Society for the Study of Social Problems

The Constant Comparative Method of Qualitative Analysis Author(s): Barney G. Glaser Source: Social Problems, Vol. 12, No. 4 (Spring, 1965), pp. 436-445 Published by: University of California Press on behalf of the Society for the Study of Social Problems Stable URL: http://www.jstor.org/stable/798843 Accessed: 23/07/2009 22:37
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=ucal. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

University of California Press and Society for the Study of Social Problems are collaborating with JSTOR to digitize, preserve and extend access to Social Problems.

http://www.jstor.org

436

SOCIAL PROBLEMS

ters. Even here it may be attenuated in areas with strong religious and family controls. Restrictionof this analysisto a particular metropolitan area precludes generalizationof the findings to the countryas a whole. The findings point to the desirability,however, of giving

increased attention to residence and status in investigating illegitimacy. It may well be that the generalizations basedlargelyon studiesof urbanpopulations do not apply uniformly either to all segments of the urban social or structure to the suburbs.

THE CONSTANT COMPARATIVE METHOD OF QUALITATIVE ANALYSIS* BARNEY G. GLASER

University of California Medical Center, San Francisco

Researchinto social problems,prob- topics,stigma,and legality,and because lems of deviation, of control and of people in these situations are usually crisis, and the like-the general sub- adept at covering the facts when necject matter to which Social Problems essary,often the only way a researcher is devoted-is still mainly feasible can obtain any data, or data that is through methods which yield qualita- accurate,is some combination of obtive data. Because these areas raise serving what is going on, talking in problems of secrecy,sensitivity, taboo ratherloose, sharing, fashion with the * This paper developedout of problems people in the situation, and reading of analysis arising during the study of some form of documentthat they have terminal care in hospitals; particularly the written. These methods best allow the interaction of staff and dying patients. researchereither to gain the trust of The study is sponsored by the National the people in the situation or, if necInstitutes of Health, Grant GN9077. Anselm Strauss, Fred Davis, and Stewart essary, to accomplish clandestine rePerry have been strong sources of encouragement in the preparationof this paper. I am particularly indebtedto the extensive editorial work of Robert K. Merton. Substantivepapersfrom this study are: Anselm Strauss, Barney G. Glaser, and Jeanne Quint, "The Non-Accountabilityof Terminal Care,"Hospitals,36 (Jan. 16, 1964), pp. 73-87; Barney G. Glaser and Anselm Strauss, "The Social Loss of Dying Patients," AmericanJournal of Nursing, 64 (June, 1964) pp. 119-121; Barney G. Glaser and Anselm Strauss, "Awareness American Contextsand Social Interaction," Sociological Review, 29 (Oct. 1964), pp. 669-678; Barney G. Glaser and Anselm Strauss,"TemporalAspects of Non-Scheduled Status Passage," (to be published in the AmericanJournal of Sociology); and a forthcomingbook, BarneyG. Glaser and Anselm Strauss, Awareness of Dying: A Chicago:Aldine Studyof Social Interaction, Press. search. In view of this distinctive rele-

vance of qualitative data collection and analysis for many areas of social problems, the constant comparative method of qualitativeanalysis will in particular,I trust, increasethe battery of alternativeapproachesuseful to researchersin these areas. My other purpose in presentingthe constant comparativemethod may be stated by a direct quotation from Robert K. Merton-a statement he made in connection with his own qualitative analysis of locals and cosmopolitans as communityinfluentials:
This part of our report, then, is a bid to the sociological fraternity for the practice of incorporating in publications a detailed account of the ways in which

Method of Analysis ConstantComparative

qualitative analyses actually developed. Only when a considerable body of such reports are available will it be possible to codify methods of qualitative analysis with something of the clarity with which quantitative methods have been articulated.1 SOME DIVERSE APPROACHES TO QUALITATIVE ANALYSIS

437

Two general current approachesto the analysis of qualitative data are as follows: (1) If the analyst wishes to convertqualitativedata into crudely quantifiableform in order to test provisionally an hypothesis, he codes the data first and then analyzes it. An effortis made to code "all relevantdata [that] can be brought to bear on a point," and then the assemblage, assessment, and analysis of this data is in accomplished systematically a fashion that will "constituteproof for a given
proposition."2

