Supporting Multiple Perspectives: Case Studies of Using Hypertext in Qualitative Research Mark A. Horney University of Oregon This paper presents three case studies of educational researchers using a hypertext editing program as part of their data analysis process. The studies come from a larger project examining hypertext authoring in a variety of contexts (Horney, 1991). I begin with a short introduction to the concepts of hypertext and follow the case studies with a short analysis of lessons to be drawn from the experiences of these researchers. Hypertext Hypertext is a form of electronic writing and although its origins go back to at least 45 years (Bush, 1945), hypertext software has been generally accessible only in the past few years. There are differing views as to what exactly constitutes hypertext, but it is generally described as either "non-sequential writing" (Nelson, 1987, p. 1/17), or as a web of informational nodes that are linked together and which readers access via a navigation system in a sequence of their own choosing (Dede, 1988, p. 96). The utility of hypertext software for qualitative researchers can be seen by considering its application to traditional þsegment and sortþ techniques of data analysis wherein researchers first þ... divide observed phenomena into units, and [then] indicate how units are like and unlike each otherþ (Goetz & LeCompte, 1984, p. 170). Operationally this process often amounts to marking up paper transcripts with margin notes or colored markers, cutting the transcripts apart, and piling or pasting the resulting text fragments into appropriate categories. A researcher using a hypertext editing system for this task follows the same general steps, but commits results to paper only, if ever, in the final stages. Segmenting is accomplished by separating transcript fragments and placing each in a separate hypertext node. Fragments can be defined by some arbitrary rule (e.g. starting a new node after each change in speaker or for each new paragraph), or for semantic reasons as judged by the researcher. Once nodes have been established, comparisons and contrasts are drawn by erecting labeled links between the nodes. Categorization is accomplished by creating new nodes, one for each category, and then linking each data node to appropriate categorical nodes. Categories can be examined by using the navigational system to traverse links electronically, or by printing their nodes to paper. Just as the term hypertext is somewhat ill-defined, software packages with hypertext features vary a good deal in their capabilities. A list, with brief descriptions is given by Fielding and Lee (1991, p. 195-199). The participants in this study used EntryWay, a hypertext editing system which I developed. EntryWay runs on the Macintosh family of computers and is based on HyperCard (Apple Computer, 1987). Like all hypertext systems EntryWay provides for nodes, links, and a system of navigation. Sample nodes are shown in *Figures 1 and 2. *Figure 1. EntryWay Data Node*Figure 2. EntryWay Thread EntryWay nodes can be configured with a variety of different spaces, called fields, for researchers to enter text. *Figure 1 has been set up as a data node with a central field to hold a data fragment, and three surrounding fields for commentary. The four corner arrows indicate links to other nodes and the menu bar across the top provides commands for navigation and node/link manipulations. Each node is named, in this case S110, which was chosen to indicate this as the 110th node of an interview with an informant called Shannon (this and all other participant names are pseudonyms). Text of this sort can be automatically imported into EntryWay documents from word processing files. *Figure 2 shows a special type of EntryWay node called a þthread.þ Threads are used to form categories. The list to the left shows the members of the thread þCatIntersect,þ each of which is another node of the document. The text held on the member nodes of a thread can be exported from the document to word processing files where it can be viewed, printed, or transferred for manipulation by yet other programs. Other than maintaining this special list, threads operate like any other node and can themselves become thread members, and the þanchorsþ of links. This versatility is useful in representing relationships among coding categories and their attendant constructs. Finally, EntryWay also has the capability to draw þmapsþ showing thread and link relationships among nodes. This feature will be explained in the third case study. This brief overview of the principles of hypertext and the capabilities of EntryWay can not do justice to either, but it should suffice to make the following case studies comprehensible. The Case Studies These case studies were part of a larger research project which aimed to gain a broad experience with hypertext and from that experience to tease out factors affecting hypertext authors as they work. To accomplish this goal, I created EntryWay, analyzed the characteristics of hypertext to identify particular tasks for which it might be useful, sought out individuals willing and needing to accomplish these tasks, taught them to use EntryWay, assisted them in their work, and closely observed their activities. The accounts presented here were based on data collected from taped interviews, direct observations, the hypertext documents created by each researcher, and from computer monitoring transcripts kept automatically by EntryWay showing activities. Dr. Peter Stone Dr. Stone was an assistant professor in a Department of Physical Education, where his research focused on teacher socialization. At the time we met, Stone and a colleague were