Social Science Computer Review 10:4, Winter 1992.
Software for analyzing qualitative data is emerging simultaneously with a resurgence of formal methods for qualitative analysis, giving contemporary sociologists unprecedented control over qualitative research projects. This article focuses on computer methods of text analysis and on software for building "logic models" that systematize interpretations of empirical data. Content analysis, network analysis, and image analysis are mentioned as additional growth areas in computer-assisted qualitative research. Keywords: qualitative sociology, text analysis, modeling.
Formal analysis of qualitative data took such a long sleep in sociology that it appeared to have died. Robinson's (1951) lambasting of analytic induction marks the beginning of the somnolent period, and Ragin, Mayer, and Drass's (1984) extraction of logical expressions from data is a possible marker of the reawakening. During the long sleep hundreds of qualitative researchers content-analyzed texts, abstracted meaning from interviews, discovered social routines through field observations, interpreted transcript with respect to cultural processes, and so on, and some of those studies, brilliantly conducted, now are sociological classics. But almost no one was working on methodologies pinned to basic concerns in qualitative research such as interpreting everyday details in terms of abstract concepts, relating conceptions of social reality to material contingencies, and making observations to ascertain existence rather than quantity.
Methodological stirrings occurred during the long sleep. Qualitative researchers produced didactic writings too numerous to list, and some of these (e.g., Glaser & Strauss, 1967) focused especially on issues of analyzing qualitative data. Sociological methodologists (who at the time were making great progress in dealing with quantitative issues) also made a number of contributions that might be End p. 531 counted as attempts to recover lost consciousness of the qualitative realm. For example, there was continuing work on Guttman's scalogram analysis (see Shye, 1978) and on latent structure analysis (Henry & Lazersfeld, 1968); Guttman (1959) developed his facet mapping sentences; and Stone, Dunphy, Smith, and Ogilvie (1968) developed the General Inquirer for content analysis. Nevertheless, we had to wait until the 1980s before writings explicitly aimed at qualitative researchers became "methodological" in the sense of performing generalized analyses that would simplify the work of qualitative researchers or that would systematize qualitative data processing in such a way as to lead to deeper or more rigorous research outcomes.
The reawakening of systematic qualitative data analysis cannot be credited to microcomputers (e.g., Charles Ragin used mainframes to develop the prototype of the QCA program that is discussed below). Microcomputers now, however, are making unprecedented computing technology accessible to qualitative researchers and giving qualitative researchers control over their data in ways they never enjoyed before. While most qualitative researchers know that their desktop computers offer new opportunities for analyzing qualitative data, I believe that many are baffled concerning exactly what the opportunities are. In this essay I will discuss some ways of processing qualitative data that literally are at a researcher's fingertips and others that can be gained by acquiring specialized software packages. The methodological opportunities divide approximately into two groups: text processing and model building.
Qualitative sociology's core data consist of text in the form of notes, transcripts, or archives. The text provides particulars that are to be abstracted, tracks processes that can be defined so as to identify contingencies in social life, and offers evidence for marking the expanse of social realities.
Since text is so important in qualitative research, it is safe to say that every qualitative researcher owns a word processor and has mastered its use as a typewriter. It may not be realized, however, that a modern word processor provides a qualitative researcher with one of the most powerful text analysis tools that can be acquired. In this section, I will mention some frequently overlooked potentials of word processors to emphasize the point that word processors are underutilized for textual analysis. The capabilities mentioned are those I have implemented in the software package that I know well, WordPerfect, though more or less the same capabilities could be developed with other powerful word processing packages, such as Microsoft Word.
Faced with tens of thousands of words in documents, a qualitative researcher's first daunting task is to gain an overview of the End p 532 content of the documents and some initial ideas about what kind of information would be useful to pinpoint and code in subsequent analyses. Unquestionably, that means studying the materials carefully. However, some initial insights into the content can be acquired through a simple type-token analysis, which lists words appearing in the documents and their frequency. A type-token analysis is a content analysis that involves no interpretation but can serve as a preliminary for interpretive analyses.
Type-token analyses can be performed easily with a word processor. The procedure outlined here is implemented through a macro that changes a copy of a document into a type-token report on the document, and the macro can be used repeatedly with different documents. Steps 1 through 5 below can be constructed using the word processor's macro learning function. Steps 6 and 7 are incorporated by editing the macro and adding commands in the word processor's macro programming language.
To transform a document into a type-token report on the words in the document you do the following:
1. Search for punctuation in the text and replace it with nothing in order to get rid of it. (My macro makes 13 passes in order to remove commas, periods, parentheses, etc.)