(2) If the analyst wishes only to generate theoretical ideas-new concepts and their properties, hypotheses and interrelatedhypotheses-the analysis cannot usefully be confined to the practice of coding first and then analyzing the data, since the analyst, in direct pursuit of his purpose, is constantly redesigning and reintegrating his theoretical notions as he reviews
1 Op. cit., p. 390. This is, of course, also the basic position of Paul F. Lazarsfeld. See Allen H. Barton and Paul F. Lazarsfeld, "Some Functions of Qualitative Analysis in Social Research," in Seymour M. Lipset and Neil J. Smelser (eds.), Sociology: The Progress of a Decade, Englewood, N.J.: Prentice-Hall, 1961. It is the position that has stimulated the work of Becker and Geer, and Berelson cited in footnote 2. 2 Howard S. Becker and Blanche Geer, "The Analysis of Qualitative Field Data" in Human Organization Research, edited by Richard N. Adams and Jack J. Preiss, Homewood: Dorsey Press, Inc., 1960, pp. 279-289. See also Howard S. Becker, "Problems of Inference and Proof in Participant Observation," American Sociological Review, Dec., 1958, pp. 652-660, and Bernard Berelson, Content Analysis, Glencoe: Free Press, 1952, Chapter III, and page 16.

his material.3Not only would analysis after a coding operation unnecessarily delay and interfere with his purpose, but explicit coding itself often seems an unnecessary, burdensometask. As a result, the analyst merely inspects his data for new properties of his theoretical categoriesand writes memos on these properties. In this paper, I wish to suggest a third approachto the analysisof qualitative data, combining, by an analytic procedureof constantcomparison,the explicit coding procedure of the first approachand the style of theorydevelopment of the second. The purpose of the constant comparativemethod of joint coding and analysisis to generate theory more systematically than allowed by the second approachby using the explicit coding and analyticprocedures. At the same time, it does not forestall the developmentof theory by adhering completely to the first approach which is designed for provisional testing, not discovering, of hypotheses. Systematizingthe second approach by this method does not supplant the skills and sensitivities required in inspection. Ratherthe constantcomparative method is designed to aid analysts with these abilities in generating a theory which is integrated,consistent, plausible, close to the data, and in a form which is clear enough to be readily, if only partially, operation3 Constantly redesigning the analysis is a well known normal tendency in qualitative research (no matter what the approach to analysis) which occurs throughout the whole research experience from initial data collection through coding to final analysis and writing. It has been noted in Becker and Geer, op. cit., 270, Berelson, op. cit., 125; and for an excellent example of how it goes on, see Robert K. Merton, Social Theory and Social Structure, New York: Free Press, 1957, pp. 390-392. However, this tendency may have to be suppressed in favor of the purpose of the first approach, but in the second approach and the approach to be presented here, it is used purposefully as an analytic strategy.

438

SOCIAL PROBLEMS

alized for testing in quantitative research. Depending as it still does on the skills and sensitivities of the analyst, the constant comparativemethod is not designed (as methodsof quantitative analysis are) to guarantee that two analysts working independently with the same data will achieve the same results; it is designed to allow, with discipline,for some of the vagueness and flexibilitywhich aid the creative generationof theory. If the person applying the first approachwishes to discover some or all of the hypothesesto be tested, his discoveriesare typicallymade by using the second approachof inspectionand memo-writing along with explicit coding. In contrast,the approachpresented here cannot be used for provisional testing as well as discovering theory, since the collecteddata, as will be seen in the foregoing description, are not coded extensively enough to yield provisional tests, as they are in the first approach.The data are coded only enough to generate,hence, to suggest, theory. Partial testing of the theory, when necessary,is left to more rigorous, usually quantitative, approacheswhich come later in the scientific enterprise. The first approachdiffersin another way from that presentedhere. The first approach is usually concernedwith a few hypotheses at the same level of generality,while the constantcomparative method is concerned with many hypotheses synthesized at different levels of generality.The reasonfor this difference is that the first approach must keep the theory tractable for provisional testing in the same presentation. Of course, the analystusing the first approachmight, after either proving or disproving his hypotheses, attempt to explain his findings with some more general ideas suggested by his data, thus achieving some synthesis at differentlevels of generality. Another approach to qualitative analysis is "analyticinduction,"which

