-image of adolescents. On analysis, twelve sub-scores are generated relating to mental health, body image and self-reliance, among other things. A 'Total Self-Image' score is also generated, with all thirteen scores presented in the context of norms for the individual's sex and age.

The Package

3.1
NUD*IST was selected to facilitate analysis of the data collected. NUD*IST stands for Non-numerical, Unstructured, Data: Indexing, Searching and Theorising. Put simply, it works with textual documents, and facilitates the indexing of components of these documents; is able to search for words and phrases very quickly; and claims to support theorising through enabling the retrieval of indexed text segments, related memos, and text and index searches; and through the construction of a hierarchically structured tree to order index categories (Richards & Richards, 1994; Qualitative Solutions and Research, 1994; Richards, 1995; Weitzman & Miles, 1995).

3.2
This paper is not attempting to review or comment on specific software packages. NUD*IST was chosen by the research team for several reasons. It was easily available, could be learned quickly, others recommended it, and it had an impressive array of functions which we vaguely thought may be useful aside from its basic ability to let us index and later retrieve categorical interview segments. It seemed 'good enough' for our needs, though, of course, a more advisable way of choosing a package would have been to consider many more, and map out how well each fitted our intended strategy of analysing the data.[1] Stanley & Temple (1995) concluded from their evaluation of five qualitative data analysis packages used on two very different data sets, that there is no one way of using these packages, rather their appropriate use is dependent on the characteristics of the data set. While agreeing with this - NUD*IST, at least, is more flexible than some writers have asserted - thorough research into the underlying assumptions of available packages is advisable.

The Data

4.1
To present a step-by step account of how NUD*IST was used would be misleading. One of its advantages over manual methods is the relative ease with which the researcher can switch between different phases of data analysis. Broadly, though, three stages of dealing with the data can be identified - its introduction; its indexing; and its retrieval in categorised form in order to assimilate and 'write-up'.

Introducing Documents to NUD*IST

5.1
Once the interviews had been conducted they were transcribed onto the word processing package used by the research team. Three decisions regarding the format of the transcriptions had to be made: 1) the size of the text units; 2) the content of the header; and 3) the use and format of sub- headers.

5.2
First, a decision was made regarding the size of each text unit. The NUD*IST manual (Qualitative Solutions and Research, 1994) instructs that a hard carriage return must be keyed at the end of each text unit. What is left up to the researcher to decide is its size. This is a very important issue as throughout the life of the project database, all segments of text coded, and therefore all segments of text retrieved, will be in multiples of text units. No segment of text smaller than one text unit can be coded or retrieved. Therefore, if one is likely to want to work with small segments, the unit of analysis should be small (perhaps a sentence). For this project, though, it was decided that each text unit would be made up of either the interviewer's question or the respondent's answer. So, every time the person speaking changed, the typist would press the hard carriage return key. This meant that, in most cases, any data retrieved would make sense - it would not be so short that no idea of the context of respondent's comments could be attained yet, by and large, it would not be so long as to contain more ideas than would be manageable.[2]

5.3
Second, NUD*IST works with what the manual calls headers (Qualitative Solutions and Research, 1994). Each document (and for this project, each interview was typed- up as a discrete document) was headed by information useful for its identification: the respondent's identity number, sex, age, and health problems; the date; duration of the interview; and a pseudonym (I found it easier to 'get to know' the transcripts if they were identified by a name, rather than by a number). As NUD*IST will print out lists of headers at the touch of a button, if one wishes to keep a track of the profile of the sample it is useful to include all desired information here.

5.4
Third, thought was given to the use of sub-headers. Retrieved text units are presented attached to the sub-header under which they appear. In this case, it was decided to use the question asked by the interviewer as a sub-heading in order that text units retrieved were placed in the context of this question. Although in many instances the answers young people gave could be understood very clearly on their own, it was often helpful to retrieve text in this 'paired question and answer' form (although NUD*IST does allow you to recontextualise any retrievals made). Each document, once transcribed, was saved as a 'Text Only' file and introduced into NUD*IST as a raw file.

The Indexing Process

6.1
NUD*IST organises data in a system of nodes, grouped together in a tree-structure (Qualitative Research and Solutions, 1994). Main categories ('nodes') are given 'parent' status, and sub-categories (also 'nodes') are 'children'. The result is a diagram which can be called up on the computer screen, with many branches at whose nodal points relevant data are stored.

6.2
The indexing process began at a very early stage, before the first document was introduced to NUD*IST. Headings from the interview schedule were used as nodal titles, as were concepts which had emerged from the literature review as being important to understanding the experience of illness in young people. Thus, among early nodes were 'impact on friendships' and 'thoughts on the future', and 'stigma' and 'infantilisation'. 'Special interest' nodes were also created at this point - for example one consultant, who was helping us to access cases, asked for anonymised comments on how his patients perceived his delivery of health-care; whilst one of the research team was particularly interested in whether media images of chronic illness impacted on young people's perceptions of themselves, and thus their experiences. Nodes were created to store data relevant to these areas.

