Weitzman

Software and Qualitative Research

Readers are reminded that copyright subsists in this extract and the work from which it was taken. Except as provided for by the terms of a rightsholder's licence or copyright law, no further copying, storage or distribution is permitted without the consent of the copyright holder. The author (or authors) of the Literary Work or Works contained within the Licensed Material is or are the author(s) and may have moral rights in the work. The Licensee shall not cause or permit the distortion, mutilation or other modification of, or other derogatory treatment of, the work which would be prejudicial to the honour or reputation of the author. Weitzman, (2000) 'Software and Qualitative Research', Denzin, N.K. & Lincoln, Y.S., Handbook of qualitative research, 803-820, Sage Publications Ltd © This is a digital version of copyright material made under licence from the rightsholder, and its accuracy cannot be guaranteed. Please refer to the original published edition. Licensed for use at the University of Bath for the course: "Professional Doctorate In Health" during the period 08/03/2006 to 31/08/2006.

Permission reference: 0761915125(803-820)60276 ISN: 0761915125

30

SOFTWARE AND QUALITATIVE RESEARCH

• Eben A. Weitzman

he array of software available to support an introduction to and overview of the role of the work of qualitative researchers is ma- software in qualitative research, (b) a discussion T turing. A wide variety of useful tools are of the critical debates and concerns in the field now available to support many different ap- about the impact and appropriateness of using proaches to qualitative research. Most qualita- qualitative data analysis (QDA) software, (c) tive researchers can now find software that is ap- guidelines for choosing software to match indi- propriate to their analysis plans, the structure vidual needs, and (d) an indication of future di- of their data, and their ease-of-use and cost pref- rections for both scholarship on the use of QDA erences. However, making that appropriate software and development of such software. match still requires systematic analysis of the needs of the project and the researcher(s), and care- ful comparison of the software options available ♦ A Minihistory of the at the time of purchase with an eye kept fixed Use of Computers in firmly on those needs. There is still no one best program. Qualitative Research To help researchers understand what soft- ware can and cannot do to support their re- search efforts, understand both the potential Traditionally, qualitative researchers have car- benefits and pitfalls of using computers in quali- ried out the mechanics of analysis by hand: typ- tative research projects, and find software that is ing up field notes and interviews, photocopying suited to their needs, I provide in this chapter (a) them, “coding” by marking them up with mark-

AUTHOR’S NOTE: My thanks to Norman Denzin, Nigel Fielding, Udo Kelle, Ray Lee, and Morten Levin for their comments on an earlier draft of this chapter.

♦ 803 804 ♦ METHODS OF COLLECTING AND ANALYZING EMPIRICAL MATERIALS Software and Qualitative Research 805

ers or pencils, cutting and pasting the marked And there has also been divergence, as develop- has been an outpouring of journal articles, a se- retrieve is the sine qua non of qualitative analysis. segments onto file cards, sorting and shuffling ers look for new and different ways to ries of international conferences on computers (p. 175) cards, and typing up their analyses. This picture conceptualize support for analysis. and qualitative methodology, thoughtful books has been slowly changing since the early to There are now tools available that can help on the topic (Fielding & Lee, 1991, 1998; I address these issues at more length through mid-1980s. At that point, some researchers were researchers who are using a wide variety of re- Kelle, 1995; Tesch, 1990; Weitzman & Miles, much of this chapter, particularly in the subsec- beginning to use word processors for the typing search and analysis methodologies, from 1995b), and special journal issues (Mangabeira, tions below headed “False Hopes and Fears,” work, and just a few were beginning to experi- grounded theory to textual analysis to narrative 1996; Tesch, 1991). “Real Hopes,” and “Real Fears,” and in the later ment with database programs for storing and ac- analysis to interpretive interactionism. It is im- Periodically, commentators have raised con- section headed “Debates in the Field.” cessing their texts. Most qualitative methods portant to emphasize that software is not now, if cerns about whether the range of available soft- textbooks at the time (e.g., Bogdan 8c Biklen, it ever was, something that is relevant only to ware is dominated by a particular approach, 1982; Goetz & LeCompte, 1984; Lofland & “positivist” or “quasi-positivist” approaches to methodology, or epistemology (see, e.g., ♦ What Software Can Lofland, 1984; Miles & Huberman, 1984) qualitative research. If you see language in this Coffey, Holbrook, & Atkinson, 1996; Lonkila, and Cannot Do made little, if any, reference to the use of com- chapter that does not match your approach, you 1995). Although there is certainly room for fur- puters. may find it helpful to do some speculative trans- ther development to support certain specific In the early 1980s, a couple of programs de- lation. For example, if the discussion is about analytic processes (I offer some suggestions signed specifically for the analysis of qualitative “verification” or “hypothesis testing” and your later in this chapter, and the list appearing in the Simply put, software can provide tools to help data began to appear (Drass, 1980; Seidel & approach is postmodern, the discussion may chapter titled “Reflections and Hopes” in you analyze qualitative data, but it cannot do the Clark, 1984; Shelly & Sibert, 1985). Early pro- seem irrelevant. But it may be that there is a way Weitzman & Miles, 1995b, has only begun to analysis for you, not in the same sense in which a grams like QUALOG and the first versions of to understand the concept that makes sense be addressed), these concerns are clearly miss- statistical package like SPSS or SAS can do, say, The Ethnograph and NUD•IST reflected the from your perspective, such as “looking to see ing the mark. In this chapter, I suggest a wide multiple regression. Many researchers have had state of computing at that time. Researchers typ- whether there is more material supporting, or variety of types of programs that are available the hope—for others it is a fear—that the com- ically accomplished the coding of texts (tagging contradicting, a certain assumption or interpre- to support a wide variety of research ap- puter could somehow read the text and decide chunks of text with labels—codes—that indicate tation.” The same software tools that someone proaches. Qualitative researchers are not lim- what it all means. That is, generally speaking, not the conceptual categories the researcher wants else might use for classical hypothesis testing ited only to coding-oriented programs, or even the case.1 Thus it is particularly important to em- to sort them into) by typing in line numbers and might be very useful for your purposes. to programs explicitly marketed to qualitative phasize that using software cannot be a substitute code names at a command prompt, and there Many programs now allow the researcher to researchers. For example, as Fielding and Lee for learning data analysis methods: The re- was little or no facility for memoing or other an- specify relationships among codes and use these (1998) point out, there are a variety of options searcher must know what needs to be done, and notation or markup of text. In comparison to relationships in analysis, and to write memos for those wishing to follow the suggestion of do it. The software provides tools to do it with. marking up text with colored pencils, this felt and link them to text and codes. Some programs Coffey and Atkinson (1996) that text retrievers The following are some of the things comput- awkward to many researchers. And computer allow the researcher to create links between dif- may be more helpful for discourse analysis than ers can be used for to facilitate the analysis pro support for the analysis of video or audio data ferent points in the text (hypertext), and a small code-and-retrieve programs. Fielding and Lee cess: was at best a fantasy. but growing handful allow the use of audio and go on to argue that But the landscape has changed dramatically, video in place of, or in addition to, text. And in terms of both software and the literature de- there are a variety of approaches to linking cate- developers of CAQDAS [computer-aided quali- 1. Making notes in the field; voted to it. By the time the late Matt Miles and I gorical and quantitative data (e.g., demograph- tative data analysis software] programs have in- 2. Writing up or transcribing field notes; wrote Computer Programs for Qualitative Data ics, test scores, quantitative ratings) to text and creasingly included facilities for proximity 3. Editing: correcting, extending, or revis- Analysis (Weitzman & Miles, 1995b), we re- for exporting categorical and quantitative data searching, which might be useful for narrative ing field notes; viewed no fewer than 24 different programs that (e.g., word frequencies or coding summaries) to analysis, and for “autocoding” which could be adapted to some kinds of semiotic analysis. The 4. Coding: attaching key words or tags to were useful for analyzing qualitative data. Half quantitative analysis programs for statistical provision of new features in CAQDAS pro- segments of text, graphics, audio, or of those programs had been developed specifi- analysis. Finally, there are now some free pro- grams reflects the generally close relationship video to permit later retrieval; cally for qualitative data analysis, whereas the grams available, notably two from the U.S. Cen- between users and developers characteristic of 5. Storage: keeping text in an organized the field, and the general willingness of devel- other half had been developed for more gen- ters for Disease Control: EZ-Text, which focuses database; eral-purpose applications, such as text search on qualitative surveys, and AnSWR, intended opers to incorporate features desired by users even if these do not always accord with the 6. Search and retrieval: locating relevant and storage. Since then, the field has continued for a more general range of qualitative data. The epistemological preferences of the developer. segments of text and making them avail- to grow rapidly. Programs are being revisesd at a software continues to vary widely, and it remains Since packages increasingly support proce- able for inspection; regular rate, new programs appear on the scene very much the case that there is no one best pro- dures, routines and features which are new to 7. Data “linking”: connecting relevant data at the rate of one or two a year, and programs gram for all needs. qualitative analysis or make procedures possible segments to each other, forming catego- that don’t find users disappear. There has been In parallel with the growth in software, liter- that were not practicable without the power of the computer, it is less and less plausible either ries, clusters, or networks of information; some convergence as good features in one pro- ature reporting studies and commenting on the to argue that the software is merely an aid to 8. Memoing: writing reflective commentar- gram are imitated by the developers of others. software has begun to appear regularly. There code-and-retrieve or to argue that code-and- ies on some aspect of the data, theory, or method as a basis for deeper analysis; 806 ♦ METHODS OF COLLECTING AND ANALYZING EMPIRICAL MATERIALS Software and Qualitative Research ♦ 807