combines the first and second approaches in a manner different from the constant comparative method.4 Analytic induction is concerned with generating and proving an integrated, limited, precise, universallyapplicable theory of causesaccountingfor a specific phenomenon,e.g., drug addiction or embezzlement.Thus, in line with the first approach, it tests a limited number of hypotheseswith all available data, which are numbersof clearly defined and carefullyselected cases of the phenomena. In line with the second approach,the theory is generated by the reformulationof hypotheses and redefinitionof the phenomena forced by constantly confronting the theory with negative cases. In contrastto analyticinduction,the constant comparativemethod is concerned with generating and plausibly suggesting (not provisionallytesting) many propertiesand hypothesesabout a general phenomenon,e.g., the distribution of services according to the social value of clients. Some of these properties may be causes; but unlike analyticinductionothersare conditions, consequences,dimensions, types, processes, etc., and, like analyticinduction, they should result in an integrated theory. Further,no attemptis made to ascertaineither the universalityor the proof of suggested causes or other properties.Since no proof is involved, the constant comparativemethod, in contrastto analyticinduction,does not, as will be seen, require consideration of all availabledata, nor is the datarestricted to one kind of clearly defined method case. The constantcomparative be applied for the same study to may any kind of qualitativeinformation,ininterviews,docucluding observations, ments, articles, books, and so forth. As a consequence,the constant com4 See Alfred R. Lindesmith, Opiate Addiction, Bloomington: Principia, 1947, pp. 12-14, and Donald R. Cressey, Other People's Money, New York: Free Press, 1953, p. 16 et passim.

Method of Analysis ConstantComparative TABLE I

USE OF APPROACHES TO QUALITATIVE ANALYSIS Provisional Testing of Theory Yes No (2) Inspection for hypotheses along with (1) coding for test, then analyzing data (4) Analytic Induction No (1) Coding for test, then analyzing data (2) Inspection for

439

hypotheses

Generating Yes Theory

(3) Constant Comparative Method Ethnographic Description

parisons required by both methods differ with respect to breadth of purpose, extent of comparing, and what data and ideas are compared. Clearly the purposes of both these methods for generating theory supplement each other as well as the first and second approaches in providing diversealternatives qualitativeanalyto sis. Table I locates the uses of these approachesto qualitative analysis and provides a scheme for locating other approaches accordingto their purposes.
THE CONSTANT COMPARATIVE METHOD

The constant comparative method can be described in four stages: (1) comparingincidents applicableto each category, (2) integrating categories and their properties, (3) delimiting the theory,and (4) writing the theory. Although this method is a continuous growth process-each stage after a time transformsitself into the nextprevious stages remain in operation throughout the analysis and provide continuousdevelopmentto the following stage until the analysis is terminated. 1. Comparing incidents applicable to each category.The analyststarts by coding each incident in his data in as many categories of analysis as possible.5 To this procedureI add the basic, defining rule for the constantcompara5 I follow the procedure for selection and coding of categories given in Becker and Geer, op. cit., pp. 271-82.

tive method: while coding an incident for a category, compare it with the previous incidents coded in the same category. For example, as the analyst codes an incident in which a nurse responds to the potential "social loss" -loss to family and occupation-of a dying patient, he comparesthis incident with others previously coded in the same category before further coding.6 Since coding qualitative data takes some study of each incident, this comparison can often be based on memory. There is usually no need to turn back to every previousincident for each comparison. This constantcomparisonof the incidents very soon starts to generate theoreticalproperties of the category. One starts thinking in terms of the full range of types or continua of the category, its dimensions, the conditions under which it is pronounced or minimized, its major consequences, the relation of the category to other categories,and other propertiesof the category. For example, in constantly comparing incidents on how nurses respond to the social loss of dying patients, we saw that some patients are perceivedas a high social loss and some as a low social loss and that patient care tended to vary positively with degree of social loss. It was also apparentthat some of the social attributes which nurses combine to establish a degree of social loss are seen
6 Illustrations in the paper will refer to "The Social Loss of Dying Patients," op. cit.