2. Search for and replace every space with a <Hard-Return> in order to put each word on a separate line.
3. Change capitals to lower-case, since capital letters sort differently from lower-case letters.
4. Search for uninteresting function words--such as a, an, the, and, of, to, is--and replace them with nothing, in order to reduce the size of the text. It is important to search for lines containing words rather than for the words alone, so that other words containing the letters do not get mangled. For example, to get rid of "the" one searches for <Hard-Return>the<Hard-Return> and replaces instances of this with <Hard-Return>.
5. Sort the lines alphabetically so as to order the words (types) with all repetitions (tokens) following one another.
6. Delete the repetitions while tallying their number, and enter the token count after each word before going on to count and delete repetitions of the next word. This procedure is implemented through programming code added to the macro, and I include the following algorithm to make the programming task less intimidating.
Move to the top of the file
Create a blank line and move to line 2
Assign 0 to a variable called Counter
Assign z to a variable called Word-1
[Emplace a label called Label-for-looping here in the code]
Add 1 to the Counter variable
Block the word on the current line End p. 533
Assign the blocked word to a variable called Word-2
[Set up the following "if-else" statement]
If Word-2 = Word-l
Delete the current line and move to the next line
Else
Go up one line, to the end of the line, and tab once
Write the number stored in the Counter variable
Move back down to the current line
Assign 0 to Counter
Assign Word-2 to Word-l
Search for <Hard-Return> to get to the next line
When a <Hard-Return> is not found, go to Label-for-sorting
[End of the if statement]
Go to Label-for-looping
[Emplace a label called Label-for-sorting here in the code]
[Continue with Step 7 of the analysis]
7. Sort the lines numerically on the count numbers, thereby ordering the words by their frequency of usage in the text.
The type-token report may have some blank lines at the beginning because of excess <Hard-Returns>. The words at the top of the list (i.e., the most frequently appearing words) are likely to be function words that were not removed. Words in the middle of the list are concepts or names that appear repeatedly, indicating they are recurrent topics in the document being analyzed. Words at the very bottom of the list identify concepts or names that were mentioned in the document just once. Comparing type-token reports for all documents in one's corpus can offer leads about what concepts provide the best comparisons and contrasts across documents.
Concordance studies are another form of text analysis that is easy to implement with a word processor. A concordance collects every sentence in a document where a particular word is used. Since each usage of the word is displayed with surrounding context, one can assess nuances in the word's meaning and functioning within the text, and a concordance is a powerful medium for exploring the meaning of a concept as used by a particular speaker or author. Assembling concordance-like listings of text that surrounds an inserted code is one of the standard functions in dedicated text analysis programs (discussed below), but a word processor easily can compile a traditional concordance based on words in the original document.
Below I outline a macro to collect all the sentences containing a focal word. The original document is retrieved as the word processor's Document 1; the concordance is created as Document 2.
1. Enter a focal word, and assign it to a variable.
2. Set up a search for the word in the variable, and find an instance of the word in Document 1. End p. 534
3. Copy the surrounding sentence. (Alternatively, copy the surrounding paragraph in order to get more context.)
4. Switch to Document 2.
5. Retrieve the copied text.
6. Move to the end of the Document 2 and add <Hard- Return> twice.
7. Return to Document 1.
8. Return to Step 2 until done.
Steps 2 through 7 can be constructed using the word processor's macro learning function. Steps 1 and 8 are incorporated by editing the macro and adding commands in the macro programming language. (In WordPerfect 5.1, steps 1 and 8 are incorporated by adding the following macro language commands at the beginning
{Macro Commands}al{Enter} {PAusE}{Enter}{LABEL loop~
and
{Go}loop~
at the end.)
Tagging a passage of text with an interpretive word or phrase is another one of the functions provided by dedicated text analysis programs, but a word processor's indexing function provides elementary capacities of this type. As you read through the text, add the desired words or phrases as hidden index codes. Then generate an index to show which pages have text relevant to each tag. If you search for and replace each <Hard-Return> in the document with a <Hard-Page> then the index will refer to paragraphs rather than to pages.