9. Content analysis: counting frequencies, Others express this more as a hope that if they analysis can be run in minutes or seconds on a sequences, or locations of words and buy the right program, they will not have to en- Consistency. Software can help with consis- desktop computer. As a result, researchers can phrases; gage in the often very time-consuming process tency. If I can search for all the places a given run factor analyses much more often, as part of 10. Data display: placing selected or re- of analyzing all that text themselves. QDA soft- key word appears, or all the places where a other analyses rather than only as major under- duced data in a condensed, organized ware provides tools that help you do these given code or combination of codes was ap- takings of their own, and on multiple sets of format, such as a matrix or network, for things; it does not do them for you. plied, or always see the relationship between scores in the same project. The speed of the com- inspection; In an extension of this concern, many re- two features of the data that I have recorded, it puter alone can change what researchers even 11. Conclusion drawing and verification: searchers have worried about the software going becomes possible for me to be more consistent contemplate undertaking. aiding in the interpretation of displayed yet a step further and “building theory.” But, as in a couple of ways. I can be much more careful data and the testing or confirmation of Miles and I also argued in 1995, “Software will about not missing the data that contradict my Representation. Software that allows dynamic, findings; never ‘do’ theory building for you ..., but it can brilliant, but wrong, new hypothesis. I can eas- real-time representation of a researcher’s think- 12. Theory building: developing systematic, explicitly support your intellectual efforts, mak- ily review all the data I assigned to a given con- ing can be a substantial aid to theorizing. Soft- conceptually coherent explanations of ing it easier for you to think coherently about the ceptual category or theme and check to see if ware that provides a graphic map of relationships findings; testing hypotheses; meaning of your data” (Weitzman & Miles, they (a) all belong together and (b) still seem to among codes, text segments, or cases can help re- 13. Graphic mapping: creating diagrams that 1995b, p. 330). support the interpretation I started out with; if searchers to visualize and extend their thinking depict findings or theories; This situation may change in the coming not, I can easily reorganize. (Note that the prob- about the data or theory at hand. Researchers of- 14. Report writing: interim and final years. There are some current efforts to use arti- lem of my making bad interpretations has not ten use drawings to depict these relationships, (adapted from Miles & Huberman, ficial intelligence (AI) approaches to get com- been removed. But the kinds of facilities men- but software can keep maps tied to the underly- 1994, p. 44). puters to interpret text. For example, the SPSS tioned here can be tremendously helpful to ing project, so that changes to the links in the module TextSmart uses information about fre- competent researchers in checking their own drawing change the links among the objects in the Obviously, many of these are things that re- quency of occurrence of words and proximity of work, as well as in allowing colleagues or re- database, and vice versa. searchers can do with a word processor. Other words to each other to categorize text responses search participants to check it and provide feed- software, which is the focus of this chapter, helps automatically. Microsoft Word97 has an “Auto back.) Consolidation. Finally, allowing the researcher to with the other tasks. The developments seen in Summarize” feature that aims to identify the record field notes, interviews, codes, memos, an- recent years have made it possible for research- most important “concepts” in a document ac- Speed. The speed of computers is a critical issue notations, reflective remarks, diagrams, audio ers to do these things more and more easily, and cording to word frequency. Other developers in making QDA software helpful. First, a cau- and video recordings, demographic variables, more and more powerfully. In the section are thinking about using AI techniques to get tion: It can take time to learn to use a program, and structural maps of the data and the theory all headed “Types and Functions of Software for software to participate in the theory-building and once you have, it can take some time to pre- in one place can be a tremendously powerful sup- QDA,” below, you will find more specific details process with researchers. These approaches rely pare and set up the data for analysis. But once port to the analysis process. If the design of the that is done, the speed of the computer quickly about what software can do. But first, consider on things like frequency to indicate importance program is such that it allows the researcher to pays for that investment. Being able to search some of the hopes and fears, both real and false, and proximity in the text to indicate relatedness. move from one intellectual activity to another and re-search almost instantaneously encour- that people have concerning QDA software. For some qualitative researchers these are ac- with minimal effort, and carry over the results of ages the researcher to conduct multiple ceptable assumptions, but for many others they one sort of thinking to others, it can both free up searches to zero in on the data that really apply are not, and for such researchers the results of large amounts of energy for the critical tasks and False Hopes and Fears to a particular question. Being able to quickly these approaches do not yield useful interpreta- help the researcher to see and keep track of con- re-sort a database, redefine codes, and reassign tions of text. nections that might otherwise easily fall through In Weitzman and Miles (1995b), we argue: chunks of text enables and encourages the re- the cracks. seatcher to revise the analysis and the thinking As Pfaffenberger ... points out, it’s equally naïve about it whenever necessary. Being able to to believe that a program is (a) a neutral technical Real Hopes Real Fears tool or (b) an overdetermining monster. The is- quickly pull together all the text for cells in a sue is understanding a program’s properties and complex matrix display enables and encourages What do we really have to be worried about? presuppositions, and how they can support or What can we really expect to gain from the the researcher to run down provocative leads Many of the advantages touted above have flip constrain your thinking to produce unantici- use of software? QDA software provides tools and new ideas—as well as worries that the cur- pated effects. (p. 330) sides. The very ease, speed, and power of the soft- for searching, marking up, linking, and reorga- rent conclusions may be way off track—much ware have the potential to encourage the kind of nizing the data, and representing and storing more often and with much less cost. thinking I have referred to as “false hopes and As already mentioned, many people apparently your own reflections, ideas, and theorizing. An example from quantitative research may fears” above. Although the software will not fig- continue to believe that QDA software intends Some of it gives you tools for further explora- be instructive. In the days of slide rules, and ure out what a complex account of a childhood to do the data analysis. Skeptical researchers tion—which in some cases might amount to hy- even of handheld calculators, before statistical trauma really means in the context of the current raise challenges to the notion of “dumping my pothesis testing or conclusion verification— software was available, doing factor analysis study, the ease of searching for key words and text into a program and seeing what comes out.” based on your theorizing and interpretive work. was a months-long enterprise. Now a factor