studying a teacher educator, a Dr. Rogers. They had collected a large body of data about Rogersþ teaching methods, and were in the final stages of data analysis. Part of their data consisted of interviews with five of Rogersþ students. Stone was interested to see if the perceptions of these students supported the interpretations he had drawn from other data. After seeing a demonstration, Dr. Stone agreed to use EntryWay in analyzing these transcripts. I originally planned an observational study of Dr. Stoneþs work with EntryWay. I expected to introduce him to hypertext, help plan an interview analysis document, observe him on some occasions while he worked, and then reconstruct his experience from the program monitors, the observations and a final interview. In practice however, I became a participant observer and collaborated with Stone at each stage of the analysis. In part, this collaboration was necessary because of problems with hardware. Stone owned a Macintosh Plus computer, which he used for word processing, but which did not have a hard disk. This deficiency threatened to limit the size and functionality of Stoneþs document and so all of our work was done on a better computer which I supplied. Since I could not leave the computer with him, we had to meet and work together. This situation had the advantage that I was able to watch Stoneþs entire process of data analysis, but ongoing scheduling problems extended the project over a period of several months. Our work with the interviews proceeded in five steps: (a) preparation of the interview transcripts; (b) primary coding of the transcript elements; (c) secondary coding; (d) examining thread intersections; and (e) exporting threads. I did all of the preparatory work and will describe that process in detail since similar procedures were followed in the other cases. I began with a disk containing text files of the interviews which had been transcribed from audio tape. I first divided the transcripts into paragraph sized chunks (referred to as transcript þelementsþ), each marked by a change in speaker, with generally one node containing a question and the following node the answer. These chunking decisions were made more or less arbitrarily, but often with a eye towards making chunks smaller rather than larger. I next imported the prepared files into an EntryWay document, which we called þROGERS.þ Once all of the data nodes were in place, I created several þorganizationalþ threads to facilitate navigation and searches. For instance the thread ELEMENTS contained all 650 data nodes. ANSWERS and INTERVIEWERS contained just nodes with student responses or just researcher questions. The S, M, T, H, and P threads each contained nodes from one of the interviews. Once these threads were in place, I made printouts for Stone showing the interview text segmented and labeled as it appeared in ROGERS. As a final step, I made ROGERSMAP, a map intended to show Stone the relationships among the organizational threads and to serve as a navigation tool. However, we never used the map after our first work session. The entire process of preparing ROGERS took approximately six hours. The next stage in working with ROGERS was the primary coding of the transcripts, which was accomplished in two work sessions. This involved placing data nodes onto 18 threads, Q1 through Q18, one for each of the questions in the interview protocol. Stone later added four additional questions. This process took approximately five and a half hours and served to re-acquaint Stone with the data, re-organize the data from its chronological ordering into a content ordering, and to eliminate material which was irrelevant. Note however, that all the data always remained available within ROGERS, and that nodes could always be viewed from either ordering. As Stone worked, I gave him instructions on the use of EntryWay. These directions had to be repeated in each session since Stone forgot various commands in the long interims between sessions. Only towards the end, when we had three sessions in two weeks, was Stone able to proceed independently. I also made suggestions about document organization and about short cuts. Stone worked steadily, moving from element to element, occasionally using the TRAIL, a list showing recently visited nodes, to access one of the questions. Some elements were handled quickly, such as statements by the interviewers, which Stone only glanced at and didnþt categorize. Others were visited for the 10-20 seconds it took to read and place them. Once Stone completed the primary coding process, he began secondary coding by re-coding the data using categories derived from other data about Dr. Rogers: So far these are the three main categories that weþre working with: experiences as a mother, her multiple roles as faculty member, and then her relationships with the students ... Thereþs also a secondary area dealing with knowledge and issues central to teaching that she thinks are particularly important. In the two work sessions comprising this secondary round of coding Stone created seven new threads (VIDEOTAPING, MOTOR LEARNING, MOTHER, ROLES, STUDENTS, BACKGROUND, STYLES) and revisited his data placing nodes into these new categories. There were two changes in his work pattern. First, rather than visiting all nodes, he only coded those which had already been placed on one of the 22 question threads, and second, he used EntryWayþs keyword search function to locate some data. For instance, for the thread VIDEOTAPING we searched for þvideo,þ þtaping,þ and þtape.