Text tables are a good way to restructure some kinds of existing texts, or to enter the text in the first place. For example, the transcript of a recorded interaction could be entered in a table with four columns: a sequence number; the name of the speaker; the name of the target person; and the text of the statement. After a transcript has been entered into the computer this way, it is easy to sort on column 2 to collect all statements made by a given speaker, on column 3 to collect all comments addressed to a specific person, or on column 1 to restore the original time-ordered sequence. Later, new columns can be added for tag words, and sorts can be made on those columns. Columns also might be added for numerical indexes, and word processor tables offer spreadsheet-like math capabilities that can be used to do sums. Finally, the table allows considerable control over formatting when it comes time to print the transcript.
Other applications of word processors to text analysis undoubtedly are possible. The crucial point is that word processors are among the most sophisticated computer software products available, and a qualitative researcher hardly can find a better investment of time than learning to be a "power user" of a word processor, including its macro programming language. Even after using End p. 535 computers for decades, I still am awed by the power, speed, and flexibility of the sorting routines in word processors, and after years of using WordPerfect, I continue to find new research applications for it.
Dedicated text analysis programs provide some functions that are not available in standard word processors. These programs are particularly of interest when one is working with a set of documents that are to be treated as cases for comparison or contrast, as when trying to make sense of dozens of interviews or when searching for patterns in some kind of repetitive archive, like state-of-the-union presidential speeches. Tesch (1990) provides an excellent review of the available text analysis packages, and I will not attempt to discuss them all, only two of the more prominent ones.
Ethnograph, a standard text analysis package for MS-DOS machines, requires that one begin by reformatting text in a word processor to fit Ethnograph's requirements. Then a copy of the entire text corpus is printed so that one can block passages and code the blocks. After entering the codes and line numbers of the blocks into the original data set, Ethnograph can create concordance-like listings of passages associated with any particular code. It also can count occurrences of codes and generate tabulations that show how often particular codes occurred in different kinds of cases. The program provides flexibility in assigning codes to overlapping passages, in changing codes, and in adding or removing codes. Ethnograph's major selling point is that it is an electronic replacement for scissors, paste, and index cards, and the program offers much greater speed and flexibility than pre-computer methods used by qualitative researchers while staying close to the spirit of those methods.
HyperResearch is a text analysis package for Macintosh microcomputers. The program is a HyperCard application, HyperCard being a database programming environment that Apple Computer Inc. distributes with its System 7 operating system. HyperResearch has text analysis capabilities similar to Ethnograph's. However, HyperResearch has such exceptional analytic capabilities that I can use it to introduce the notion of qualitative models.
One begins a project by importing text files prepared with a word processor. Then one does the usual blocking and coding of text, though HyperResearch greatly reduces the tediousness of the work by providing a tagging menu. Codes are typed once when adding them to the menu; thereafter one uses a mouse to block a text passage and to point and click on the desired code within the tagging menu. After the coding is done, HyperResearch can generate concordance-like listings of passages that are associated with a code, and it can count code occurrences and co-occurrences.
The unusual analytic power is as follows: HyperResearch allows End p. 536 the user to formulate and test interpretive rules involving the codes. For example, one might decide that any case that contains both codes A and B manifests code C as well; this decision is implemented as a rule to be applied to every case. Furthermore, one might hypothesize that any case with either code C or code D is to be interpreted as an X, where X is a new categorization not in the original codes. In this simple illustration, the interpretations would be formulated as two rules:
If A and B, then C.
If C or D, then X.
Now with this model having been formulated, HyperResearch will go through the cases, one by one, applying the rules in sequence and arriving at a judgment regarding whether condition X prevails in the case or not. A report is written for each case telling which rules could be applied successfully and which attempts at applying rules failed. Given this information, and the facility to examine the original text for any case, the user is able to adjust either the codes or the rules so as to get a better success rate. Ultimately, one ends up with a rule system for explaining details within cases through a few generalizations. That is, the end product is a logic model consisting of a set of rationally coherent propositions for interpreting the cases.
The illustration above is a good medium for emphasizing some of the hallmarks of qualitative analysis as they occur in textual analysis. First, counts are irrelevant in a final logic model. It really does not matter if code C is manifested in a case once or a dozen times: once is enough to decide that condition X prevails, and any other manifestations of C are merely redundant. Counts of successful applications of rules may be useful in judging whether one is getting close to a set of applicable rules or not, and once a model is completed counts may be useful in weighing the empirical importance of one logical path to the outcome as opposed to another. The logic model itself does not deal with counts, however, only with occurrence versus nonoccurrence of codes. Ragin (1987) has provided extensive discussion of this nonquantitative perspective on data.