808 ♦ METHODS OF COLLECTING AND ANALYZING EMPIRICAL MATERIALS Software and Qualitative Research ♦ 809

“autocoding” them may encourage the re- ♦ Types and Functions of They may allow you to define fields in the fixed to categorize or code your memos) or more so- manner of a traditional database such as phisticated search-and-retrieval functions than searcher to take shortcuts. We may fail to check Software for QDA to see what passages were actually coded in the Microsoft Access® or FileMaker Pro®, or they code-and-retrieve programs. They may have ex- autocoding process, and to use their own intelli- may allow significantly more flexibility, for ex- tended and sophisticated hyperlinking features, gence to analyze whether they fit. There is the ample, allowing different records to have dif- allowing you to link segments of text together In this section I offer a rough sorting of available real potential that we will get lazy. As Lee and ferent field structures. Their search operations or to create links among segments of text, graph- software into types. There is naturally quite a bit Fielding (1991) have noted, “There is the possi- may be as good as, or sometimes even better ics, photos, video, audio, Web sites, and more. of overlap among categories, with individual than, those of some text retrievers. Examples of They may also offer capabilities for “system clo- bility that the use of computers may tempt quali- programs having functions that would seem to textbase managers are askSam, Folio Views, sure,” allowing you to feed results of your ana- tative researchers into ‘quick and dirty’ research belong to more than one type. However, it is Idealist, InfoTree32 XT, and TEXTBASE AL- lyses (such as search results or memos) back with its attendant danger of premature theoreti- possible to focus on the “heart and soul” of a PHA. into the system as data. Examples of code-based cal closure” (p. 8). program: what it is mainly intended for. This theory builders are AFTER, AnSWR, AQUAD, It is also possible that the availability of soft- categorization scheme was first presented in ATLAS/ti, Code-A-Text, HyperRESEARCH, ware may tempt researchers to skip over the pro- Code-and-Retrieve Programs Weitzman and Miles (1995b). NUD●IST, NVivo, QCA, the Ethnograph, and cess of learning properly about research. Again winMAX. Two of these programs, AQUAD and from Lee and Fielding (1991): Code-and-retrieve programs are often de- QCA, support cross-case configural analysis Text Retrievers veloped by qualitative researchers specifically (Ragin, 1987), QCA being dedicated wholly to Of course, the ultimate fear here is of Franken- for the purpose of qualitative data analysis. The Text retrievers specialize in finding all the in- this method and not having any text-coding capa- stein’s monster. It is susceptible to the same cave- programs in this category specialize in allowing bilities. ats, too. Like the monster, the programs are mis- stances of words and phrases in text, in one or you to apply category tags (codes) to passages of understood. The programs are innocent of guile. several files. They typically also allow you to text and later retrieve and display the text ac- It is their misapplication which poses the threat. search for places where two or more words or Conceptual Network Builders It was exposure to human depravity which made cording to your coding. These programs have at phrases coincide within a specified distance (a a threat of Frankenstein’s creation. Equally, the least some search capacity, allowing you to untutored use of analysis programs can certainly number of words, sentences, pages, and so on) search either for codes or words and phrases in Conceptual network builders are programs produce banal, unedifying and off-target analy- and allow you to sort the resulting passages into the text. They may have a capacity to store that emphasize the creation and analysis of net- ses. But the fault would lie with the user. This is different output files and reports. They may do memos. Even the weakest of these programs work displays. Some of them are focused on al- why teaching the use of the programs to novice other things as well, such as content analysis lowing you to create network drawings: graphic researchers has to be embedded in a pedagogy represent a quantum leap forward from the old functions like counting, displaying key words in representations of the relationships among con- which has a sense of the exemplars of qualitative scissors-and-paper approach: they’re more sys- analysis, rather than as skills and techniques to be context or creating concordances (organized tematic, more thorough, less likely to miss cepts. Examples of these are Inspiration, Meta- mechanically applied, (p. 8) lists of all words and phrases in their contexts), things, more flexible, and much, much faster. Design, and Visio. Others are focused on the or they may allow you to attach annotations or Examples of code-and-retrieve programs are analysis of cognitive or semantic networks, for example, the program MECA. Still others offer The final fear that has some truth to it is that even variable values (for things like demograph- HyperQual2, Kwalitan, QUALPRO, Martin, some combination of the two approaches, for ex- the conceptual assumptions behind the pro- ics or source information) to points in the text. and the Data Collector. ample, SemNet and Decision Explorer. Finally, gram—for example, that the relationships Examples of text retrievers are Sonar Profes- ATLAS/ti, a program also mentioned above un- among codes are always strictly hierarchi- sional, the Text Collector, and ZyINDEX; there Code-Based Theory Builders der code-based theory builders, also has a fine cal—will shape the analysis. This fear both has are also a variety of free (but hard to use) GREP graphical network builder connected to the ana- truth to it and is often overstated. For example, tools available on the World Wide Web. Most of the code-based theory-building pro- lytic work you do with your text and codes. if the program allows you to directly represent grams are also based on a code-and-retrieve hierarchical relationships among codes, but not Textbase Managers model, but they go beyond the functions of nonhierarchical relationships, such as circular code-and-retrieve programs. They do not, nor Summary loops or unstructured networks, it will probably Textbase managers are database programs would you want them to, build theory for you. encourage you to think primarily or only in specialized for storing text in more or less orga- Rather, they have special features or routines In concluding this discussion of the five main terms of hierarchical relationships among your nized fashion. They are good at holding text, to- that go beyond those of code-and-retrieve pro- software family types, I want to emphasize that codes/concepts. If you are aware of the assump- gether with information about it, and allowing grams in supporting your theory-building ef- functions often cross type boundaries. For exam- tions behind a program, you have a couple of op- you to quickly organize and sort your data in a forts. For example, they may allow you to rep- ple, Folio VIEWS can code and retrieve, and tions: You can choose another program or you variety of ways and retrieve it according to dif- resent relations among codes, build has an excellent text search facility. ATLAS/ti, can find a way to work around the assump- ferent criteria. Some are better suited to highly higher-order classifications and categories, or NUD•IST, NVivo, the Ethnograph, and tions in the program—for example, by keeping structured data that can be organized into “re- formulate and test theoretical propositions winMAX graphically represent the relation- an ever-changing, nonhierarchical code map cords” (that is, specific cases) and “fields” (vari- about the data. They may have more powerful ships among codes, although among these, only pinned to the wall. More on this in the section ables—information that appears for each case), memoing features (allowing you, for example, ATLAS/ti allows you to work with and manipu- whereas others easily manage “free-form” text. headed “Debates in the Field,” below. 810 ♦ METHODS OF COLLECTING AND ANALYZING EMPIRICAL MATERIALS Software and Qualitative Research ♦ 811