6.3
Although it was not clear at this stage what respondents would tell us about their illness experience, the team was under no illusion that the interviews were being approached 'anew', or 'in a theoretical vacuum' as Mason (1996) puts it. Insights had been gained from the literature on chronic illness, and members of the research team (three clinicians - two psychiatrists and a G.P., and two social scientists - a sociologist and a psychologist) had their own disciplinary backgrounds as well as their own personal circumstances to bring to the understanding of the illness experience. Indeed, the content of the interview schedule was a product of all this. Charmaz's (1983), essentially sociological, work had been drawn on, for example, with her conceptualisation of the 'loss of self' seen as a potentially useful concept in examining the experiences of a young age group. The psychiatrists were more interested in understanding the experience in terms of mental health outcomes, with their medicalised view of suffering coming to the fore. A preliminary tree was thus drawn-up, based on a range of ideas and theories, and regarded as a useful starting point from which to consider the interviews.

6.4
It was understood, though, that the tree structure would not be used to constrain the ways in which the respondents' dialogue was considered. It was expected that new and more developed ideas would emerge from the interview transcripts, although it was always acknowledged that the 'baggage' described in the preceding paragraph would shape precisely what was coded and how. Kelle said:

Qualitative researchers ... always bring with them their own lenses and conceptual networks. They cannot drop them, for in this case they would not be able to perceive, observe and describe meaningful events any longer - confronted with chaotic, meaningless and fragmented phenomena they would have to give up their scientific endeavour (Kelle, 1997: ¶4.2).

6.5
In this way, he says, coding textual data is neither inductive nor deductive, but is based on qualitative induction and abduction. He explains: 'With qualitative induction a specific empirical phenomenon is described (or explained) by subsuming it under an already existing category or rule; whereas abductive inference helps find hitherto unknown concepts or rules on the basis of surprising and anomalous events' (Kelle, 1997: ¶4.4). Kelle's words aptly describe the way in which we viewed the analytical process here - we had 'theoretical preconceptions' (Kelle, 1997, ¶4.5), but revisions and modifications of these were possible, and to be expected, on immersion in the data.

6.6
With the first version of the tree structure in place, then, the first interview transcript (or 'document') was introduced to NUD*IST. This took place as soon as a transcript became available. This document, as with all those that followed, was read through in its entirety, and each section of the text of relevance in answering the main research question was deposited in the appropriate node(s). New coding categories were created as new ideas and ways of looking at the issues became apparent from examining the stories of the young people. Indeed, the preliminary tree structure expanded, to around ten times its original size, as the analytical process proceeded. It attempted to be as exhaustive as possible - making sure every illness-related detail was coded in some way, and that every potentially important idea was allocated a node.

6.7
NUD*IST's capacity to search texts was very useful in facilitating 'retrospective' coding. For example, after around 15 interview transcripts with asthmatics had been indexed, it became clear that the shortage of affordable nebulisers was a common concern. As this had not been anticipated, or picked-up during the first readings of transcripts, a simple search on the word 'nebuliser' was carried out on all the asthmatic documents and the results stored in a node of that name. This took around 30 seconds to do. On those occasions where it was not so straightforward to identify key words that encapsulated a new idea about the data, further indexing was added on retrieval of already coded text. Or, specific documents were recalled mentally, retrieved and read again in their entirety, and were indexed further in light of these new hunches.

6.8
Inevitably, some text units were indexed under many different nodal headings, whereas others were not indexed at all (except as base data - see below). One sentence uttered by a respondent could contain information relevant to understanding eight or nine concepts and headings, whereas a long monologue could be completely irrelevant to answering the research question. Some of the nodes contained hundreds of text units, coming from many of the interviews, while others contained only one or two, coming from, perhaps, a single interview.

6.9
In the case of these heavily utilised nodes, sub-categories were often created as ways in which to divide the data became apparent. This made retrievals more manageable. For example, so much data was stored in the 'experience of the NHS' node that once around half of the interviews had been analysed, retrievals of this category were taking an impractical amount of time to read. Thus, sub-categories were created, for example: hospitalisation, positive comments, and negative comments.[3]

6.10
For the project described here, the 'final' tree structure is very simple. At the end of the indexing process there were, at most, two levels of data, where parent nodes had been split into sub-categories to facilitate management and prevent unfeasibly long print-outs (although further levels were added during the write-up stage). Nodes were numbered as they were created, with nodes that may be related to each other often far apart diagrammatically. In essence, the tree structure was primarily a dumping ground for ideas and textual segments, with memos (discussed below) a forum for recording any ideas about linkages and relationships. Indeed, Richards herself talks of 'parking' nodes in the index system and warns of spending too much time making sure they 'belong' in a tidy system, pointing out that NUD*IST does not mind if they are never linked to other categories, though at any point nodes can be shifted, explored or deleted as the analyst sees appropriate (Richards, 1995: p. 123).