Second, the logic model works only if one defines the codes correctly. Suppose heuristically that code C in the illustration above is the phrase "socially deprived childhood" and that, originally, it was applied when interview respondents reported that playmates and peers were lacking while the respondent was growing up. Then C may not imply condition X if, say, X stands for the phrase "adult social skills lacking." One broadens the definition of "socially deprived childhood," however, by including the combination of A and B, which, say, stand for the code phrases "both parents worked" and "siblings absent." Then perhaps it is true that being deprived of peers or being deprived of childhood family interactions implies a lack of adult social skills. In any case, the point is that one has to End p. 537 construct the meanings of general categories in particular ways in order to have one general category materially imply another. That is why coding schemes are so important and may have to be modified repeatedly before a project is done. The researcher has to devise concepts that capture the essence of textual details and that also relate to one another in useful ways, using the text as evidence of what is factual in the social reality of the people being studied. Qualitative sociological analysis is concerned with the formulation of meanings that help us understand the social world rationally.
Qualitative Comparative Analysis, or QCA, is another program for constructing logic models, but it runs on MS-DOS computers, and its approach is very different from HyperResearch's. In the first place, QCA is not text oriented at all. Rather, the data for analysis have to consist of arrays in which each row is a case, each column is a code, and each cell contains a one or a zero telling whether the case manifests the code or not. The coding categories are entirely tacit, and frequently the definition and application of codes is left to people other than the interested researcher. Since there is no linkage to original texts, it is extremely awkward to redefine codes in order to improve a model. On the other hand, QCA has more than the usual facility for counting codes and complex configuration of codes.
Second, whereas HyperResearch requires the researcher to grope intuitively for an applicable rule system and work for improvement by trial and error, QCA extracts a valid logic model from the cases and simply reports the set of logical relations that best describe the patterns of codes within the given data set. Moreover, QCA conducts its analyses of cases far faster than does HyperResearch. Once an optimal logic model has been derived, QCA allows the researcher to test alternative models, and the program reports which existing configurations of codes fit a hypothesized model and which do not. Also, QCA will present a logical formulation of what is not found empirically, which amounts to a description of a "possible world" that is just the opposite of reality.
Finally, the model formulation with QCA is different from that with HyperResearch. A user constructs a HyperResearch model as a set of if-then rules. In contrast, QCA reports its models in an alternative symbolic logic form phrased entirely in terms of and, or, and not. Let us return to the illustration above to see the difference. The symbolic logic representation of the HyperResearch model is as follows:
(Symbols have the usual meanings: /\ and; V or; ~ not; --> implies.) The QCA model, saying exactly the same thing logically, has the following symbolic logic form:
End p. 538
Obviously, implicational structuring is a simpler and more natural way of representing generalizations. The QCA representation can be converted to implicational structuring, but doing so requires the user to perform symbolic logic manipulations. The required manipulations in the example above would be the opposite of what I did to convert the HyperResearch form to the QCA form. (Assuming I got it right!)
HyperResearch and QCA offer true methodological advances to qualitative researchers trying to formulate rational propositions from their data. HyperResearch is the better choice when working with original texts because it offers the flexibility of a text analysis program in inventing and re-inventing codes and in examining the instances of codes in texts. HyperResearch also produces models that are easy to understand and easy to communicate. QCA is the better choice when one has more than a few cases (say, 20 or more), the codes are numerous and no longer subject to much revision, and one wants to move quickly and efficiently to a logic formulation that describes the data perfectly.
Ethno is another MS-DOS program for constructing logic models. Ethno's typical application is the longitudinal examination of happenings, whereas HyperResearch and QCA deal with cross-sections of cases. For the sake of continuity, however, let us first see how Ethno would compare to the other programs were it used to deal with the example above. First, Ethno's approach to data is something of a compromise between HyperResearch and QCA: we would have to reduce each case to its codes (as in QCA), but the codes would be verbal rather than numerical. Thus, for example, Case 1 might reduce to: "siblings absent in childhood"; Case 2 might be "absent from playmates and peers in childhood; lacking adult social skills," and so on.
Second, Ethno prompts the analyst to build a model as the program reads cases. Each time the program encounters a code it has not encountered, it asks the analyst to specify how that code relates to other codes. For example, while processing Case 2, the program would ask a question something like this: Does "lacking adult social skills" imply "absent from playmates and peers in childhood?" The analyst formulates the model by answering yes or no to such questions. Thereafter, whenever possible, the program uses the analyst's previous answers to deduce answers to questions, thereby minimizing the number of questions the analyst needs to answer. Thus, Ethno's approach to model building is like HyperResearch's in that the analyst is responsible for defining logical relations, and it is like QCA'S in that every possible relation is considered systematically.