late the drawing.2 The Ethnograph and win- program for a school ethnography? or, What’s such as the one proposed in Weitzman & Miles, Internet. People at different levels seem to have MAX both have systems for attaching variable the best program for doing grounded theory? or, 1995b, which has rows for each of the questions quite different reactions to the same programs. values (text, date, numeric, and so on) to text What’s the best program for analyzing focus below and columns for answers, implica- So, for example, a person at Level 2 or 3 might files and/or cases. Sphinx Survey allows you to groups? None of these questions has a good an- tions/notes, and candidate programs.) Work like a program that puts the maximum informa- work with survey data consisting of a mix of swer. Instead, analysts need to approach choice your way from answering questions, to the im- tion on one screen because this allows her to find qualitative and quantitative data. The implica- based on the structure of the data, the specific plications of those answers for program choice, what she wants quickly, and she might learn the tion: Do not decide too early which family you things they will want to do as part of the analysis, to candidate programs. For example, if you are program very quickly. A person at Level 1 might want to choose from. Instead, stay focused on and their needs around issues such as ease of use, working on a complex evaluation study, with a find all that information overwhelming at first, the functions you need. cost, time available, and collaboration. combination of structured interviews, focus and might take a little longer to learn that pro- Researchers can ask themselves four broad groups, and case studies, you will need strong gram because of it. But, once he has learned the Multimedia. Multimedia capabilities are just be- questions, as well as consider two cut-across is- tools for tracking cases through different docu- program, our Level 1 person would probably ginning to emerge as a significant issue in soft- sues, to help guide their choices (Weitzman & ments. You might find good support for this in a benefit from the layout in the same way as the ware choice. There are now several programs in Miles, 1995a, 1995b). These guidelines for program’s code structures, or through the use more advanced computer user. the code-based theory builder category that al- choice have seen wide use in practice since their of speaker identifiers that track individuals low you to use audio and video, as well as text, as original formulation, and have proven to be ef- throughout the database (see Question 3, be- Question 2: Are You Choosing for One data: AFTER, ATLAS/ti, and Code-A-Text all al- fective for guiding researchers to appropriate low). Such suggestions are elaborated below. Project or for the Next Few Years? low you to code and annotate audio and video choices. Because this approach to choice empha- files, and search and retrieve from them, in ways sizes matching functions, rather than specific Question 1: What Kind of A word processor does not care what you are quite similar to the ways they let you manipulate programs, to particular needs, these guidelines Computer User Are You? writing about, so most people pick one and stick text, as does version 2 of HyperRESEARCH, can continue to be useful long after the programs with it until something better comes along and which is under development at the time of this referenced here as examples have evolved into they feel motivated to learn it. But particular writing. In these programs, you can play a media new versions and new programs have arrived on Your present level of computer use is an im- qualitative analysis programs tend to be good for file (audio or video), mark the beginning and the scene.3 portant factor in choice of a program. If you are certain types of analyses. Switching will cost you ending points of segments, and then treat those Specifically, there are four key questions you new to computers, your best bet is probably to learning time and money. Think about whether segments much like segments of text. A program need to ask and answer as you move toward choose a word-processing program with advice you should choose the best program for this pro- now in beta testing called InterClipper is de- choosing one or more software packages: from friends and begin using it, learning to use ject or the program that best covers the kinds of signed primarily for audio files, with the as- your computer’s operating system (e.g., MS- projects you are considering over the next few sumption that you only bother to transcribe the 1. What kind of computer user am I? DOS, Windows, or Mac) and getting com- years. For example, a particular code-and- segments that you find most important (it is tar- 2. Am I choosing for one project or for the fortable with the idea of creating text, moving retrieve program might look adequate for the geted at focus group researchers in commercial next few years? around in it, and revising it. That would bring current project and be cheaper or look easier to environments who need to be able to generate 3. What kind of project(s) and database(s) you to what we’ll call Level 1. Or you may have learn than some other program. But if you are analyses and reports quickly). This program will will I be working on? gotten acquainted with several different pro- likely to need a more fully featured code-based probably fall in the code-and-retrieve family 4. What kinds of analyses am I planning to grams, use your operating system easily, and theory builder down the road, it might make when it is ready for release. There is also a grow- do? feel comfortable with the idea of exploring and more sense to get started with one of those now ing field of software dedicated exclusively to learning new programs (Level 2). Or you may (assuming you choose one that includes good managing video. In addition to these four key questions, there are be a person with active interest in the ins and code-and-retrieve capabilities). two cut-across issues to bear in mind: outs of how programs work (Level 3) and feel easy with customization, writing macros, and Question 3: What Kind of Database ♦ How to Make Intelligent, • How important is it to you to maintain a the like. (I will not deal here with the “hacker,” and Project Will You Be Working On? Individualized Software Choices sense of “closeness” to your data? a Level 4 person who lives and breathes com- • What are your financial constraints when puting.) buying software and the hardware it needs Being more of a novice does not mean you Here the questions begin to get a bit more spe- to run on? have to choose a “baby” program, or even that cific. As you look at detailed software features, I have emphasized from the beginning of this you shouldn’t choose a very complex program. you need to play them against a series of detailed chapter that there is no one best software pro- With these basic issues clear, you will be able It does mean, however, allowing for extra issues. Because of the nature of computers, it be- gram for analyzing qualitative data. Further- to look at specific programs in a more active, de- learning time, perhaps placing more emphasis comes essential to give careful attention to the is- more, there is no one best program for a particu- liberate way, seeing what does or does not meet on user-friendliness, and finding sources of sup- sue of understanding the nature and structure of lar type of research or analytic method. your needs. (You may find it helpful to organize port for your learning, such as friends or col- qualitative data sets. The issues here have to do Researchers will sometimes ask, What’s the best your answers to these questions on a worksheet, leagues, or on-line discussion groups on the with the physical and logical form of the data:

812 ♦ METHODS OF COLLECTING AND ANALYZING EMPIRICAL MATERIALS Software and Qualitative Research ♦ 813

how structured and how consistent it is, how schools, which in turn might be nested in dis- narrow and stern in their requirements. If you methodology requires. If you are doing narrative data about a case are organized, and so on. In tricts. Look for software that will easily select will have diverse entries, look for software de- analysis you may need to track temporal or narra- terms of the issues presented below, there may be different portions of the database, and/or do signed to handle multiple sources and types of tive structures in certain ways. Or you may be fo- great variation from project to project, even configurational analysis (Ragin, 1987) across data, with good source tags and good linking cusing on building a web of hypertext links as a within a given analytic approach (say, grounded your cases; software that can help you create features in a hypertext mode. The ability to way of understanding the phenomena in your theory, ethnography, or narrative analysis). multiple-case matrix displays, usually by gather- handle “off-line” data—referring you to mate- data, and your needs may be better served by one Epistemological issues, such as the interpretive ing together the data that correspond to the dif- rial not actually loaded into your program—is a or another way of creating and representing that nature of observational notes, coding, or ferent cells of the matrix, is also useful. plus. Many programs can be tricked into doing hypertext web. Coding is probably the best-sup- memos, or the social construction of interview this if you are clever about it. If you want to be ported approach at the current writing, and data, although very important for research Fixed records versus revised. Will you be working able to code, and then retrieve, audio or video, many researchers who use other approaches may methodology, do not come into play here; the with data that are fixed (such as official docu- look for programs like AFTER, ATLAS/ti, find that their best option is to use a “coding” sys- question is whether the program you choose ments or survey responses) or data that will be Code-A-Text, and InterClipper, which let you tem for their own purposes—for example, to provides the organizational tools for the text, revised (with corrections, added codes, annota- treat these media much like text. mark up the narrative structure of a text. More graphics, audio, or video you want to put into it. tions, memos, and so on)? Some programs make on this in the section headed “The Future,” be- database revision easy, whereas others are quite Size of database. A program’s database capacity low. Data sources per case: single versus multiple. You rigid, so that revising can use up a lot of time and may be expressed in terms of numbers of cases, The subsections below, laying out the parts of may be collecting data on a case from many dif- energy. Some will let you revise annotations and numbers of data documents (files), size of indi- Question 4, should help you move beyond just ferent sources (say your case is defined as a stu- coding easily, but not the underlying text, and vidual files, and/or total database size, often ex- the name of your methodology. They should help dent, and you talk with several teachers, the stu- some will let you revise both. Although this has pressed in kilobytes (K) or megabytes (MB). you identify the specific analytic moves you will dent’s parents and friends, the principal, and the been a constraining issue up to now, the trend in (Roughly, consider that a single-spaced page of need to make; the specific operations you will student herself). Some programs are specifically new programs and upcoming revisions of exist- printed text is about 2 to 3K.) Estimate your to- need to perform; the kinds of insights, infer- designed to handle data organized like this, oth- ing ones is toward programs that allow you to tal size in whatever terms the program’s limits ences, and interpretations you will need to re- ers are not designed this way but can handle mul- edit underlying text easily. are expressed, and at least double it. Most pro- cord; and the manner in which you plan to record tiple sources pretty well, and some really do not grams today are generous in terms of total data- them. In other words, they should help you to get have the flexibility you’ll need. As mentioned Structured versus open. Are your data strictly or- base size. A few are still stingy when it comes to specific about the things you would do if working above, you should look for strong tools for ganized (for example, responses to a standard the size of individual texts. For example, in with paper and to translate these into the func- tracking cases through different documents. questionnaire or interview) or free-form (run- some programs when a document goes beyond tions you will need from software. Some programs provide good support for this in ning field notes, participant observation, and so about 10 pages, the program will insist on code structures (particularly programs with on)? Highly organized data can usually be more breaking it into smaller chunks or will open it Exploratory versus confirmatory. Are you mainly highly structured code systems, like NUD•IST easily, quickly, and powerfully managed in pro- only in a “read-only mode” browser.4 planning to poke around in your data to see what and NVivo, and to lesser degrees in programs grams set up to accommodate them—for exam- they are like, evolving your ideas inductively? Or with flexible code systems like ATLAS/ti) or do you have some specific hypotheses in mind ple, those with well-defined “records” for each Question 4: What Kind of Analyses through the use of speaker identifiers in pro- case and “fields” (or variables) with data for each linked to an existing theory to test deductively? If Are You Planning to Do? grams like AFTER, the Ethnograph, or record. Structured surveys may benefit from sur- the former, it is especially important that you Code-A-Text. Also, look for programs that are vey-oriented programs like Sphinx Survey or have features of fast and powerful search and re- good at making links, such as those with hyper- EZ-Text, which take advantage of predictable As mentioned above, identifying the name trieval, easy coding and revision, along with good text capability, and that attach “source tags” tell- structure to provide good data-manipulation of your analysis methodology really won’t do text and/or graphic display. ing you where information is coming from. tools. Free-form text demands a more flexible the trick here. Your choice of software depends If, on the other hand, you have a beginning program. There are programs that specialize in on how you expect to go about analysis. This theory and want to test some specific hypotheses, Single versus multiple cases. If you have multiple one or the other type of data, and some that does not mean a detailed analysis plan, but a programs with strong theory-building and -test- cases, you usually will want to sort them out ac- work fairly well with either. general sense of the style and approach you are ing features are better bets. Look for programs cording to different patterns or configurations, expecting, which in turn will tell you the kinds that test propositions, or those that help you de- and/or work with only some of the cases, and/or Uniform versus diverse entries. Your data may all of things you will need to be able to do with the velop and extend conceptual networks. do cross-case comparisons. Multicase studies come from interviews, or you may have infor- data. For an excellent overview of a range of ap- can get complicated. For example, your cases mation of many sorts: archival documents, field proaches to qualitative data analysis, and a dis- Coding scheme firm at start versus evolving. Does might be students (and you might have data observations, questionnaires, pictures, audio- cussion of some of the procedures associated your study have a fairly well defined a priori from multiple sources for each student). Your tapes, videotapes (this issue overlaps with single with them, see Fielding and Lee (1998, chap. 2). scheme for codes (categories, key words), per- students might all be “nested” in (grouped by) versus multiple sources, above). Some programs If you will be coding your data, you need a haps theoretically derived, that you will apply to classrooms, which might be nested within handle diverse data types easily, and others are program that will let you code the way your your data? Or will such a scheme evolve as you

814 ♦ METHODS OF COLLECTING AND ANALYZING EMPIRICAL MATERIALS Software and Qualitative Research ♦ 815