þ EntryWay automatically placed any entry containing those words on the VIDEOTAPING thread. These searches did not eliminate the need to read each node, but did provide Stone with some confidence that information relating to his categories did exist within the data set. During this time Stone added five additional þstudentþ categories (Practice, Lectures, Substitute, Confident, Planning) which he thought complemented the original seven þteacherþ categories: þThis is one of the things I thought was going to happen if I went through the questions. Iþd begin to see things that I missed, and I don't have a category for.þ When finished with this secondary coding, Stone felt his work with ROGERS was essentially complete. His only further requirement was to obtain printouts of the student and teacher threads. In these he was going to search for, on paper, dominat themes and find þstudent quotes to fit into some of the category definitions we already have.þ I suggested that such dominate themes might be found in the intersection of his teacher and student categories. I proposed to show this intersection by making a map of the relationships between these categories. I was particularly eager to do this because at this point Iþd been considering the usefulness of maps for some time, but hadnþt made or found one of any real value. I was also interested to see if I could push Stone into a deeper analysis of the relationships among categories, the sort of secondary analysis that might better justify the use of hypertext. Stone agreed the intersection might be useful. Creating an intersection thread (*shown in Figure 2) of the teacher and students categories was relatively simple. Making the map proved to be more difficult and for technical reasons this was abandoned. Therefore, I simply made a printout of the intersection thread, which contained 35 data nodes, and sent it along to Stone with printouts of each of the other categories. In our final meeting Stone discussed the process of creating and using hypertext in qualitative research. Stone allowed that at the beginning he wasnþt sure of exactly what I wanted and let us work on transcripts he thought would be easy to analyze. None the less, Stone found the process þenormously useful ... it certainly saved time.þ What I found most useful were these parts where the sections were broken down. Originally we started off by looking at questions, Iþd say OK letþs put question one together and question two together and so forth. That was ok, that was useful and that's about where I usually wind up with this kind of analysis anyway. But when we were able to breakdown the categories and actually fit the quotes in with the category, at least a preliminary category. It just makes it that much easier to analyze, to immediately see in front of me and realize we do have a category, that thereþs enough there, that itþs a legitimate category. At least close enough to being a final category that I could begin drafting the document. It also makes it easy to have the quotes right there. What I [normally] do is take the transcripts and either cut them or highlight them in some way, and group them. I think about them for a while, and come back and see if I can regroup them. Then, maybe write a draft of just this one section, see how that comes out, and then go back and see if I need to regroup them again, does everything fit, is there anything that doesnþt make sense? The number of steps I take are a lot longer than [using EntryWay] because what we did on the computer in a matter of not even seconds, what would be a physical movement, and so would take much longer. I try not to cut up the transcripts because once you do that, its kind of like your thread elements, once you cut up the transcript you more or less loose the thread elements. I suggested to Stone that even more time could have been saved if we skipped the first round of coding (organizing the transcripts according to the interview protocol), and had gone straight to categorizing by the teacher and student themes used in the secondary coding. Stone objected: I don't know if I could have, because when you gave me the printouts of the questions, I could see it was very labored. The questions didnþt give us the answers we were after. I saw that, and in reading the answers I began to see what the questions should have been. That's what the categories became. Stone noted a change in the way he treated quotes: ... I donþt know if it was the program that allowed me to do this, or a change Iþm making in my own way of analyzing data. I find sometimes I get a good quote and Iþll hang too much on it, I'll put more value in it because itþs worded well or expresses an idea that I find attractive, and it blows the weight of that idea out of proportion to what it should be. So this time, this program kind of allowed me to focus on a bigger picture. I think the speed had something to do with it. For example, in question one, we sat down that one morning and had almost all the questions sorted in two or three hours. Ordinarily I'd take two or three hours for one question and try to sort one question out, get that set, and then move on to the next one. I wasnþt focused on the specifics immediately. I was focused on trying to analyze the larger data set rather than just the specific question. Stone believed the use of EntryWay saved enough time that he was able to do a secondary analysis of materials he would normally only go through once, and that this in turn allowed him to avoid becoming too attached to the easily found items that might dominate his thinking inappropriately. Stoneþs task was one of data reduction. The student interviews were not the main focus of his study of Dr. Rogers, they existed only as support. What we did with EntryWay was to sift out a few pertinent statements from the hundreds of transcript elements. Noel