Evaluation and Program Planning 26 (2003) 41–44 www.elsevier.com/locate/evalprogplan
Book Review Fuzzy-set social science C.C. Ragin (Ed.); University of Chicago Press, Chicago, Illinois, 2001, 352 pages, paperback, ISBN 0-225-70277-4.
1. Introduction Tolstoy is supposed to have said that all happy families are the same, but they are not. Even a short list of examples makes clear that the recipe for family happiness can contain various combinations of ingredients. To explore this diversity, it often does not take us very far to use a variable (e.g. leisure, high levels of communication, or physical proximity) to explain a portion of the variance in the distribution of happiness in a population of families. More intriguing by far than a variable’s average influence across families is its role as an element in the specific configurations of variables that make up the special chemistry of family happiness. No understanding is complete that leaves out the numerous ironies of family happiness, the way a single variable (e.g. large size) may in some circumstances promote happiness, yet in others work against it—sometimes in the same family at different times. Charles Ragin may be able to help researchers who find it frustrating to observe irony, paradox, and causal complexity in the social processes they study, yet find they must struggle to represent them with the tools at their disposal. In Ragin (1987) first book, The Comparative Method: Moving Beyond Qualitative and Quantitative Strategies, he drew on logic and set theory to formalize Qualitative Comparative Analysis (QCA), a rigorous technique for holistic qualitative comparison that uses categorical variables to specify sets of necessary and/or sufficient conditions associated with outcomes. Fundamentally non-statistical, QCA’s strengths are its explicit concern with the exploration and construction of types and kinds of cases, allowing the investigator to see meaningful subtypes in an apparently homogenous group, and its interest in situations when variety of pathways may lead to a single outcome. QCA has now been extended to incorporate probabilistic criteria in the book under review, Fuzzy Set Social Science (FSSS). In it, he offers a bold, carefully conceived extension of his prior work that is particularly well-suited for evaluation and program planning professionals who want to expand their methodological tool kit to include choices other than case studies or multivariate modeling. PII: S 0 1 4 9 - 7 1 8 9 ( 0 2 ) 0 0 0 9 3 - 9
The technical material is demanding in a few of the chapters, but the reader is assisted by the generous use of tables and illustrations, and by succinct chapter introductions and conclusions that orient the reader to the progress of the argument. The software needed for both the original, and the updated, techniques is free and can be downloaded from his website at the University of Arizona (http://www.u. arizona.edu/~cragin/QCA.htm) where Ragin teaches. Readers may also be interested to know that, between these two books, he published in 1994 a fine introductory text on research methods, Constructing Social Science.
1.1. Outline of the book Ragin admits that “without the frame of the diversity oriented approach, the value of fuzzy sets to social science is limited” (p. 12). Accordingly, the first half of FSSS consists of five chapters on ‘diversity oriented research’ that recapitulate and update material and arguments from the first book. Any QCA analysis has three steps, case selection and definition; testing the necessity and sufficiency of various causes or conditions in relation to an outcome, and evaluating the results. Ragins believe this important first step is the most difficult. In mainstream research, much of the conceptual work involved in constituting and defining the boundaries of a sample or a population ordinarily takes place off-stage, with the sometimes unfortunate result that, as the boundaries become fixed, and the homogeneity of the cases becomes a taken-for-granted fact of research life. Ragin wants to throw a spotlight onto this work. (Is this group of coresidents a family? Is that group of blood relations a family?) Decisions are made more explicit, and more meaningful, he argues, when cases are conceptualized as configurations of attributes or set memberships. Moreover, this approach is better equipped to capture the high degree of causal complexity typical of social phenomena. Many social processes entail set-theoretic relationships that are not brought into the open by correlation methods, Ragin believes. For example, consider a somewhat oversimplified illustration (based on Ragin, p. 11) in which there are two garden-variety claims that might appear in a state program evaluation. (1) For-profit methadone clinics have a high volume of patients. (2) Mental health facilities billing federal entitlement programs, such as Medicaid, are licensed by the state. It is simple enough to approach each