go, in a grounded theory style, using the “con- segments when you first import the data, ingful way (in terms of the analytic approach counts for educational users and multiple-user stant comparative” method (Glaser, 1965; Glaser whereas others let you mix and match chunk you are taking). For example, do you need to be “site licenses.”) In addition, programs vary a lot & Strauss, 1967; Strauss & Corbin, 1998)? If sizes as you go. able to select subsets of your qualitative data in the hardware they require to run efficiently. the latter, it is especially important that you have based on quantitative scores or demographics? You obviously cannot use a program if it is too ex- easy on-screen coding (rather than being re- Interest in context of data. When the program Or do you need to use your qualitative coding pensive for you, if it requires a machine you can- quired to code on hard copy, or having to deal pulls out chunks of text in response to your to generate scaled variables for statistical analy- not afford, or if it runs on the wrong plat- with cumbersome on-screen coding procedures) search requests, how much surrounding infor- sis in SPSS? Or do you want to be able to gener- form—say, PC instead of Mac. Happily, the U.S. and features supporting easy or automated revi- mation do you want to have? Do you need only ate word or code frequency tables for statistical Centers for Disease Control now distributes two sion of codes. Hypertext linking capabilities are the word, phrase, or line itself? Do you want the analysis? programs free via the Web: EZ-Text for qualita- helpful here too. “Automated” coding (in which preceding and following lines/sentences/para- tive surveys and AnSWR for unstructured text. the program applies a code according to a rule graphs? Do you want to see the entire file? Do Collaboration. If you will be working with a Also happily, reports from the field are that the you set up, such as when a certain phrase or a you need to be able to jump right to that place in team and more than one of you will be working Macintosh computers being sold today (the G3 is combination of other codes exists) can be help- the file and do some work on it (e.g., code, edit, on data analysis, look to see how the program today’s top Mac) run even the most powerful ful in either case. annotate)? Do you want the information to be supports collaboration. Some are fine if you just new PC programs satisfactorily with PC “emula- marked with a “source tag” that tells you where want to divide up the work with each of you tion” software. This will, presumably, continue Multiple versus single coding. Some programs let it came from (e.g., Interview 3 with Janice coding different parts of the data and then com- to be the case with future generations of Macs. you assign several different codes to the same Chang, page 22, line 6)? Or do you just want the bining the work. Others will support compar- The remaining issue, closeness to the data, is segment of text, including higher-order codes, source information without the text itself? Pro- ing multiple researchers’ interpretations of the more complex, and I also address it below in the and may let you overlap or nest coded “chunks” grams vary widely on this. same data. Some programs will allow multiple section headed “Debates in the Field.” For choice (the ranges of text you apply codes to). Others users to access a shared database over a net- purposes, remember to think about what kind of are stern: one chunk, one code. Still other pro- Intentions for displays. Analysis goes much work; others will allow you to merge periodi- closeness to the data is important to you. Many grams will let you apply more than one code to a better when you can see organized, compressed cally separate copies of the database that differ- researchers fear that working with qualitative chunk, but will not “know” that there are multi- information in one place rather than in page af- ent researchers have been working on. data on a computer will have the effect of “dis- ple codes on the chunk—they’ll treat it like two ter page of unreduced text. Some programs pro- Programs differ in how much control they give tancing” them from their data. This can in fact be chunks, one for each code. duce output in list form (lists of text segments, you over the merge process. Some allow you to the case. You may wind up looking at only small hits, codes, and so on). Some can help you pro- specify what the program will do if it finds, say, chunks of text at a time, or maybe even just Iterative versus one pass. Do you want—and do duce matrix displays. They may list text seg- codes or memos with the same name in each of line-number references to where the text is. This you have the time—to keep walking through ments or codes for each cell of a matrix, al- the copies being merged, whereas others follow is a far cry from the feeling of deep immersion in your data several times, taking different and re- though you will have to actually arrange them in a fixed rule. Some programs are good at letting the data that comes from reading and flipping vised cuts? Or will you limit yourself to one a matrix for display. Look for programs that let you tell which copy a code, memo, or other ob- through piles of paper. pass? An iterative intent should point you to- you edit, reduce, or summarize hits before you ject came from; others lose all identifying infor- But other programs minimize this effect. They ward programs that give you a good display of put them into a text-filled matrix with your mation so you have to use tricks like using dif- typically keep your data files onscreen in front of your previous coding, are flexible, invite re- word processor. Some programs can give you ferent names in each copy (e.g., I might start the you at all times; show you search results by scroll- peated runs, make coding revision easy, have quantitative data (generally frequencies) in a names of all codes I create with my initials, and ing to the hit, so that you see it in its full context; good search and autocoding features, allow you matrix. Others can give you networks or hierar- you start yours with your initials). Some pro- and allow you to execute most or all actions from to track connections between different parts of chical diagrams, the other major form of data grams offer specific features for letting you the same data-viewing screen. Programs that al- the text with hypertext, and can make a log of display. compare the coding of two different research- low you to build in hypertext links between dif- your work as you go. (See also the question of ers, for example, by showing you a table in ferent points in your data, provide good facilities whether your records are fixed or revisable dur- Qualitative only or numbers included. If your which you can see the coding done by each. for keeping track of where you are in the data- ing analysis.) data, and/or your analyses, include the possibil- base, display your coding and memoing, and al- ity of number crunching, look to see whether the Cut-Across Issues low you to pull together related data quickly can Fineness of analysis. Will your analysis focus on program will count things and/or whether it can in some ways help you get even closer to the data specific words? Or lines of text? Or sentences? share information with quantitative analysis The two main cut-across issues are closeness than you can with paper transcripts. If you Paragraphs? Pages? Whole files? Look to see programs such as SPSS or SAS. Think carefully to the data and financial resources. Let’s dis- choose with this consideration in mind, software what the program permits (or requires, or for- about what kind of quantitative analysis you’ll pense quickly with the latter question first. can help, rather than hinder, your work at staying bids) you to do. How flexible is it? Can you look be doing, and make sure the program you are Software varies dramatically in price. The range close to the data. It can, in fact, help keep you at varying sizes of chunks in your data? Can you thinking about can arrange the data appropri- of prices for the programs we reviewed in from drowning in those piles of paper. define free-form segments with ease? Some pro- ately. Consider, too, whether the program can Weitzman and Miles (1995b) was $0 to $1,644 However, having software that enhances the grams make you choose the size of your codable link qualitative and quantitative data in a mean- per user. That is still the range. (Look for dis- sense of closeness to the data may not be a crucial

816 ♦ METHODS OF COLLECTING AND ANALYZING EMPIRICAL MATERIALS Software and Qualitative Research ♦ 817