Salo Noel Salo was a teacher on sabbatical leave doing a graduate degree in educational computing. He had completed the data collection phase of his dissertation when he asked for a demonstration of EntryWay. Some weeks later, somewhat to my surprise, Salo called with some specific questions and indicated he was already using the program. He called several more times with questions or problems, all of which I was able to diagnose over the phone. After he finished with EntryWay, I met with him to get copies of his documents and to interview him about his experiences. Of all the participants, Salo had the least support, either in designing his documents, or in operating the program. Saloþs task was simple and straightforward. His dissertation data consisted of transcripts taken from interviews with ten educators concerning goals for mathematics education, and totaled some 250 pages of typescript. Salo used EntryWay to sort transcript elements into 19 previously defined categories using procedures similar to those used by Stone. This took approximately 20 hours spread over 10 days. Once the coded elements had been exported, Salo used a word processing program to read, print, search and manipulate the categories. Because he was familiar with the material, he found it necessary to see only a quarter of the categories in print and read the rest from the computer screen. To find particular segments in the word processing file, Salo used the FIND function built into the word processor while moving between several files, each viewed through a separate window on the computer screen. Once items were located, they were cut and pasted into the dissertation. Salo used this technique to create a simulated panel discussion among his informants. When asked whether this task could have been accomplished more easily in EntryWay, Salo replied that EntryWay was too slow and that he hadnþt learned enough about the program to do the required operations, such as keyword searches. Before EntryWay came along, Salo planned to print out his transcripts, cut them up and thumb tack them to the walls of his apartment. He was happy to have avoided this. To sum up his experience Salo said þI found it very valuable in organizing the paragraphs from my transcripts into themes.þ While Salo found EntryWay to be an effective tool for accomplishing the simple straight forward segmentation and sorting of his interview transcripts, it did not seem to present him with any new or different insight into either his task, or his data. Dr. Fiona Shaw Dr. Shaw held a position in a School of Education where she was involved in a study of a mathematics classroom. She was replicating in the 7th grade her previous work from a 5th grade class on how functions and graphing are taught and how teachers use instructional representations. Shawþs data consisted of audio transcriptions from video tapes of seven math lessons. She planned a four step analysis process: (a) Code the transcript elements by þactivity structures.þ That is, the kind of activity taking place: assignment, development, shared presentation, enrichment, guided practice, review, seat work, management, closure, etc.; (b) Prepare a þcontent outlineþ and cross reference this with the activity structures; (c) Read the prepared transcripts for interesting features including: comparison of the presented content to the received content, tracing differences in the use of a particular activity structures across different lessons, and pulling out all of the ideas being presented using the teacherþs own words; and (d) Represent the information from the first three steps as a semantic network. It was in the last step that Shaw found the most interesting application of EntryWay. She preferred to work with her data in visual forms and so our direct purpose became to set up a document where Shaw could manipulate EntryWay map of the content and processes in the lessons she was investigating. As I had done for Stone, I prepared an EntryWay document, LESSONS, to contain Shawþs transcripts. And also as with Stone we worked together on most occasions, with myself at the keyboard, and Shaw working from printed EntryWay threads. In part this arrangement was driven by the need for efficiency since it was difficult to schedule time on the communal computer we found it necessary to employ. We used this computer because of its power, and also because Shaw planned to have some of the coding work done by graduate students and this was the only machine with open access. This part of the plan collapsed, however, when Shaw was unable to get enough students, or enough money to pay them, and we were pressed to move forward faster than the students could work. However, we continued to use this machine for some time. After we had coded one lesson (called Lesson A) for activity structures and content, we began our first experiments in map making. Our first effort, the Concepts A map (*Figure 3) was intended to show how the content of the lesson was distributed among the transcript elements. *Figure 3. ConceptsA Map EntryWay maps represent nodes by small fields containing node titles, and show link and thread relationships by lines drawn between the fields. Fields can be added to a map one at a time, or in groups with commands such as þAdd all members of thread A,þ or þAdd any node linked to any node already present.