42
Book Reviews / Evaluation and Program Planning 26 (2003) 41–44
by checking for a correlation. If a facility’s billing of federal programs correlates with possession of a state license, and if a methadone program’s for-profit status correlates with high volume, then there is evidence for a possible causal link in each case. In each case, the subset term comes first (for profit clinics in the first example, facilities that bill in the second). But Ragin stresses that the causal logic is dramatically different. In the first case, the need to make a profit drives the volume, but for-profit status is only one possible source of high volume. (Another might be location in a high-need area with no other clinics in the vicinity.) For the mental health clinics, the practice of billing entitlement programs is made possible by possession of a license. The direction of influence is different, running from the second term (licensing) to the first (billing), and possession of a license is not simply one route among others. It is, for all practical purposes, a condition that must be met to bill Medicaid. Thinking of set-theoretic relations can force these issues to the researcher’s attention. To use QCA, one needs to treat all outcomes and variables as categorical. This is most straightforward when each can be described based on the presence or absence of a single attribute that defines membership in a set (happy family vs. not-happy family), but a little fiddling of the sort familiar from traditional approaches can be used to accommodate data that are continuous or involve multiple categories. Using cases as rows and attributes as columns, a matrix is created listing the evidence on the cases. When cases have the same attributes, they should have the same outcome. The matrix is then transformed into a truth table with rows that describe each configuration of attributes, including its outcome, as a configuration of attributes that are present or absent. In contrast to covariation techniques, which look for the independent influence of each variable, QCA focuses from the start on combinations, comparing to one another the various configurations of causes found in the rows of the matrix. When two combinations differ only by a single attribute, and the outcome is unaffected, the attribute is ruled out as a cause and dropped. When repeated, this process produces a series of simplified expressions. What might an evaluation researcher learn? To evaluate the success of minority professionals in overcoming glass ceilings in traditionally non-minority agencies, relevant variables might include the presence of determined mentors, aggressive recruitment of minority candidates, transparent criteria for promotion (presumed to cut back on ‘old boy’ promotions), a strong regional training program producing minority candidates, limited local alternatives for minority professionals, etc. When reduced, the data indicate two combinations of conditions can produce success: (a) determined mentors and aggressive recruiting; or (b) transparent criteria for promotion and limited local alternatives for minority professionals. No single condition (or combination of conditions) is necessary for the outcome. No single condition is sufficient. Only the two combinations of conditions are sufficient. The practical importance of
sufficiency findings is to bring home the message that institutions operate in open systems and that there may be more than one way to accomplish a goal. Necessity findings (e.g. no program succeeds without determined mentors) show how determined efforts can come to nothing if they lack a feature and how long successful programs can collapse when a necessary element erodes. The second half of FSSS uses six chapters to spell out his technical extension of QCA, introducing fuzzy sets as a research tool. As Ragin uses them, fuzzy sets are the duckbilled platypus of the measure world. They mix characteristics that aren’t supposed to go together. As with a nominal scale, a fuzzy set indicates membership in a class. The notion of ‘degree of membership’ in a well-defined set allows one to say, for example, that a given family’s membership in the type ‘happy family’ can vary from 0 to 1.0. With quantified membership, fuzzy sets can have equal intervals, a fixed meaningful zero point (for non-members) and a fixed maximum (full membership), offering a precision and flexibility not possible with a nominal scale. Despite appearances, this quantification of set membership is not a simple transformation of a binary variable into a continuous one; it is instead, he urges, an ‘interpretive algebra’ that is ‘half-verbal-conceptual’ and ‘half-mathematical analytic’ (p. 5). Often early efforts to develop a QCA truth table with a minimum of contradictions don’t click into place. The investigator must re-examine and re-think the cases and the variables selected. These are far from being ad