Base Data

7.1
Not only did concepts and broad ideas and headings have their own nodes, but what the NUD*IST manual calls 'base data' were indexed (Qualitative Solutions and Research, 1994). 'Base data' is the term used for the key characteristics relating to each interview - such as is the respondent male or female, how old is he/she, what are his/her health problems? Although one will probably have a good idea of what kind of data he or she wishes to code as base at an early stage, NUD*IST does not preclude additional characteristics being coded in this way at any time as new demographic or contextual factors of importance in better understanding the data become apparent. Indexing each document, in its totality, in this way makes it possible to cut across the data set on the basis of these characteristics when making retrievals.

7.2
For this project each transcript was indexed on the basis of the respondent's health problems (asthma, sequelae of head injury or mental disorder), sex (male or female), age (14, 15, 16, 17, 18, 19 or 20 years), and scores on the OSIQ (very low, low, normal, high or very high on each of the 12 sub-scores and on Total Self-Image). Each was seen as a potentially useful point at which the data-base could be cut across, thematically, to examine and better understand experiences. Questions such as: 'was it only young people with mental health problems who talked about stigma?', 'does body image tend to be a concern only to young women?' or 'are respondents with a high self-image found to have a different view of the future than respondents with a low self-image?' could be explored easily by using the base data nodes.

Memos

8.1
Another of NUD*IST's functions is memo-writing. This has been pointed to as useful in aiding theory-building (Weaver & Atkinson, 1994). Taking note of ideas as they occur to one throughout the analytical process has been seen as an integral part of much qualitative research. Glaser described a memo as 'the theorizing write-up of ideas about codes and their relationships as they strike the analyst while coding......it exhausts the analyst's momentary ideation based on data with perhaps a little conceptual elaboration' (Glaser, 1978: pp. 83 - 84). While using NUD*IST, such memos can be written at any point. One can attach these memos to nodes or documents. They can be free-standing, as simple memos, or can themselves be indexed and searched alongside the interview transcripts.

8.2
Here, after the coding of each transcript, a document memo was attached, recording thoughts about the case. These were largely impressionistic, drawing on the interview text itself, medical case notes, the results of the OSIQ and any field notes made at the time of the interview. Thus, lines from one memo read:

Brilliant interview but very sad case. Two months on we hear that she is now in hospital, close to death. She has a very great insight into her illness and identifies closely with others with anorexia. [Name of psychiatrist] comments on both these points in the case notes too. Clare talks a lot about her sister - there seems to be resentment there, Clare sees her as getting everything while she gets nothing. Similar sibling stuff to interviews with Stephanie and Julie [also suffering from anorexia nervosa]. Lot on family relationships and control.

8.3
This memo goes on to discuss briefly interesting points made by Clare, reproducing a couple of salient quotes and making further observations on how her case fits in with other interviews.

8.4
The length of the document memos varied. Those for the interviews that were first input to NUD*IST tended to be fairly brief, but as more interviews were read thoroughly and indexed memos became longer reflecting the further development of thoughts (including thoughts on the interview in the context of other interviews). As early interviews were revisited too, memos tended to be augmented. As such, these memos could be seen as a log of the research process as basic ideas became more sophisticated as time went on.

8.5
These memos were invaluable when it came to the later stages of the project, facilitating consolidation of thoughts on a variety of issues. Looked at together, or in sub-groups, they acted to spark ideas which could be investigated further by a closer look at individual interviews or nodes. They could also be used to identify quickly particularly interesting cases meriting further investigation, and helped avoid the problem of decontextualisation whereby text units become meaningless when seen on their own, sliced away from the interview as a whole and the circumstances in which the interview was carried out.

8.6
Nodal memos were also written. These documented certain patterns observed in the data - perhaps ideas about the illness experience for certain sub-groups, or hypotheses relating to particular concepts. For example, under the parent node 'medication' I had noted:

Steroids seem to be of particular concern in the case of asthma, see interview with Dr. Martyn Partridge. Talks about 'steroid phobia', says it often arises because those with asthma are not given sufficient verbal and written info to permit them to appreciate the 'risk versus benefit' argument in favour of inhaled steroids. It says that this is fault of communication between health professionals and people with asthma. Look at this point in relation to our data.

8.7
Such memos would often contain hypotheses or ideas, either generated by me (or another team member) on considering the data, or by something one of us had read. The intention was to go back to these and assess their utility again at a later stage. In the meantime the memo window attached to each node was a useful storage point for partially formed thoughts on the work in progress - another way of 'parking' ideas.

8.8
By the end of the indexing process, there were 165 nodes, all storing data relevant to answering the research question 'what is it like to be young and to have persistent health problems?'. Although NUD*IST does not preclude adding to this system at any time in the future, the research team was satisfied that this was as exhaustive as possible given the limited time available in which to write-up and disseminate results.

8.9
This indexing process was greatly facilitated by NUD*IST. At this point, the data-set was ready to be used for final analyses and write-up. It is this 'stage', the crux of the knowledge-building process, that will now be examined.

Writing-Up - Retrieving Indexed Data and Assimilating Ideas for Dissemination

9.1
The time eventually came for reviewing the final 'big picture' in order that the research could be written-up and disseminated. The interviews had now been indexed under the 165 category headings, each comprising a node. Three broad types of categorisations can be identified: descriptive, conceptual and base data. Each was used in a slightly different way at this stage.