440

SOCIAL PROBLEMS

immediately(age, ethnic, social class) and some are learnedafter a time with the patient (occupationalworth, marital status, education). This further led us to the realization that perceived social loss can change as new attributes of the patients are learned. It also became apparentunder what conditions (types of wards and hospitals) we would find clusters of patients with differentdegrees of social loss. After coding for a categoryperhaps three or four times, the analyst will experience a conflict in emphasis of thought. He will both muse over these theoretical notions and try to concentrateon the study of the next incident to determine the alternate ways in which it should be coded and compared. At this point, the second rule method is: of the constantcomparative stop coding and record a memo on ideas. This rule is designed to tap the initial freshness of the analyst'stheoretical notions and to relieve the conflict in thought. In doing so, the analyst should take as much time as necessaryfor reflectingand taking his thinking to its most logical (grounded in the data, not speculative) conclusions. If one is working on a team, it is also a good idea to sit down with a teammateand discuss theoreticalnotions with him. The teammate can help bring out points missed, add points he has run across in his own coding and data collection, and crosscheck points. He, too, begins to compare the analyst'snotions with his own ideas and knowledge of the data, which generatesmore theoreticalideas. With clearer ideas on the emerging recorded,the anatheorysystematically lyst then returnsto the data for more coding and constantcomparison. 2. Integrating categories and their properties. This process starts out in a small way; memos and possible conferences are short. But as the coding continues the constant comparative units change from comparisonof incident with incident to incident with

properties of the category which resulted from initial comparisonof incidents. For example,in comparingincident with incident we discovered the property that nurses are constantly recalculatinga patient's social loss as they learn more about him. From then on each incident on calculation was compared to accumulatedknowledge on calculating, not to all other incidents of calculation. Thus, once we found that age was the most important in characteristic calculatingsocial loss, we could discernhow age affectedthe recalculation social loss as the nurses of found out more about the patient's education. We found that education was most importantin calculatingthe social loss of a middle year adult, since at this time in life educationwas likely to be of most social worth. This example also shows that the accumulated knowledge on a propertyof the category-because of constant comparison -readily starts to become integrated; that is, related in many diverse ways, resulting in a unified whole. In addition, the diverse properties of the category start to become integrated.We soon found that calculating and recalculatingsocial loss was related to the development of a social loss "story" about the patient. When asked about a patient, nurses would tell what amountedto a story about a dying patient, the ingredientsof which were her continual balancing out of social'loss factors as she learned more about the patient. We also found that the calculus of social loss and the social loss story were related to her strategies for coping with the upsetting impact on her professional composure of, say, a dying patient with a high social loss (e.g., a mother with two children). This example further shows that the categorybecomes integrated with other categoriesof analysis: the social loss of the dying patient is related to nurses' maintainingtheir professional composure while attending his dying. Thus the theory devel-

ConstantComparative Method of Analysis ops as different categories and their properties tend to become integrated through constant comparisons which force the analystto make some related theoretical sense of each comparison. 3. Delimiting the theory. As the theory develops, various delimiting features of the constant comparative method set in to curb what could otherwise become an overwhelming task. This delimiting occurs at two levels: (1) the theory and (2) the original list of categoriesproposedfor coding. First, the theory solidifies in the sense that major modificationsbecome fewer and fewer as one compares the next incidents of a category to are propertiesof it. Latermodifications mainly on the order of logical clarity; paring off non-relevantproperties;integratingelaboratingdetails of properties into the major outline of interrelated categories; and most important, reduction. By reduction I mean that a higher level, smaller set of concepts, based on discovering underlying uniformities in the original set of categories or their properties,might occur to the analyst by which to write the theory, hence, delimiting its terminology and text. An illustrationshowing both integration of more details into the theoryand some consequentreduction is the following. We decided to elaboratethe theoryby adding detailed strategies which the nurses used to maintain their professional composure while taking care of patients with varying degrees of social loss. We discovered that the rationaleswhich they used among themselves could all be considered "loss rationales."The underlying uniformity was that all rationales indicated why the patient, given his degree of social loss, would, if he lived, now be socially worthless; in spite of the social loss, he would be better off dead. (For example, he would have brain damage, be in constant, unendurablepain, or have no chance for a normal life.) By further reductionof terminology