þ The fields are placed randomly on the screen and then moved into some meaningful arrangement by the author. The size, font, and style of the text can also be adjusted. Pointing at and clicking on a field allows readers to move to that node. Lines showing link and thread relationships are drawn automatically between associated fields, a thin line showing links, such as from A30 to A31, or with a thick line drawn from item to item of a thread list, as with the line from the þcommunicationþ thread to each of its members, A22b, A23, A26, A52, etc. Map making thus is an iterative process of adding node fields, and then moving them about the screen, unscrambling thread and link lines, as some meaningful pattern evolves. As we worked with ConceptsA it quickly became apparent that no such pattern was likely to emerge since the thread lines were always impossibly tangled. I thought the map useless, but Shaw claimed ConceptsA was þimmensely helpful:þ The reason this is helpful is that I find this kind of data analysis overwhelming. Itþs very easy to get lost in your data, and the connections are what I'm most interested in. I'm trying to overcome the chronological organization thatþs implicit in this type of a transcript. This is very helpful in beginning to rearrange and reorganize the data. I think the links will begin to tease out patterns and the threads will help me decide whether these are the appropriate categories, because again, these are categories that are going to emerge from each transcript. If I find that may not be the most powerful way to think about a particular lesson, then its time to take a look at that category. Because of the difficulties with ConceptsA I rewrote the EntryWay mapping software so that thread membership was shown by lines radiating from the thread field. Shaw dubbed these as þConesþ. By the time Iþd finished, Shaw had some new ideas about the data: I'm fairly certain I want to do an analysis of questions she asks, because the questions she asks are very rich, they serve many purposes, and theyþre very provocative. One analysis of the questions is [to ask] where do the questions happen, in which activity structure so to speak. I was thinking the activity structure map would have the different activity structures, what ever they are, and then I would have linked that to all the questions that happened in each activity structure. So, we created threads for different categories of questions (e.g. JustifyThinking, Management, OpenEndedKnowledge, OpenEndedProcess, OpinionInference, OpinionJustify, RightAnswerInference, RightAnswerKnowledge etc.), and began coding the elements of the A and B threads by these categories, as well as for activity structures and content. At this time Shaw began working on LESSONS alone at her home away from her office where she was constantly interrupted. She found that even though her home computer was much slower than the office machine, it was sufficient for map making where there are long intervals when the software operates alone. At home she was able to do other work while waiting on the computer, and found the difference between a 5 minute gap and a 20 minute gap unimportant. By the time of our next meeting, Shaw had created question threads for the second lesson, B, and had it partially coded. Together we finished coding B and began QuestBMap (*See Figure 4). This work was my last substantial contribution to LESSONS. A change took place in Shawþs thinking about maps at this point. Beyond serving as devices for her to think about the data, they could also be used to communicate: Another way to think of it, Qrankin is right answer knowledge questions which are a big part of everyone's teaching. So it might be worthwhile to have one picture establishing that point: Yes this teacher is like lots of other teachers, she asks lots of questions. *Figure 4. QuestBMap QuestBMap is organized with transcript elements containing teacher questions scattered along the diagonal and its sides of the map. Threads representing activity structures (e.g. ASreview, ASDevA, ASdevB, etc.) are located above the transcript elements with cones drawn down to their members. Threads for the different types of questions are shown below the elements along with their cones. Shaw drew a number interpretations from this map: I realized in looking at the map that Warm and Review weren't different [categories], that really theyþre the same thing. When a teacher is reviewing content, whether the purpose is to get kids ready for the day or to close up the day, that to me is the same thing in this situation. By seeing those two separate cones I realized, wait a minute I really don't think Warmup and Review are different, I think theyþre the same thing. Just seeing the cones evolve says something. This teacher tends to move from one activity structure to the next. Sheþs not jumping around back and forth between the activity structures, which to me indicates this teacher is real together for her management and her design of classroom instruction. The other nice thing I like about the maps is that I have this chronological trace of the questions. Whatþs intriguing now when you look to the other side of the map is that the types of questions she asks vary quite a bit from one activity structure to the next. Itþs not like she's only asking one kind of question. ... Let me make an exception to that. During the review time she is asking these right answer knowledge questions, what in Bloomþs taxonomy would be called lower level questions. Thatþs pretty much the kind of questions sheþs asking during this review. Her purpose here is to go over some concepts and some procedures the studentþs have already learned and so she wants the right answer and she wants to go through it. She sticks to that kind of question there pretty much all through her review. ... She ask these right answer knowledge questions through out the lesson and we find that the open ended questions are scattered throughout the lesson as well. What the map makes me want to do is kind of highlighting areas I want