9.2
'Money problems' and 'future worries' are both examples of descriptive nodes. Each respondent was asked to talk about both of these areas, with relevant answers placed in the relevant node. These nodes, then, contain a full record of what respondents had to say about these aspects of their lives.

9.3
The conceptual nodes, on the other hand, contain textual segments which had been interpreted by me as having some kind of common meaning that may not have been made explicit by the respondents, but which I thought were important. Examples of such nodes include 'stigma' and 'denial'. Although a very small number of respondents did actually use the words 'stigma' or 'denial' in talking about their experiences, these issues were never directly examined by the interviewer. A young woman with Chronic Fatigue Syndrome, for example, talked about how she and her mother initially refused to see a psychiatrist because she 'wasn't mad'. This was coded under 'stigma'. A young man with asthma insisted he could go running if he wanted to, even though his case notes and his account in an earlier section of the interview suggested he would find this extremely difficult. Relevant parts of this interview were coded as 'denial'.[4]

9.4
The importance of pointing out these different kinds of nodes lies in the need to use them slightly differently in analysis. I was asked, for example, what the young people had to say about money problems, and was able to retrieve everything in this node while being sure that all problems talked about were stored there. This retrieval comprised a simple log of the kinds of problems experienced, of the numbers of people in our sample who admitted they had such problems, and, when the headers were consulted, of the kinds of young people that talked about money being problematic in this way (ie. mostly those with mental health problems and/or those in their late teens).

9.5
Retrieving text units coded under 'denial' though did not provide a log in this way. I was less interested here in the numbers of cases which contained such indexing, as a percentage of the sample, than with the content. Retrievals were useful as a starting point for exploring the ways in which young people may play down or disregard symptoms, perhaps using denial as a coping strategy. The bringing together of text units with common themes in this way under 'stigma', 'denial' or 'infantilisation', for example, was enormously valuable in allowing me to explore many of the concepts common in the chronic illness literature.

9.6
Whether retrievals are conceptual or more descriptive, the issue of context is important. By coding text units and parking them in nodes, and later going back and looking at groups of these text units, one is at risk of giving these units a life of their own at the write-up stage. Each interview as a meaningful entity in itself may be forgotten as chunks with common themes from a variety of interviews are read and re-read on retrieval. This is where memos may be useful, written about the interview as a whole. It should also be remembered that just because NUD*IST has encouraged the slicing-up of interviews and division of these slices categorically, the original interview in its entirety is still there and can (and sometimes should) be used at the write-up stage (and if it is not seen as necessary to re- read the whole interview, NUD*IST can show the retrieved text units in the context of a specified number of surrounding text units).

9.7
The base data nodes were used in conjunction with the conceptual and descriptive nodes. Those used most were the illness nodes: asthma, mental disorder and head injury. In the case of money problems, for example, it was very useful to use illness as a way of cutting across the data in order to see whether the kinds of money problems experienced were different according to illness group as, indeed, they were. Some writers have argued that packages such as NUD*IST can lure the analyst into the trap of using a form of variable analysis inappropriately on qualitative data, by encouraging the matching of demographic and other variables with particular experiences (Mason, 1996). NUD*IST was used here, though, to facilitate the cutting across of nodes on the basis of interesting characteristics such as sex, illness and self-image score. Sectioning data in this way and examining it aided the fuller understanding of the lives of chronically ill adolescents. It was important to look at stigma as experienced by asthmatics separately from stigma as experienced by mentally ill young people, for example. The experiences of these two sets of young people were qualitatively different. Such a process of analysis is very different from any kind of quasi-variable analysis which seeks simply to reduce data and correlate it in an attempt to understand how variables are related.

9.8
Illness-based enquiry became very important. As was outlined above, one of the motivations for carrying out the project had been to try and find commonalities in experience across illness groups, on the basis of which new health-based initiatives could be built. It became increasingly apparent, though, as the data collection progressed and analysis proceeded, that the experiences of the young people were very much entwined with their particular diagnoses. They were 'asthmatics', or they were 'mentally ill', or they were 'head injured'. It was as such that the young people identified, the label 'chronically ill' meant little to them. Therefore, at this stage, many intersections of conceptual and descriptive nodes with the illness base data nodes were performed, saved as new 'children' nodes of the former, and explored further from here. This saving of retrievals is seen by Richards as part of NUD*IST's ability to enable theory discovery and construction 'allowing the researcher to build on and interrogate the results of past analyses, express and test hypotheses in processes of iterative theory-building and theory testing' (Richards, 1995: p. 111).

9.9
The demands of the clinicians with whom I was working played a part in how I approached 'writing-up'. I would be asked to write briefing papers on certain issues. These issues had invariably been flagged-up earlier in the project so a node existed. What was now required was to print out the contents of that node, draw together 'findings' and present these. Whether or not our data supported what had been found in other studies (usually more quantitatively-based) was something that was of particular interest to the clinicians involved, and findings were often presented framed in these terms.