issue for everyone. Some researchers do not Does Software Drive Methodology? more flexible networks. HyperRESEARCH about your data. The different programs all have mind relying heavily on printed transcripts to emphasizes the relationship between codes and different strengths and weaknesses. It is also true get a feeling of closeness, whereas others think Another concern has been that researchers cases, rather than codes and chunks of text. that a clever user will be able to bend each of such heavy reliance defeats the purpose of QDA might wind up adapting their research to the When you code a chunk of text, you create an these flexible packages to a wide variety of differ- software. Furthermore, some projects simply do software they use, rather than the other way entry on what looks like an index card for a par- ent tasks, overcoming many of the differences be- not require intense closeness to the data. You around (Coffey et al., 1996; Kelle, 1997; ticular case. This strongly supports and encour- tween them. may be doing more abstract work, and in fact Lonkila, 1995)—that is, that the software will ages thinking that stresses casewise and Each of these assumptions is both a benefit for may want to move away from the raw data. impose a methodological or conceptual ap- cross-case phenomena, but makes it harder to some modes of analysis and a constraint. The key, proach. In fact, software developers bring as- look for and think about relationships among then, is not to get trapped by the assumptions of sumptions, conceptual frameworks, and some- codes within a text (although version 2, under the program. If you are aware of what they are, times even methodological and theoretical development at the time of this writing, appears you can be clever and work around them. The ♦ Debates in the Field ideologies to the development of their products. to be solving this problem). Finally, Code- program should serve your analytic needs, goals, These have important implications for the im- A-Text offers a quite different set of coding met- and assumptions, not the other way around. Re- pact that using a particular program will have on searchers interested in empirical work on the im- A number of debates have taken place over the aphors: (a) codes arranged into “scales” (you your analyses. However, as I have argued else- pacts of different programs on research are again past two decades in the qualitative research com- can assign only one code from each scale to a where, you need not, and in fact should not, be encouraged to refer to Fielding and Lee (1998). munity about whether the use of software is a segment of text, useful if you want to code your trapped by these assumptions and frameworks; good idea and, if so, what kinds of software are a text chunks by making mutually exclusive judg- there are often ways of bending a program to good idea. I will address four of the debated is- ments on a variety of factors), (b) codes auto- Should New Researchers Start Off your own purposes (see Weitzman, in press). For sues here: closeness to the data, whether soft- matically assigned according to words in the Doing Analysis by Hand? example, a program may allow you to define ware drives methodology, whether new re- text, and (c) open-ended “interpretations” you only hierarchical relations among codes. You searchers should start off doing analysis by hand, write about each text chunk. Working with this There is no clear-cut answer to this question. might work around this by creating redundant and whether software really affects rigor, consis- collection of coding metaphors could be ex- Certainly, it is important that new researchers be- codes in different parts of the hierarchy, or by pected to lead you to consider your text in gin by learning about how to do good analysis, 5 keeping track of the extra relationships you tency, and thoroughness. somewhat different ways than with one of the rather than just how to use a program. Whether want to define with memos and network dia- other metaphors described above. that means doing a first project by hand or learn- grams. Similar issues exist in the choice of other ing about analysis and software to do it with in Closeness to the Data The fact that developers bring conceptual as- types of software, such as textbase managers. the same course is a question best left to teachers. The issue of closeness to the data, which I sumptions to their work is in fact one of the InfoTree32 XT allows you to arrange texts in a I have taught both ways, and in my experience have just discussed in terms of program choice, strengths of the field. Many of the developers, hierarchical tree and lets you drag texts around students benefit from having some experience has been one of the big concerns raised by quali- particularly of code-and-retrieve and code- to rearrange them. Folio Views also allows you with manual methods, if only a few coding exer- tative researchers over the years. Experienced based theory builder programs, are researchers to create a hierarchical outline, but gives you cises, so that they can get the feel of what is hap- researchers have often found that as difficult as themselves. They have invested enormous intel- the additional capability of creating multiple, it was, the process of spending endless hours sit- lectual energy in finding the right tools for anal- pening analytically before they start worrying nonsequential “groupings” of texts that you can about using the software. ting on the floor surrounded by piles and piles of yses of different types, and the user can benefit activate any time you wish. Folio Views and paper led them, by necessity, to a very rich and greatly from their investment. askSam let you insert fields whenever and wher- thorough familiarity with their data. But as I It is also true that the design of the software ever you want in any given record, whereas Does Software Really Affect Rigor? have tried to argue above, software need not cut can have an impart on analysis. For example, InfoTree32 XT requires that you use a standard Consistency? Thoroughness? down on this familiarity. Software neither makes different programs work with different “meta- field set (of your own design) throughout the it better nor worse, it simply changes it. Al- phors”—that is, different ways of presenting the whole database. Idealist not only lets you cus- Some researchers are dedicated to the notion though some programs still create the sense that relationships among codes, and between codes tomize the field set for each record, you can also that software makes for more rigorous research. you are staring at just a small window of text and text. Kwalitan, NUD●IST, the Ethnograph, create a variety of record types, each with its There are even rumors floating around of federal with no sense of what lies around it, there are and winMAX all allow hierarchical relations own set of fields, and mix them in the same da- funding agencies requiring the use of software in now many programs available that provide rich among codes. For studies in which you are orga- tabase. Finally, whereas most of these programs grant proposals. Yet, as I have argued above, soft- contextual information (such as source infor- nizing your conceptual categories hierarchically, show you just one record at a time, Folio Views ware will not pull good work out of a poor re- mation, graphical maps of hypertext links, navi- these programs offer significant strengths. If you searcher. On the other hand, for all the reasons gable outlines, and linked lists of codes, docu- want to represent nonhierarchical relationships, shows you a word processor-like view in which outlined above in the subsection headed “Real ments, and text segments), and may in fact even if you choose to try to work around this records appear one right after the other as para- Hopes,” software can in fact help competent re- help you get to know your data better than ever metaphor, it may be less comfortable than using graphs. Clearly, each of these programs shows searchers do more rigorous, consistent, and thor- a program like ATLAS/ti that explicitly supports you a quite different view of your data, and so before. each may encourage different ways of thinking ough analysis than they otherwise might. The is-

818 ♦ METHODS OF COLLECTING AND ANALYZING EMPIRICAL MATERIALS Software and Qualitative Research ♦ 819