441

we were also discovering that our theory could be generalized to one which concerns the care of all, not just dying, patients by all staff, not just nurses. Even more generally, it could be a theoryof how social values of professionals will affect the distribution of their services to clients: for example, how they decide who among many waiting clients should next receive a service and what calibre of the service to give him. Thus, with reduction of terminology and consequent generalizingwhich are forced by constant comparisons-some of which can now be based on incidents found in the literatureof other professional areas-the analyststartsto achievetwo foremost requirements theory: (1) of parsimony of variables and formulation and (2) scope in the applicability of the theory to a wide range of situations,7 while keeping a close correspondenceof the theoryto data. Second, delimiting the theoryresults in a delimiting of the original list of proposed categories for coding. As the theorygrows, reduces,and increasingly works better in ordering a mass of qualitativedata, the analystbecomes committedto it. This commitment now allows him to delimit the original list of categoriesfor coding according to the boundariesof his theory. In turn, his consideration, coding, and analyzing of incidentsbecomemore select and focused. He can devote more time to the constantcomparison incidents of clearly applicable to a smaller set of categories. Another factor, which then further delimits the list of categoriesfor coding, is that categories become theoreticallysaturated.After one has coded incidentsfor the same categorya number of times, it becomesa quick operation to see whether or not the next applicable incident points to a new aspect of the category. If yes, then the incident is coded and compared.
7 Merton, op. cit., p. 260.

442

SOCIAL PROBLEMS

If no, the incident is not coded, since it only adds bulk to the coded data and nothing to the theory.8For example, once we had established age as the base line for calculating social loss, it was no longer necessary to code incidents referring to age in calculating social loss. However, if we came across a case where age did not appearto be the baseline (a negative case), it was coded and then compared.In the case of an 85-yearold, dying woman who was considered a great social loss, we discoveredher "wonderful personality" outweighed her age as the most important factor in calculatingher social loss. The fact that categories become can saturated be employed theoretically as a strategy in coping with another problem: new categories will emerge after hundredsof pages of coding. The question is whether or not to go back and re-codeall previouslycoded pages. The answer for large studies is "no," not until starting to code for the new categoryat the page when it occurs,and waiting for a few hundred pages of coding, or when the remaining data have been coded to see whetheror not the new category has become theoIf reticallysaturated. yes, then it is not to go backbecausetheoretical necessary saturationsuggests that what has been missed will in all probability have little modifying effect on theory. If the category does not saturate, then it is necessaryto go back and try to

saturate it, if the category is central to the theory. Theoretical saturation helps solve anotherproblemconcerningcategories. If the analyst has also collected the data, then he will be remembering from time to time other incidents he observed or heard that were not recorded. What does he do? If the unrecordedincident applies to an established category, it can, after comparison, either be neglected as a saturated point or, if it is a new property of the category, it can be added into the next memo and thus integrated into the theory. If the remembered incident generates a new category, both incident and category can be included in a memo bearing on their place in the theory.This may be enough data if the category is minor. However, if the category becomes a central part of the theory, the memo becomes a directive either for returning to the notes for more coding, or for returningto the field or library for more data or for future research. The universe of data used in the constant comparativemethod is based on the reductionof the theory and the delimination and saturation of categories. Thus, the collected universeof data is theoreticallydelimited and, if necessary,carefully extended by a return to data collection according to theoretical requirements. This theoreticaldelimiting of the universeeconomizes research resources, since it 8 If the purpose of the analyst, besides forces the analyst to spend his time developing theory, is also to count inci- and effort on data relevantonly to his dents for a category to establish provisional For large field studies with proofs, then he must code the incident. categories. Professor Merton has made long lists of possibly useful categories Furthermore, that and thousandsof pages of notes emthe additionalpoint in correspondence countingfor establishingprovisionalproofs bodying thousands of incidents, each may also feed back to the developmentof of which could be coded a multitude theory, since frequencyand cross-tabulation of ways, theoretical criteria are of of frequenciescan also generatenew theoretical ideas. See Berelson on conditions great necessity in paring down an under which one can justify time consum- otherwise monstrous task to the reing, careful counting, op. cit., pp. 128-134.