9.10
In terms of writing-up interesting findings from the data-set more generally, for a variety of academic journals, enquiry was more grounded in the data itself. Certain themes or categories had come to my attention, on interaction with and immersion in the data, as being particularly salient to understanding the illness experiences of these young people. Stigma 'jumped out' as being incredibly important in understanding the day-to-day experiences of many of the young people with mental disorders; asthma as common, and 'not really an illness' was a theme that arose in many of the interviews with asthmatics and became a primary focus of enquiry; and the concept of 'sudden/dramatic change' became something that was investigated further as framing the experiences of many of the head injured young people.[5]

9.11
Whether responding to issues of interest to the clinicians involved, or following lines of enquiry that had emerged from the data the key question is how, precisely, I got from printing-out node retrievals to the presentation of findings? This is the crux of the analytical process, and the step that many writers fail to discuss. The utility of NUD*IST as a superior filing cabinet must be clear by now. Beyond that how useful was it?

9.12
Kelle points out: 'The process of coding the data is the preliminary for the actual analysis in which the analyst tries to make sense of the data' (Kelle, 1997: ¶15.5). Preliminaries over, now was the time to look for structures in the data and to try and explain these. With relevant print-outs in front of me, I had now to study what the young people had to say, making comparisons and noting similarities whilst observing differences and accounting for these. In this way, I had to construct theories. I will use an example from my data-set to try and illustrate how this worked in practice.

Constructing a Theory

10.1
I wanted to look at asthmatics attitudes to their medication. I had originally become interested in this aspect when investigating 'disregard of symptoms' amongst asthmatics and I wanted to discover whether or not such disregard might play a part in 'non-compliance' with medication regimes, as well as being interested in the whole issue of asthmatics' attitudes to medication in the broad context of how they perceived their symptoms. As part of this quest, I intersected the 'compliance' node (child of 'medication') with the base data 'asthma' node and printed out the result (other children of 'medication' were also looked at in this way). Many text units were retrieved, but to make this illustration clearer here are just three units from this retrieval:

* How well do you stick to taking your medication?

Before I went into hospital I was very bad. I didn't like the asthma - still don't - but I really grudged having the asthma. I used to be very very active. I took part in a lot of sports and some limitations really annoyed me so I just wanted to forget about it and get rid of it and I didn't take the inhalers properly, but I got a fright in hospital and now I take it just like that [Catherine].


* How well do you stick to taking your medication?

At first when I was getting used to it I used to forget to take it all the time and I wasn't taking it for ages and I was getting worse and all that and then once I started taking it regularly I just realised how much it helped so I just started taking it all the time [Maureen].


* So you always take your medication when you should?

Yes, the way I see it, if I didn't take it, you know. When I first got diagnosed as having asthma I would avoid taking my medicine. I would say 'yes Mum, I've took it, I've took it' until I ended up ill and I was in hospital and I thought I've got to get my act together and I started taking my medicine all the time, so from when I was about 12 I started taking it regularly. When I was younger I'd take a puff here and there cos I was quite embarrassed at first of having asthma cos I was left out of things at school because of it so I just didn't bother. It was after I took ill and going into hospital I thought I'd better stick to it [Britt].

10.2
These three segments, along with the many others retrieved, were scrutinised and coded again for subsidiary themes. The aim was to identify different dimensions present in order that a typology could be built. The three reproduced here were recoded under a node whose definition was 'medication not taken at first, but after bad experience(s) the regime is now followed'. Sister nodes were defined as follows: 'medication always taken properly', 'medication rarely taken properly', 'medication usually taken properly'. Within these 'types', reasons given by the asthmatics for their 'position' were examined - with deviant cases accounted for.

10.3
Going back to the three quotes again, then, all have in common the claim that medication was not taken as it should have been, the result was ill-health, and thereafter the regime was complied with as it was now seen as important in preventing illness. Other themes present differed between cases, though. Britt said she didn't used to take her medication because she was embarrassed, Catherine didn't take hers because she was in denial, and Maureen said she didn't take hers simply because she used to forget. So, although all were 'shocked' into taking their medication, reasons for non-compliance were different. Text units were further 'sorted' in terms of these kinds of differences.

10.4
The result was a better understanding of patterns of non-compliance amongst young asthmatics, with reasons for non-compliance explained with reference to the young people who took part in this study. Such an analysis has obvious implications for education campaigns targeted at the problem of young asthmatics not taking their medication as they should (the reason for too many unnecessary fatalities).

10.5
Let me go back to the role of NUD*IST in this - the step which took me from usefully ordered retrievals to theory. Of course, those usefully ordered retrievals were down to NUD*IST. It is doubtful I would have found time, using manual methods, to create such a comprehensive and systematic indexing system. In addition, NUD*IST allowed me to play around with the data however I wished, print-outs were always available quickly for me to scribble thoughts on, facilitating the development of theory in a highly organised and systematic way (I doubt manual methods would have enabled such a degree of flexibility). The thinking was all mine though. The human brain was needed to complete this process successfully. The superb organisational skills and facilities for automation offered by NUD*IST made mundane tasks easier to get through, but as Kelle said: 'the role of the computer remains restricted to an intelligent archiving ('code-and-retrieve') system, the analysis itself is always done by a human interpreter' (Kelle, 1997: ¶15.7). This is a point that is sometimes lost sight of in talking of software packages as 'overdetermining monsters', capable of the autonomous power to make methodological change.