hoc or arbitrary efforts to fit a model to the data, however. Ragin has this process follow a logic of disciplined discovery and clarification. When degrees of membership are permitted, population boundaries no longer enforce an assumption that cases are homogenous. (The degree of membership in the set ‘families’ might be 0.60 for this group of co-residents, and 0.75 for that group of blood relations.) It becomes possible to describe variation without surrendering the emphasis on types that characterized QCA. Ragin recognizes, however, that contradictions may sometimes remain. One could just ‘resolve’ them using various interpretive rules of thumb, keeping these resolutions from being arbitrary by requiring that the rules of thumb be based in theory, but in FSSS he shows how adoption of probabilistic criteria can add precision to these rules of thumb. If, for example, the combination of physical proximity and leisure was present in . 80% of the happy families under study, then we might conclude this combination was ‘almost always sufficient’. Data analysis using fuzzy sets often follows a logic articulated in the first book, but, because it is arithmetic, the procedure for determining subset membership is more technical, and not as intuitive as the earlier QCA approach. If the fuzzy set membership scores on the outcome (e.g. happy family) are uniformly less than or equal to the scores in a cause (e.g. leisure), then leisure can be considered a necessary condition for the family happiness outcome. If the
Book Reviews / Evaluation and Program Planning 26 (2003) 41–44
outcome scores are uniformly greater than or equal to the scores on the cause, then leisure can be considered a sufficient condition. (If one plots membership in the outcome along the vertical axis and membership in the causal condition along the horizontal, then cases in the upper left triangle are consistent with sufficiency, and cases in the lower right triangle are consistent with necessity.) More intuitive are fuzzy set rules for combinations. For logical ‘and’, the researcher uses the minimum membership score. A case might be a member of the set of families with devout religious faith, at 0.91 (not a full member, but almost), and the set of families with leisure time 0.34 (more out of the set than in, yet still not a bona fide non-member). In that case, its membership in the set of families with devout religious faith and leisure is 0.34. With logical ‘or’, the researcher focuses on the maximum instead. Membership in the set of families with either devout religious faith or leisure is 0.34. (I find this approach is consistent with common sense intuitions. A dark-haired happy family is not a member of the set of families that are happy and blond, no matter how happy it is. However, it is fully a member of the set of families that are happy or blond, no matter how darkhaired it is.)
1.2. The quantitative/qualitative divide and diversityoriented research Part of the appeal of both books comes from the author’s ability to mix sharp observation with insightful interpretation, an intellectual style that is nicely displayed by an intriguing graph on page 25 of FSSS. It plots the relative number of studies conducted in various areas of the social science against the N of cases in each study. The simple U-shaped curve one finds there calls attention to the fact that there are lots of studies with many cases, and lots of studies of a single case, or perhaps a few cases, a pattern that can be seen as a reflection of a real, perhaps unavoidable, trade-off between the number of cases and the number of features of cases that can be included in any given study. On the right side of the Ushaped curve are typical quantitative studies that condense data to focus on the covariation of a few features across many cases. On the left side are typical qualitative studies that examine many features of a few cases to see how the features fit together, identify meaningful commonalities that can help clarify concepts, and, ultimately, elucidate the significance of practices. By almost every measure, the two seem worlds apart. Vocal regrets over the qualitative/quantitative division in social science are common. Many amount to little more than nostalgia for an era of larger-than-life ‘renaissance researchers,’ able to combine theory, quantitative and qualitative methods, practical relevance. But there are serious efforts besides Ragin’s. Two are worth mentioning, both because Ragin explicitly tries to rebut their proposed
43