See Beckerand Geer for a new method of

counting frequency of incidents, op. cit., pp. 283-287.

sources of the people and the project's allotted time and money. Without these criteria the delimiting of a

ConstantComparative Method of Analysis universe of collected data, if done at all, can become very arbitrary, less likely to yield an integrated product; and the analystis more likely to waste time on what might later prove to be irrelevantincidents and categories. 4. Writing theory. At the end of this processthe analysthas coded data, a series of memos, and a theory. The discussionsin the memos provide the content behind the categories, which are the major themes of the theory as written in papers or books. For example, the major themes (section titles) for our paper on social loss are "calculating social loss," "the patient's social loss story," and "the impact of social loss on the nurse's professional composure." To start writing one's to theory,it is firstnecessary collate the memos on' each category, which is easy since the memoshave been written according to categories. Thus, all memos on calculatingsocial loss were brought together for summarizing and, perhaps,further analyzing before writing about it. The coded data is the resourceto return to when necessary for validating a suggested point, "pinpointing"data behind an hypothesis or gaps in the theory,9and providing illustrations.

443

also attempting provisional proofs, accompanying crude tables. If the theory encompasses a multitude of ideas, it becomes too cumbersometo illustrate each idea and, even if space were allowed, too burdensometo read many illustrationswhich interruptthe flow of general ideas.1' Thus qualitative analystswill usually present only enough material to facilitate comprehension, which is typicallynot enough data to use in evaluating all suggestions. Another way to convey credibility of the theory along with the use of illustrationsis to use a codified procedure for analyzing data, such as presented here, which allows readers to understand how the analyst obtained his theoryfrom the data. In qualitative analyses the transition from data to theory is hard, if not impossible, to grasp when no codified procedure is used.12And in his turn the reader is likely to feel that the theory is somewhat impressionistic,even if the analyst strongly asserts he has based it on hard study of data gatheredduring months or years of field or library research. Even such codified proceduresas a search for negative cases or a consideration of alternative hypotheses13 DISCUSSION will leave a readerat a loss, since these Conveying credibility. A perennial analyticproceduresare not linked with problem with qualitative analysis is proceduresfor using qualitative data. conveyingthe credibilityof a theory.10 The standardapproachto this problem in 11 See detailed discussionon this point Strauss, et al., op. cit. is presenting data as evidence for con12 Following Merton's quotation (page of clusions, thus indicating the way by 437), we need moredescriptions methods which the analyst obtained the theory of transitionfrom qualitativedata to qualifrom his data. However, since qualita- tative analysis.Barton and Lazarsfeld(op. cit.) delimiting the various functions of tive data do not lend themselves to qualitative analysis indicate a full range the readysummary, analystusuallypre- of purposes for which other methods of illustrationsand, if transition can be developed. In focusing sents characteristic
discussionon these purposesthey hit upon what might be considered elements of possible such methods. To analyze a purpose and the analytic operations involved in its final achievementis not, however, to be construedas a method of transition that guides one the full route from raw qualitativedata to accomplished purpose. 13
Becker, op. cit., p. 290.

9 On "pinpointing" see Anselm Strauss, Leonard Schatzman, Rue Bucher, Danuta Ehrlich and Melvin Shabshin, Psychiatric Ideologies and Institutions, New York: Free Press of Glencoe, 1964, Chapter 2, "Logic, Techniques and Strategies of Team Fieldwork." 10 Becker, op. cit., p. 659.

444

SOCIAL PROBLEMS

They do not specify how and how long to search for negative cases or how to find alternativehypothesesgiven a specifiedbody of qualitativedata.Thus the analyst can still be suspect in making his theory appear credible by biasing his search for negative cases or his reasonable alternative hypothmethod eses. The constantcomparative with standardanalyticprocedures joins directives for using the data systematically. In addition, keeping track of one's ideas, as requiredby the constantcomparativemethod, raises the probability that the theory will be well integrated and clear,since the analystis forced to make theoretical sense of each comparison. Making sure the categories and their propertiesof the theory are meaningfully interrelated is difficult enough; keeping all the interrelations clearlydelineatedis an addeddifficulty. The integration and clarity of the theory will in turn raise the probability that it will be understood and believed credible by colleagues. Properties of the theory. The constant comparativemethod raises the probability of achieving a complex theory which corresponds closely to the data, since the constant comparisons force considerationof much diversityin the data.By diversity,I mean that each incidentis comparedto other incidentsor to propertiesof a category by as many of its similar and diverse aspects as possible. This way of comparing may be seen in contrast to coding for crude proofs, which only establisheswhether or not an incident indicatesthe few propertiesof the category which are being counted. The constant comparisonsof incidents on the basis of as many of their similaritiesand differencesas possible tend to resultin the analyst'screatinga developmentaltheory.14In comparing
14 Recent calls for more developmental, as opposed to static, theories have been made by Wilbert Moore, "Predicting Discontinuities in Social Change," American