Conclusion

11.1
That, then, is how NUD*IST was used in this project examining the experiences of chronically ill young people. It was made clear at the beginning of this paper that its purpose was not to look at what NUD*IST can do, but at what NUD*IST did do in this one study. The package can perform many more functions than those outlined above: command files can be written and used to automate indexing tasks further, for example; complex searches can be carried out, using a variety of Boolean operators; nodes can be merged, shifted and copied. Richards (1997), Richards (1995) and the manuals (Qualitative Solutions and Research 1994; Qualitative Solutions and Research, 1997) can be read to give the beginner a wider view of how NUD*IST can be used. It is hoped, though, that the above description may be useful to those wishing to be introduced to NUD*IST and its capabilities in a way that is more concrete than are abstract ideas about the management and analysis of text, by one who does not have a self-interest in promoting NUD*IST.[6]

11.2
On another level, though, it is hoped that the above description can be used in considering the recent advances on the methodological debate on CAQDAS. The remainder of the paper will consider further some of these recently discussed methodological issues, some of which have been touched upon in the main body of the paper.

11.3
The central thread running through the CAQDAS debate has been around the influence this software has on the way in which qualitative analysis is carried out. The extremes of this debate have been presented as: packages having the autonomous power to make methodological change, versus packages not changing anything at all about the way in which researchers work (Richards, 1997). Or, as Pfaffenberger (1988) stated these extremes several years earlier: CAQDAS as overdetermining monster versus neutral technical tool. The consensus is that packages such as NUD*IST are neither monstrous nor neutral, but that they affect some moderate degree of influence on the process of analysis.

Overdetermining Monster?

12.1
To what extent did NUD*IST affect methodological change in this project? Compare the above description of NUD*IST's utility with that of Richards' (1995) where she describes the way in which she and her team used the package on a project examining menopause and mid-life as a social construction. Richards' (1995) study was a large-scale team project, aiming to combine being systematic with a very large data-set with the discovery of themes normally associated with small-scale projects. Many of the issues she and her team had to consider differed from those described in the case of the chronic illness project: how to deal with over a thousand open-ended responses in a questionnaire, on top of all the other data collected; how to use NUD*IST in a large-team environment; what to do about the enormous tasks of factual coding. The sheer size of the project meant NUD*IST was used differently in several respects - symbols were inserted systematically in the text of documents, for example, to be searched on later should that information be needed; and matrices were used to see patterning in the huge data-set. The central issue confronting Richards (1995) and her team was resisting the temptation offered by NUD*IST's superior technical capabilities. Use of the package made the management of this huge quantity of data relatively easy, but the team came to realise that only a small part of these data could be analysed fully (the part of the analytical process that NUD*IST may be able to facilitate, but only the human brain can do).

12.2
I have recently begun to use NUD*IST on anther project.[7] It is a much bigger project than the chronic illness study, with different aims and different methods of data collection, and there is a far greater element of team-work. Like in the menopause project (Richards, 1995), searches will probably be used to a much greater extent, indexing all answers to particular standard questions together, for example, or searching for key words in order to save time. As this new project was preceded by two years pilot work, during which many key issues were identified, analyses will probably be conducted with a greater number of specific questions in mind, driven to a lesser extent by the data than was the case for the chronic illness project.

12.3
The key point these comparisons are intended to demonstrate is that NUD*IST is flexible - it can be used to support a range of studies. It has its constraints, of course. Coffey et al (1996: ¶7.1) are correct in saying it, like other similar packages, is based on the coding and retrieval model. Essentially, though, it is a resource to be used in a manner set-out by the researcher; not something which ensnares the researcher, dictating precisely how it should be used. Empirically, Lee & Fielding's (1995) findings in their work with users is testimony to this. They report that analysts were quick to abandon programme use when the software did nor meet their needs, or was challenging their epistemological position. Users are not mere slaves to the computer.

12.4
I said earlier that this paper is not attempting to review other packages but it is worth mentioning hypertext here as it has been discussed recently in this journal (Coffey et al, 1996; Lee & Fielding, 1996). The position set out in the previous paragraph can be held without disagreeing with Coffey et al's (1996) argument that hypertext has exciting possibilities. They also said 'unlike much of the code-and-retrieve computing this approach represents a genuine and generic advance over manual methods of data management' (Coffey et al, 1996: ¶7.4). Perhaps NUD*IST is not 'revolutionary' in this way, but why does something have to be a 'generic advance' in order to be immensely useful? Kelle (1997) was right to clarify what has often been misunderstood, or not articulated clearly enough, in some work on CAQDAS - NUD*IST and its ilk directly help 'analysis' only in the broad sense of the word. The preliminaries to theory-building are part of this broad analytical process - in the project described here I could not have built theory without first indexing the data, and NUD*IST was enormously helpful here The actual theory- building, though, could not have been done without human intelligence. NUD*IST, other similar packages, and, indeed, hypertext are all without utility in making the jump from ordered data to theories (and will be without such utility until they incorporate sophisticated artificial intelligence).