Third, Ethno presents its models in diagrammatic form rather than propositionally. Thus, the two-rule model presented above, would be presented by Ethno as shown in Figure 1. Following a line upward from an item reveals the other items that are implied by the End p. 539 starting item. The implied items are grouped by and if the starting item is in lower-case, by or if the starting item is in capitals. Thus the above diagram is equivalent to the propositions: X implies C or D; and C implies A and B.
Although Ethno can be used to diagram propositions about the logical structure of codes, HyperResearch and QCA are both much better programs for analyzing cases in order to formulate and evaluate such structures. Ethno was designed for a different task: modeling the logical structure of events within happenings. The kind of code phrases Ethno expects to see are brief event descriptions like "Tom enters office," "Janet greets Tom," and so forth, and Ethno expects the events to be temporally ordered because the program analyzes sequences of events. Ethno does indeed draw a logic diagram as one of its products, but instead of talking about items implying others, it is more natural to interpret an Ethno diagram in terms of events pointing to their prerequisites. The diagram shows the logical constraints on how a given kind of happening can unfold.
Once a kind of happening has been modeled, Ethno can be used to see if the model accounts for the sequence of events in that happening and in other similar happenings. The program assumes that an event's prerequisites have to happen before the event itself can happen and that people do not ordinarily repeat an event unless it is a prerequisite for another event that needs to be enabled. So Ethno proceeds through the sequence one event at a time, checking to see whether each event's prerequisites are fulfilled and whether occurrence of the event enables another event. When Ethno finds an event that does not fit the model, Ethno reports the problem and suggests every logical way of achieving a fit between data and model, letting the analyst make the final decision about what is to be adjusted. In this way, the model can be improved until it accounts for every recorded happening of a particular type.
Ethno also allows an analyst to define an abstract model to represent superficially different happenings. For example, Corsaro and End p. 540 Heise (1990) constructed models for an approach-and-flee game played by preschoolers. The models for the American and Italian versions of the game looked quite different. Yet a more abstract rendition was developed that aligned with the culture-specific models and that accounted for game sequences in both cultures.
Formal analysis of event sequences, or narratives, is a lively topic in sociology at present (e.g., see Abell, 1987; Fararo, 1989; Heise, 1989), and Ethno's approach to event sequences is not the only one being taken. Abbott (1992) reviews most of the alternatives, including his own procedures for similarity clustering of event sequences using algorithms borrowed from biologists who classify DNA structures. Shye's (1985) extension of Guttman scales into multiple cumulative paths is another promising approach. At the time of this writing, though, the alternative procedures have not yet developed into readily accessible, user-oriented programs for microcomputers.
Much of what a qualitative researcher does could be called content analysis, and I hoped that I would be able to describe some new computer-based approaches to content analysis in this article. Programs are under development (e.g., see Franzosi, 1990; Carley, in press), but at the time of this writing they were not readily available in user-friendly forms, so the topic will have to be deferred to another year.
Network analysis deals with the presence or absence of particular kinds of relations between people, and it attempts to abstract general social structures from empirically manifest relations among individuals. Thus, network analysis can be viewed as a form of qualitative analysis, and qualitative researchers at least should be aware of Structure--the preeminent available microcomputer program for analyzing network data (see Heise, 1990, for a review). Qualitative researchers working in the field might be particularly interested in Structure's ability to convert co-presences into social networks. For example, records of who sits together in a cafeteria lunch room could be converted into a network that can be analyzed for the existence of cliques or paths of influence. I have not discussed network analysis in this essay because the topic deserves separate treatment, being one of the most mathematically advanced areas in sociology, with several new computer programs under development.
Although the core data of qualitative sociology consist of texts, sound-image recordings are becoming increasingly important (Grimshaw, 1982). I am aware of no programs existing at this time that help a qualitative researcher analyze sound-image data. (HyperResearch is promoted as a tool for handling audio and image data sets as well as text files, but the current version is limited to text only.) This unfortunate situation is bound to change in the 1990s with the diffusion of computer hardware and system software End p. 541 that is oriented toward multimedia recordings and presentations. A hint of what is coming may be glimpsed in a program called Mannequin, which I have reviewed elsewhere (Heise, 1991): after a photo of a social scene is converted to a Mannequin representation, the scene can be viewed through the eyes of any participant.