to go back into the transcript and look at again, or even into the video tapes. My hunch is that weþll see a real progression where these justify your thinking questions become more and more dominate as the lessons progress through the unit. The maps will help me make ferret out that particular hunch. Iþd like a map like that for all seven lessons, because I think that will be a real strong picture of how this teacher breaks up a class period. In summing up the usefulness of hypertext and her experience with EntryWay Shaw said: What this does is confirm my hunch, my impressions of this teachers classroom, of her teaching, which to me is a challenge in qualitative research. Weþre constantly exposing ourselves to the richness of a situation and then weþre getting these insights, these hunches. Often its very difficult to go back and reconstruct all the bits of evidence that lead to that particular hunch or insight. With this I'm feeling confident now. I feel I've got a nice base to support my hunches or my analyses of this particular teacher. I'd say the maps are the most helpful to me when I get to that stage. You know coding data, that's not too different from what I do with paper and pencil. [I] sit down and break it into elements and code things. But being able to pull together that coding into visual forms, thatþs real helpful for me. The second way I think its real helpful, as kind of a hidden benefit in a way, is that it really does force you to be systematic and to clean up your categories. I guess I was impressed with that as I worked with it because inconsistencies hit me very quickly. Theyþre not always that apparent when you have 40 pages of a transcript that youþve scribbled on in the margins. Where as in this, it was very clear to me that my Review and Warmup [categories] were the same thing. When this study ended Shaw had only completed maps for the first two of the seven lessons making up this math unit. She planned to make maps similar to QuestBMap for each lesson and used them as the basis for the analysis of her data. Shaw was the only author in this study to succeed in conceptualizing and executing useful hypertext maps. She was also the only author to organize entries with independent sets of threads. With these two accomplishments, LESSONS is the most advanced document produced in this study. Implications The experiences of these qualitative researchers elicit three observations about the utility of hypertext for data analysis. First, the use of hypertext burdens researchers with a new set of interrelated demands on their skills and resources: Researchers must have adequate access to appropriate hardware and software. Researchers must acquire both the computer skills and the hypertext skills necessary to exploit the advantages offered by hypertext, and to minimize its disadvantages. Researchers must be able to efficiently move data and data representations into and out of different software packages, and across the barriers separating different media. The multimedia capabilities of hypertext editing systems require researchers to acquire multimedia production and editing skills in addition to the traditional writing skills that are already and always needed. Because hypertext provides greater fluidity in data manipulation, representation, and interpretation, extra responsibilities are imposed on researchers to maintain document/data integrity. Second, this study suggests that hypertext editing systems can reduce the time and effort required to perform the basic segmenting and sorting procedures that underlie many data analysis procedures. This advantage is augmented by the fact that hypertext furnishes researchers with computer files, rather than the piles of paper scraps and note cards, that can be easily manipulated by other computer applications such as word processors, data bases, and presentation graphics programs. Third, researchers in this study reported positive changes in their analytic methods. They believed hypertext allowed them to be more focused, systematic, and organized. Further, the flexibility of hypertext allowed them to maintain a larger perspective than was ordinarily possible during data analysis. These observations suggest the possibility that hypertext may at some point in the future, significantly alter the methods and expectations of qualitative research. Research continues on the applications of hypertext in qualitative research. This work includes improvements in the editing software, the further development of analytic techniques appropriate for hypertext, the integration of hypertext based processes with other analytic procedures, creating training materials, and investigating the cognitive states involved during the use of hypertext. *Figures were not compatible with the electronic publishing methods used. Figures are available from author on request References Apple Computer, Inc. HyperCard [Computer Program]. Cupertino, CA: Apple Computer, Inc. Bush, V. (1945). þAs we may think.þ Atlantic Monthly, 176(1), p.101-108. Dede, C. (1988). The role of hypertext in transforming information into knowledge. In W.C. Ryan (Ed.), Proceedings of the National Educational Computing Conference '88 (pp. 95-102). Eugene, OR: International Council on Computers for Education. Goetz, J. P., & LeCompte, M. D. (1984). Ethnography and qualitative design in educational research. San Diego: Academic Press, Inc. Fielding N. G., & Lee R. M. (Eds.). (1991). Using computers in qualitative research. London: Sage. Horney, M. A. (1991). Case studies of authoring in hypertext. Unpublished doctoral dissertation, University of Oregon, Eugene, Oregon. Nelson, T. H. (1987). Literary machines. South Bend, IN: The Distributors.