Neutral Tool?

13.1
It is simplistic, though, to think of NUD*IST as a neutral tool. Working with NUD*IST is not exactly the same as working using manual methods only. Just as with any technological innovation, the nature of work changes. Word processing packages, statistical packages, graphics packages all have particular structures and features which lead the user in certain directions (how many presentations at conferences now have in common a structure and interface recognisable as Powerpoint's, how few social science publications report statistical tests that may be appropriate but are not available on SPSS?).

13.2
Had manual methods been used solely it is certain that much longer periods would have been spent doing routine tasks. The flip-side of this, as Richards recognises is that the researcher using a package such as NUD*IST now has time to develop a 'coding fetishism', indexing anything and everything obsessively and unnecessarily - treating indexing as an end in itself and leaving little time for retrieval and 'real' analysis (Richards, 1997: p. 429). She also warns of the danger of 'keeping on going', ignoring common sense signs to limit the size of the data-set, driven on by the almost unlimited storage space, with no concrete limits such as the size of the office floor to act as a warning that an unmanageably large amount of data may be being gathered.

13.3
Lee & Fielding (1995) found, when studying users of qualitative data analysis packages such as NUD*IST, that defining text units as lines may lead one towards analyses that are too fine. Although the danger of this is lessened if one uses paragraphs as text units, relying on simple node retrievals of any size of text units, as can be churned out very easily by NUD*IST, may lead one to lose the context which may be all-important in understanding what is going on.

13.4
In these sorts of ways, NUD*IST changes the nature of components of the analytical process - often to the good, hastening routine and mundane tasks so that more time can be spent on the brain-work - sometimes to the bad if it is used without awareness of its limited power in facilitating theory-building.

13.5
In conclusion let me cite Richards: 'NUD*IST requires nothing, but invites a lot' (Richards, 1995: p. 120) - and Weitzman & Miles: 'you don't have to accept the invitation' (Weitzman & Miles, 1995: p. 243). What is important is for the researcher to be aware of why he/she is doing things, and to ensure that these things are not just being done because the package can do them, quickly and easily! The capabilities of the package should not be allowed to dictate the shape of any part of the analytical process - the researcher should have a clear idea of this process and then use the software to benefit it. Researchers should heed the advice of Stanley & Temple: 'all such packages produce epistemological effects and ... this in itself is no bad thing, so long as researchers are aware of it and choose software aids that do not undermine or vitiate their particular epistemological stance' (Stanley & Temple, 1995: p. 191). Moreover, packages such as NUD*IST should be seen for what they are - very useful, but powerless without a human brain leading them. As Kelle (1997: ¶6.3 ) pointed out: we do not fear an index card system taking over the way we work. It is worth remembering, then that packages such as NUD*IST, at the moment at least, have no more power to actually think than does that index card system.

Notes

1 Like the users in Fielding & Lee's (1995) study, informal networks of colleagues were our main source of information, and it was solely NUD*IST and The Ethnograph on which this information focused.

2 The study context is important in making a decision on the size of text units. What is desirable will depend very much on the style of interviewing and the talkativeness of the respondents. Respondents were young in this study, many were unused to talking about themselves in this way, and some were fairly inarticulate so continuous answers, uninterrupted by an interviewer's prompt, probe or further question, tended to be short (rarely longer than 10 sentences or around 20 lines). Furthermore, most of the time most of the respondents answered questions directly, not digressing from their main point.

3 This was not the only way in which the tree structure could have been used to organise the data. The manual, tutorial and some users have all pointed to the central tree structure as encouraging the researcher not only to arrange categories in relation to one another so that unmanageably large nodes can be split, but also so that a conceptual framework can be built that can be used, in itself, to shed light on how ideas and concepts are related.

4 Data in the descriptive and conceptual nodes are not mutually exclusive, there is much overlap. A young man's answer to the question about 'future worries', where he tells the interviewer how he fears he will never find a job because he has been labelled as 'head injured' and 'not right in the head', is indexed conceptually under 'stigma' in addition to being coded under 'future worries'.

5 Although it tended to be the case that the nodes of most interest at this stage were the fuller nodes, with the emptier ones being disregarded, this was not always the case. Whilst almost all of the respondents had something to say about hospital visits, for example, with the result that this node was very large, I did not feel that this data was of central importance in answering the research question. On the other hand, the stigma node was much smaller, but its contents were of central relevance to understanding the day-to-day lives of many of the mentally ill young people. Enquiry was directed by a knowledge of the data, built-up during the lengthy process of indexing, not simply by the size of nodes.

6 Although Lynn Richards, a co-creator of NUD*IST, has never claimed or pretended in her papers to be neutral in this matter and, indeed, has been critical of the package and her use of it on occasion.

7 Version 4 of NUD*IST became available early in 1997. It is not, essentially, different from Version 3, but has an improved interface, is more user-friendly in several other respects and has several new functions (including the ability to import and export from any table-based software). Had this version been available during the chronic illness project described here, little in the substance of the procedures described would have been altered.