solutions, and because each helps bring out what is distinctive about his approach. One approach is frank pluralism, using common sense and educated judgment to combine the best features of various methods. Robert Alford’s (1998) Craft of Inquiry, for example, urges students to recognize that ‘diversity of methods’ is a strength, not a weakness. It is hard to argue with this attitude (which I think most readers of this journal would endorse), but a second alternative has been forcefully articulated in Designing Social Inquiry: Scientific Inference in Qualitative Research (1994). Authors King, Keohane, & Verba (1994) reject eclectic pluralism and argue that differences between qualitative and quantitative methods are stylistic, not substantive or methodological, that a single, unified logic governs inference in each domain, and that, as a practical matter, qualitative research can be improved by systematically applying methodological demands typically associated with quantitative research. Ragin’s starting point is quite different, and his sharp eye is very much in evidence. His main concern is with the scarcity of studies focused on a middling number of cases, small enough to allow for some in-depth knowledge of each case, but large enough to make sense of diversity by unraveling patterns of similarities and differences. His home discipline is comparative sociology, with its focus on determining how differing causal conditions yield different outcomes. Two facts about this area of study provide an informative context for his methodological innovations. First, many important events are comparatively infrequent (e.g. armed revolt, political secession by ethnic minorities, collapse of democratic government). Second, the circumstances of such an occurrence can seem to be present, yet the event does not occur. The investigator wants to figure out why pressure from the International Monetary Fund, government scandals, and an active labor movement sometimes produces a revolt, and sometimes do not. Analogously, the program evaluation researcher faces similar questions with rare events, such as the closing of a state mental hospital, or trying to figure out why one community -based program retain the political agenda of its founders and another depoliticizes, hires professional staff, and moves to the mainstream. A not uncommon outcome of this sort of approach to data is that in that consideration of causes ends up bringing into relief subgroups that were not initially apparent (e.g., two distinct types of communitybased programs now in the mainstream).
2. Criticisms Ragin several times insists that he believes strongly in the importance of conventional variable-oriented research. Sometimes diversity-oriented research is officially envisioned as a complement to case studies and large-N studies, but other times the reader suspects that appropriately
44
Book Reviews / Evaluation and Program Planning 26 (2003) 41–44
employed variable-oriented research might have a very limited role in Ragin’s ideal world. Readers ought to understand, and I expect Ragin would admit, that his working image of the ‘conventional’ researcher is something of a rhetorically useful simplification, significantly more literal minded than the flesh and blood sort usually encountered. Perhaps inevitably, the contrasts sometimes seem overdrawn. FSSS goes to great lengths to compare and contrast its approach with more conventional methods; nevertheless the book has an odd hermetic quality. Perhaps it is an inevitable price to be paid for the extended focus and singleness of vision required to develop the fuzzy-set approach. The only intellectual source given extended treatment is Paul Lazarsfield’ notion of a ‘property space’ (Barton, 1955). There is not much recognition of other maverick approach to causal complexity (e.g. Abbott, 2001). It is clear that the incorporation of fuzzy set logic allows Ragin to overcome limitations of QCA, but he often fails to link innovations to the criticisms QCA received. This book articulates a strong vision, and these criticisms add up to very little in that context. As Ragin would admit, I am sure, the real value of fuzzy set research will be difficult to judge for some time, since it ought to be judged by the fruit it bears. The yield has already begun for QCA, which
* Corresponding author. Tel.: þ 1-732-932-1171.
has been applied to an impressive range of topics. There is every reason to believe that fuzzy set analysis will make a major contribution.
References Abbott, (2001). Time Matters. Chicago,Ill.: University of Chicago Press. Alford, R. R. (1998). The craft of inquiry: Theories, methods, evidence. New York: Oxford. Barton, A. A. (1955). The concept of property space in social research. In P. F. Lazarsfeld, & M. Rosenberg (Eds.), The language of social research. Glencoe: Free Press. King, G., Keohane, R., & Verba, S. (1994). Designing social inquiry: Scientific inference in qualitative research. Princeton: Princeton University Press. Ragin, C. (1987). The comparative method: Moving beyond qualitative and quantitative strategies. Berkeley: University of California Press. Ragin, C. (1994). Constructing social research. London: Pine Forge Press.
J. Walkup* Rutgers University, Graduate School of Applied and Professional Psychology, Pitcataway, NJ, USA E-mail address:
[email protected]