Qualitative research traditionally was considered the soft side of sociology because there seemed to be no standards for judging the procedures and products of such research. A more realistic view is that the procedures and products of such work are so enormously complex that standards of adequate sophistication are hard to formulate. Now, computer software is beginning to systematize procedures and help qualitative researchers manage their analyses more effectively. As a result, we might expect to see more rigorous qualitative studies and a greater appreciation of such work as science.
David Heise, Rudy Professor of Sociology at Indiana University, developed a mathematically implemented symbolic interactionist theory positing that people try to create impressions that confirm sentiments about social roles, behaviors, and settings (Understanding Events: Affect and the Construction of Social Action, Cambridge University Press, 1979). He is a past editor of Sociological Methodology and of Sociological Methods & Research. His methodological research has proceeded from variable-oriented quantitative modeling (Causal Analysis, New York: Wiley, 1975) to methodological problems in qualitative research (with Alex Durig, "Qualitative models," Encyclopedia of Sociology, Macmillan, 1992). He has written five published microcomputer programs and is a past chair of the American Sociological Association's Microcomputing Section. He can be reached at the Department of Sociology, Indiana University, Bloomington, IN 47405; 812-855-7231; Bitnet: HEISE@IUBACS.
Abell, P. 1987. The syntax of social life: The theory and method of comparative narratives. New York: Oxford University Press.
Abbott, A. 1992. From causes to events. Sociological Methods & Research 20:428-55.
Carley, K. M. In press. Textual analysis using maps. In P. Marsden (ed.), Sociological Methodology.
Corsaro, W., & Heise, D. 1990. Event structure models from Ethnographic data. In C. Clogg (ed.), Sociological Methodology.
Fararo, T. J. 1989. The meaning of general theoretical sociology: Tradition and formalization. New York: Cambridge University Press.
Franzosi, R. 1990. Computer-assisted coding of textual data: An application to semantic grammars. Sociological Methods & Research 19:225-57.
Glaser, B. G., & Strauss, A. L. 1967. The discovery of grounded theory: Strategies for qualitative research. Chicago: Aldine.
Grimshaw, A. D., ed. 1982. Special issue devoted to sound-image records. Sociological Methods & Research 11.
Guttman, L. 1959. A structural theory for intergroup beliefs and action. American Sociological Review 24:318-28. End p.542
Heise, D. R. 1989. Modeling event structures. Journal of Mathematical Sociology 14:139-69.
___.1990. Review of Ronald Burt, Structure: Network analysis system. Contemporary Sociology 19:799-804.
___.1991. Previewing image analysis. Contemporary Sociology 20:857-60.
Henry, N. W., & Lazersfeld, P. F. 1968. Latent structure analysis. Boston: Houghton Mifflin.
Ragin, C. C. 1987. The comparative method: Moving beyond qualitative and quantitative strategies. Berkeley: University of California Press.
Ragin, C. C., Mayer, S. E., & Drass, K. A. 1984. Assessing discrimination: A Boolean approach. American Sociological Review 49:221-34.
Robinson, W. S. 1951. The logical structure of analytic induction. American Sociological Review 16:812-18.
Shye, S., ed. 1978. Theory construction and data analysis in the behavioral sciences: A volume in honor of Louis Guttman. San Francisco: Jossey-Bass.
___.1985. Multiple scaling: The theory and application of partial order scalogram analysis. New York: Elsevier.
Stone, P. J., Dunphy, D. C., Smith, M. S., & Ogilvie, D. M. 1968. The general inquirer: A computer approach to content analysis. Cambridge, MA: MIT Press.
Tesch, R. 1990. Qualitative research: Analysis types and software tools. Bristol, PA: Falmer Press.
Ethno, Version 2. Wm. C. Brown Publishers, Software, 2460 Kerper Blvd., Dubuque, IA52001.
Ethnograph. Qualis Research Associates. POB 3129, Littleton, co 80161.
HyperResearch. Researchware, 20 Soren St., Randolph, MA 02368.
Mannequin. HUMANCAD, 1800 Walt Whitman Rd., Melville, NY 11747.
Microsoft Word. Microsoft Corp., One Microsoft Way, Redmond, WA 98052.
QCA. Professor Kriss A. Drass, Department of Criminal Justice, University of Nevada-Las Vegas, 4505 Maryland Pkwy, Las Vegas, NV 89154.
Structure Network Analysis System, Version 4.1. Research Program in Structural Analysis, Center for the Social Sciences, 420 West 118th St., 8th floor, Columbia University, New York, NY 10027.
WordPerfect. WordPerfect Corp., 1555 N. Technology Way, Orem, UT 84057.