Acknowledgements

I wish to thank Daniel Wight, Medical Research Council Medical Sociology Unit, Glasgow and Sue Scott, Department of Applied Social Sciences, University of Stirling, as well as the two anonymous reviewers for their helpful comments on earlier drafts of the paper.

References

BURGESS, R. G. (editor) (1995) Studies in Qualitative Methodology: Computing and Qualitative Research: Volume 5. London: JAI Press.

CHARMAZ, K. (1983) 'Loss of Self: A Fundamental Form of Suffering in the Chronically Ill', Sociology of Health and Illness, Vol 5, No 3, pp. 168 - 195.

COFFEY, A., HOLBROOK, B., and ATKINSON, P. (1996) 'Qualitative Data Analysis: Technologies and Representations', Sociological Research Online, vol. 1, no. 1, <http://www.socre sonline.org.uk/socresonline/1/1/4.html>.

DRASS, K. (1980) 'The Analysis of Qualitative Data: A Computer Program', Urban Life, vol. 9, no. 3, pp. 332 - 353.

FIELDING, N. G. and LEE, R. M. (editors) (1991) Using Computers in Qualitative Research. London: Sage.

FIELDING, N.G. and LEE, R.M. (1995) 'Confronting CAQDAS: Choice and Contingency', in R.G. Burgess (editor) Studies in Qualitative Methodology: Computing and Qualitative Research: Volume 5. London: JAI Press.

GLASER, B.G. (1978) Theoretical Sensitivity. Mill Valley: The Sociological Press.

HESSE-BIBER, S. (1995) 'Unleashing Frankenstein's Monster? The Use of Computers in Qualitative Research' in R. G. Burgess (editor) Studies in Qualitative Methodology: Computing and Qualitative Research: Volume 5. London: JAI Press Inc.

KELLE, U. (editor) (1995) Computer-aided Qualitative Data Analysis: Theory, Methods and Practices. London: Sage.

KELLE, U. (1997) 'Theory-Building in Qualitative Research and Computer Programs for the Management of Textual Data', Sociological Research Online, vol. 2, no. 2, <http://www.socre sonline.org.uk/socresonline/2/2/1.html>.

LEE, R.M. and FIELDING, N.G. (1995) 'User's Experiences of Qualitative Data Analysis Software', in U. Kelle (editor) Computer Aided Qualitative Data Analysis: Theory, Methods and Practice. London: Sage.

LEE, R.M. and FIELDING, N.G. (1996) 'Qualitative Data Analysis: Representations of a Technology: A comment on Coffey, Holbrook and Atkinson', Sociological Research Online, vol. 1, no. 4, <http://www.socre sonline.org.uk/socresonline/1/4/lf.html>.

MASON, J. (1996) Qualitative Researching. London: Sage.

OFFER, D., OSTROV, J. D., HOWARD, K. I. and DOLAN, S. (1992) Offer Self-Image Questionnaire, Revised (OSIQ-R) Manual. California: Western Psychological Services.

PATTON, M. Q. (1980) Qualitative Evaluation Methods. Beverly Hills, CA: Sage.

PFAFFENBERGER, B. (1988) Microcomputer Applications in Qualitative Research. Newbury Park, CA: Sage.

QUALITATIVE SOLUTIONS AND RESEARCH (1994) Nudist Revision 3.0 for Windows. Melbourne: Qualitative Solutions and Research.

QUALITATIVE SOLUTIONS AND RESEARCH (1997) QSR NUD*IST 4: User Guide. Victoria: Scolari.

RICHARDS, L. (1995) 'Transition Work! Reflections on a Three-Year NUD.IST Project', in R. G. Burgess (editor) Studies in Qualitative Methodology: Computing and Qualitative Research: Volume 5. London: JAI Press.

RICHARDS, L. (1997) 'User's Mistake as Developer's Challenge: Designing the New NUD*IST', Qualitative Health Research, vol. 7, no. 3, pp. 425 - 433.

RICHARDS, L. and RICHARDS, T. (1991) 'Computing in Qualitative Analysis: A Healthy Development?', Qualitative Computing, vol. 1, no. 2, pp. 234 - 262.

RICHARDS, T. J. and RICHARDS, L. (1994) 'Using Computers in Qualitative Research' in N. K. Denzin and Y. S. Lincoln (editors) Handbook of Qualitative Research. Thousand Oaks: Sage.

STANLEY, L. and TEMPLE, B. (1995) 'Doing the Business? Evaluating Software Packages to Aid the Analysis of Qualitative Data Sets' in R. G. Burgess (editor) Studies in Qualitative Methodology: Computing and Qualitative Research: Volume 5. London: JAI Press.

TESCH, R. (1990) Qualitative Research: Analysis Types and Software Tools. London: Falmer Press.

WEAVER, A. and ATKINSON, P. (1994) Microcomputing and Qualitative Data Analysis. Aldershot: Avebury.

WEITZMAN, E. and MILES, M. B. (1995) Computer Programs for Qualitative Data Analysis. Thousand Oaks, CA: Sage.

Copyright Sociological Research Online, 1997