sue should be conceptualized not as whether the ware are nice, but we also need to continue to a new markup language that may succeed 2. The first release of NVivo lets you draw di- software makes the work more rigorous, but subject our hypotheses to empirical research. HTML, the World Wide Web formatting stan- agrams, but any connections you draw are repre- whether the researcher uses the software to do dard. A common standard like this, if adopted sented only in the diagram, they are not represen- more rigorous work than he or she could with- by other developers, would allow researchers to tations of the defined relationships among codes Needs for Software Development out it. move fully developed projects easily from one and other objects, as in ATLAS/ti. You see the ac- tual relationships among codes in a hierarchical program to another, just as we can now move “explorer” with expandable and collapsible We can at this point identify some of the tabular data among multiple spreadsheet and branches, as in NUD.IST, the Ethnograph, and ♦ The Future needs of researchers that are not yet met. For ex- database programs. winMAX. ample, the field is still lagging in its support for 3. This section does not contain much in the case-oriented work. A few programs have fea- way of references to specific software, both be- tures built in for explicitly tracking individual ♦ Conclusion cause the landscape changes every few years and It is my hope that the future will see a continua- cases through multiple documents, but few pro- because a single chapter does not allow for re- tion of current trends, both in scholarship and in grams are set up with a strong case-oriented sponsible comparisons among programs. software development. Some of my specific structure. 4. For any warning like this, check at the time hopes are outlined below. Display building, especially of matrices, still Unlike the situation just a decade or so ago, you are choosing to see if the program under con- needs much development. A product newly re- qualitative researchers now have available to sideration presents this problem. This type of problem is worked at so regularly by developers leased at the time of this writing, NVivo (from them an array of very good software tools to as- Needs for Scholarship on the Topic that it would be unfair and unhelpful for me to the developers of NUD●IST), allows you to sist in their research, and the use of soft- name particular programs. Things change. build an interactive matrix in which you can ware—including, but not limited to, word pro- Ongoing review work. In addition to books like 5. The reader interested in pursuing these click on cells to call up the corresponding text. cessors—seems more and more to be a regular Weitzman and Miles (1995b) and its upcoming questions further is referred to Fielding and Lee Matt Miles’s dream of a program that would part of the qualitative research process. There is revision (which I am coauthoring with Nigel (1998) for reports of users’ experiences of these combine this sort of functionality with the abil- still no one “best” program, not even for a par- and other issues when using different programs. Fielding and Ray Lee), which offer comprehen- ity to actually compose the summary text for the ticular methodology, and that’s good. It means 6. You can, in fact, attach not only date val- sive comparative reviews of the range of soft- output matrix (rather than switching to a word that researchers have to think through their ues, but text or numerical values as well. ware available at a particular time, there is a processor) is still one step away. methods and choose programs that fit, which need for regularly appearing reviews of new and Tools for narrative and discourse analysis are should keep them from becoming reliant on the revised programs as they appear. The journal still lagging as well. Researchers using these ap- software to lead them. As researchers continue ■ References Field Methods (formerly Cultural Anthropology proaches continue to call for features that let to hunt around for programs that will do the Methods) offers regular software reviews (of them flexibly describe the structure of text and things they want, and do them better, software Bogdan, R. C., & Biklen, S. K. (1982). Qualita- quantitative as well as qualitative programs) in discourse, and longitudinal researchers do not developers will likely continue to respond by tive research for education: An introduction to the same way that many journals feature regular yet have much in the way of tools built explicitly making their programs more and more useful. theory and methods. Boston: Allyn & Bacon. book reviews. More journals that serve qualita- for tracking cases over time, though NVivo has What else can we hope will come out of this Coffey, A., & Atkinson, P. (1996). Making sense tive research audiences should follow this lead. an “attributes” feature that allows you to attach collaboration between users and developers in of qualitative data: Complementary research date values to codes or documents.6 In each of the near future? More and better tools for shar- strategies. Thousand Oaks, CA: Sage. Debate on methodological questions. The kind these cases, researchers can either adapt coding ing analyses and raw data, perhaps by allowing Coffey, A., Holbrook, B., & Atkinson, P. (1996). of controversial issues addressed in this chapter systems to their needs or look for yet other kinds posting of project databases, with analytic Qualitative data analysis: Technologies and representations. Sociological Research On- need to be subjected to continued debate in the of software (such as hypertext authoring pro- markups, links, and memos, to the World Wide line, 1(1). Available Internet: http://www. literature and among researchers. We need to be grams, project schedulers, and so on) that they Web, as ATLAS/ti allows, or on CDs; tools for socresonline.org.uk/socresonline/l/l/4.html both wary of unintended influences of software can adapt to their needs. building complex reports that include analyses Drass, K. A. (1980). The analysis of qualitative and actively participating in shaping the future Finally, because no one program will ever do and data right in the report itself; and more and data: A computer program. Urban Life, 9, development of software by arguing (construc- it all best, researchers need developers to create better tools for supporting collaboration 322-353. tively) with developers about what we need and the possibility of importing and exporting among research teams, and for involving infor- Fielding, N. G., & Lee, R. M. (Eds.). (1991). what we do not like. marked-up, coded, annotated data from one mants in the research process without intensive Using computers in qualitative research. Lon- program to another. At this writing, there is just computer training. don: Sage. More empirical work. The kind of empirical a little of this beginning to happen. The develop- Fielding, N. G., & Lee, R. M. (1998). Computer work on the impact of software on analysis that ers of Code-A-Text, the Ethnograph, and win- analysis and qualitative research. London: has been pioneered by Fielding and Lee (1998), MAX have agreed to work on a common struc- ■ Notes Sage. Weaver and Atkinson (1995), Horney and ture, partially realized at this point. And Glaser, B. G. (1965). The constant comparative method of qualitative analysis. Social Prob- Healey (1991), and Walker (1993) needs to be ATLAS/ti has become the first program to sup- 1. I discuss some exceptions in the subsec- lems, 12, 436-445. continued. Opinions about the impact of soft- port export of fully developed projects in XML, tion below headed “False Hopes and Fears.”

820 ♦ METHODS OF COLLECTING AND ANALYZING EMPIRICAL MATERIALS

Glaser, B. G., & Strauss, A. L. (1967). The dis- Miles, M. B., & Huberman, A. M. (1994). Qual- covery of grounded theory: Strategies for itative data analysis: An expanded source- qualitative research. Chicago: Aldine. book (2nd ed.). Thousand Oaks, CA: Sage. Goetz, J. P., & LeCompte, M. D. (1984). Eth- Ragin, C. C. (1987). The comparative method: nography and qualitative design in educa- Moving beyond qualitative and quantitative tional research. New York: Academic Press. strategies. Berkeley: University of California Horney, M. A., & Healey, D. (1991, April). Hy- Press. pertext and database tools for qualitative re- Seidel, J. V., & Clark, J. A. (1984). The search. Paper presented at the annual meeting Ethnograph: A computer program for the of the American Educational Research Asso- analysis of qualitative data. Qualitative Soci- ciation, Chicago. ology, 7, 110-125. Kelle, U. (Ed.). (1995). Computer-aided qualita- Shelly, A., & Sibert, E. (1985). The QUALOG us- tive data analysis: Theory, methods and prac- ers’.manual. Syracuse, NY: Syracuse Univer- tice. London: Sage. sity, School of Computer and Information Kelle, U. (1997). Theory building in qualitative Science. research and computer programs for the Strauss, A. L., & Corbin, J. (1998). Basics of management of textual data. Sociological Re- qualitative research: Techniques and proce- search Online, 2(2). Available Internet: dures for developing grounded theory (2nd http://www.socresonline.org.uk/socresonlin ed.). Thousand Oaks, CA: Sage. e/2/2/l.html Tesch, R. (1990). Qualitative research: Analysis Lee, R. M., & Fielding, N. G. (1991). Com- types and software tools. New York: Falmer. puting for qualitative research: Options, Tesch, R. (1991). Computers and qualitative problems and potential. In N. G. Fielding & data II. Qualitative Sociology, 14(3). R. M. Lee (Eds.), Using computers in qualita- Walker, B. L. (1993). Computer analysis of qual- tive research (pp. 1-13). London: Sage. itative data: A comparison of three packages. Lofland, J., & Lofland, L. H. (1984). Analyzing Qualitative Health Research, 3(1), 91-111. social settings: A guide to qualitative observa- Weaver, A., & Atkinson, P. (1995). From coding tion and analysis (2nd ed.). Belmont, CA: to hypertext: Strategies for microcomputing Wadsworth. and qualitative data analysis. In R. G. Burgess Lonkila, M. (1995). Grounded theory as an (Ed.), Studies in qualitative methodology. emerging paradigm for computer-assisted Greenwich, CT: JAI. qualitative data analysis. In U. Kelle (Ed.), Weitzman, E. A. (1999). Analyzing qualitative Computer-aided qualitative data analysis: data with computer software. Health Services Theory, methods and practice (pp. 41-51). Research, 34(5), 1241-1263. London: Sage. Weitzman, E. A., & Miles, M. B. (1995a). Mangabeira, W. (Ed.). (1996). Qualitative soci- Choosing software for qualitative data analy- ology and computer programs: Advent and sis: An overview. Cultural Anthropology diffusion of CAQDAS [Special issue]. Cur- Methods, 7(1), 1-5. rent Sociology, 44(1). Weitzman, E. A., & Miles, M. B. (1995b). Com- Miles, M. B., & Huberman, A. M. (1984). Qual- puter programs for qualitative data analysis: itative data analysis: A sourcebook of new A software sourcebook. Thousand Oaks, CA: methods. Beverly Hills, CA: Sage. Sage.