incidents, the analyst learns to see his categories as having both an internal developmentand changing relationsto other categories. For example, as the nurse learns more about the patient, her calculationsof social loss change; and recalculationschange her social loss stories, her loss rationales and her care of the patient. Thus, while this method can be used to generate static theories, it especially facilitates the generation of theories of process, sequence,and change which pertainto organizations,positions, and social interaction. This is an inductive method of theory development. In making theoretical sense of much diversity in his data, the analyst is forced to develop ideas on a level of generality which is higher than the qualitativematerial being analyzed.He is forced to bring out underlyinguniformitiesand diversities and to account for differences with single, higher level concepts. He is forced to engage in reduction of terminology, as discussed above, to achieve mastery of his data. If the analyst starts with raw data, he will at first end up with a substantive theory: a theory for the substantive area on which he has done researchfor example, patient care or gang behavior. If the analyst starts with the findings from many studies which pertain to an abstract sociologicalcategory, he will end up with a formal theory for a conceptualarea such as stigma, deviance, lower class, status congruency, or referencegroups. To be sure, the level of generalityof a substantive theorycan be raisedto a formal theory (our theory of social loss of dying patients could be raised to the level of how professional people give service to clients accordingto their social value). This requiresadditionalanalySociological Review, June, 1964, p. 332; Howard S. Becker, Outsiders, New York: Free Press, 1962, pp. 22-25; and Barney G. Glaser and Strauss, Awareness Contexts and Social Interaction, op. cit.

Method of Analysis ConstantComparative sis of one's substantivetheory, and the analyst should include material from other studies with the same formal theoreticalimport,however diversethe substantive content.15 The analyst should be aware of the level of generality at which he starts in relation to the level at which he wishes to end up. The constant comparative method can yield either property or propositional theory. The analyst may wish to proliferate many properties of a categoryor he may wish to write propositions about a category. Property theory is often sufficient at the exploratorystage of theory development and can easilybe translatedinto propositions if the work of the reader requires a formal hypothesis. For ex-

445

ample, two related properties of a dying patient are his social loss and the amount of attention he receives from nurses. This can easily be restated as a proposition: patients considered a high social loss compared to those considered a low social loss will tend to receive more attention from nurses.
of these coherent analytic perspectivesis not likely to come from those who restrict their interestexclusivelyto one substantive area," Erving Goffman, Stigma: Notes on the Management Spoiled Identity,Engleof wood Cliffs, N.J.: Prentice-Hall, Inc., 1963, p. 147. See also ReinhardBendix, "Concepts and Generalizationsin Comparative Sociological SociologicalStudies,"American Review, August, 1963, pp. 532-539.
15 ". . the development of any one

THE TEXTBOOK WORLD OF FAMILY SOCIOLOGY HYMAN RODMAN Institute Merrill-Palmer Many things have been much said about family sociology: it deals with sensitiveissues and thereforethe objecand tivity of researchers the acceptability of researchby the public have come slowly; everybody is an expert in family sociology, having lived most of his life in families, and thereforemuch of family sociology is trivial and commonsensical; many groups have a vested interest in "the family" and therefore family sociology is either conservativeor insolently subserviently radical; family sociology boasts of many diverse studies but few binding theories;it has low status,unless sprinkled with terms like "kinship" or "comparative" or "structural-functional"; it is popular with students (either becauseof intrinsic interest or easy grading) and therefore suspect. Despite this rather gloomy picture, families of one form or another are universallyfound, and universallyperform importantfunctions for individuals and society, and as a consequence family sociology is an importantarea of researchand has producedsome of the most importantstudies in the social sciences. The above issues have been dealt with manytimes, and I shall therefore not elaborate. Nor shall I go into a general review of family research, because the family area has perhaps had more than its fair share of such reviews and critiques.Textbooks,however, have seldom been looked at critically, except in the course of reviews about a single text at a time. I shall therefore,in a criticalvein, concentrate upon family sociologytextbooks,to the relativeneglect of readers,and of texts that are primarily practical (family life) or cross-cultural(anthropological) in orientation. I have made no attempt to rank the texts in order of
their excellence; all of those that I