Goldthorpe+Mjøset 2005 CanGroundedTheorySolveTheProblemsOf ItsCritics

Sosiologisk Can Grounded 13, 379408 Universitetsforlaget 2005 tidsskrift: VOL Theory Solve the Problems of its Critics?
CAN GROUNDED THEORY SOLVE THE PROBLEMS OF ITS CRITICS?1
Lars Mjset
lars.mjoset@sosiologi.uio.no
Summary
With reference to a typology distinguishing six notions of theory, grounded theory is a case of the explanation-based type of theory reflecting a pragmatist attitude. This essay evaluates the criticism waged against grounded theory. John Goldthorpes criticism, a side effect of his attempt to synthesize two standard notions of theory, leads him to establish an unbridgeable theory/context divide which, in the end, he cannot sustain. Grounded theory is not committed to such a divide and can thus solve one of Goldthorpes main problems, which is that he cannot establish any theory (and thus not explain) without some delimitation of scope or context. Michael Burawoys criticism reflects his commitment to a critical notion of theory, but he fails to argue convincingly how his own extended case method can be distinguished from grounded theory. His ability to reflect on ethnographic research is hampered by a social-philosophical attitude (more specifically, a Marxist variety of that attitude). Grounded theory can help Burawoy out of at least one of his main problems, which is that successful pursuit of the extended case method requires that the determining macrocontext is analysed in a grounded way. In terms of concepts that we define in this essay, both the standard and the social-philosophical attitude pursue high-level notions of theory. Our discussion supports the broader conclusion that such notions tend to hamper the accumulation of empirical knowledge in the social sciences. If the pragmatist attitude is one that also standard and social-philosophical social scientists will always be forced to accommodate in some sense, why not consider it and the low/middle-level notions of theory it implies in its own right? Keywords: Grovnded theory, philosphy of the social sciences, the extended case method, John Goldthorpe
SOSIOLOGISK TIDSSKRIFT
379
Lars Mjset
Social science has its science wars. At stake is whether one or the other highlevel notion can claim the status of being the relevant notion of theory (and thus of science) in this area of systematic search for knowledge. When the battle is over, social scientists return to their daily activities, and while the general staffs may still care for the high-level notions of theory, the rank and file continue to analyse and explain, implying notions of theory at lower levels than the high notions they fought over in the war. Sometimes the generals worry, perhaps due to intelligence reports about lack of attention to the cannons of science among the footsoldiers. On both the warring partners sides, the generals then issue declarations which imply that these low or middle-level notions of theory are no better than the enemys notion: they betray the nation that is, they betray science! In this essay we look at two such declarations. They represent different camps that now and then go to war against each other, but both see the discovery of grounded theory (Glaser & Strauss 1967) as an enemy activity: John Goldthorpe sees it as yet another expression of the anti-positivist attack on proper social science, while Michael Burawoy regards it as a positivist assault to be resisted by anyone who wants to defend reflexive and critical social science. By means of critical discussion of both, we arrive at the conclusion that no general can manage without taking seriously what the rank and file actually do! Grounded theory can solve some of the problems of its critics!
A typology of notions of theory

We shall start from two schemes that are drawn from a comparison of the various meanings given to the term theory in social science (Mjset 2001, revised in Mjset 2006, forthcoming). Table 1 challenges the persistent dualism invoked in discussions on the philosophy of the social sciences. A distinction between three practical philosophies of social science common attitudes (habits of thought) among groups of researchers are offered as a more productive alternative. Very briefly, the arguments behind this distinction is as follows: Any scholar who has done social research for some time knows that a large group of social scientists conduct research and legitimate what they are doing in ways that refer to routines similar to those employed in various natural sciences (experimental designs, mathematical modelling techniques, etc). Let us call this the standard attitude among social researchers. We also know that there is another group of scholars who proceed in ways that remind us of work in the humanities (interpreting texts, reconstructing culturally significant events, reflecting on the existential challenges of our time, etc). Let us call this the social-philosophical attitude among social scientists.
380
Can Grounded Theory Solve the Problems of its Critics?
The claim here is not that researchers with such attitudes rigidly copy these other fields of science, but that their inspiration is mainly drawn from models and philosophies of science relating to these fields, and that each type of social researcher communicates reasonably well with researchers in their respective fields of science (either natural science or the humanities), but not with those within the other field. The dualism between natural science and humaniaties cultures is not new. Less common, however, is the separation of a third gorup of researchers who simply behave as social scientists, with no external inspiration. Let us call this the pragmatist or participationist attitude. The suggestion here is that when we reflect on the philosophy of the social sciences, we distinguish these three attitudes, not just the usual two. A more specific term for these attitudes is: practical philosophies of social science. The distinction helps us avoid the engrained polarizations between e.g. between explanation and understanding or between positivism and anti-positivism that has too long haunted the discussion on the fundamentals of social science.
Table 1. Six notions of theory

Standard dualism Practical philosophy of social science Model of science Notion of theory Main empirical procedures Positivism Critique of positivism
Standard
Pragmatist (participatory)
Social-philosophical
Natural science Middle range theory Variables oriented, cases treated in variablesterms Idealizing
Social science Explanation- based Critical
Humanities Transcendental Diverse sources on modernity Deconstructionist Conceptual history, discourse/ conversation analysis
Primary Cases as the Cases explanatory reference to mainly contemporary basis treated as legitimate illustrations cases of legitimate claims for social change
Table 1 further shows how each attitude involves two different notions of theory. Among researchers whose attitude is the standard one, we find either middle range theory or the idealizing notion of theory (Mjset 2001). Researchers with a pragmatist attitude invoke either explanation-based or critical
381
Lars Mjset
notions of theory, while social philosophers entertain either transcendental or deconstructionist notions of theory. This six-fold typology is a modest contribution to the classification and comparison of notions of theory that are implied in contemporary social science research. The typology intends to cover all notions of theory that are currently accepted in the various disciplines of social science (e.g. when phd dissertations or journal articles are evaluated). Several of the notions may be marginal in certain disciplines (in economics, for instance, the two standard ones largely dominate), but they are all involved in more than marginal positions in some discipline. Sociology it must be admitted is extreme, since all six notions seem to flourish simultaneously in our discipline. Table 2 locates the six notions of theory at different levels, referring to the case level at the most concrete (low) end of a continuum that has decontextualized general theory at the other end. The table also indicates the basic gap between understandings of theory as pre- or non-empirical knowledge and understandings of theory as knowledge that is empirical/substantive. Theory at the low and middle level is regarded as such empirical or substantive knowledge.
Table 2. Notions of theory and levels of analysis

Attitude Notion of theory High level Middle level Low level X Standard Law-oriented Idealiazing Pragmatist Explanation- Critical based Social-philosophical Transcendental X X X X X X Deconstructionist
Note: X marks the main level which the notion of theory relates to, marks an ambition to move upwards to the next level, marks an ambition to move downwards.
The rest of this essay employs this framework for a specific purpose. As was signalled by the introductory metaphor of social science wars, we should worry about the dominance of high level notions of theory in these wars. As long as the wars that erupt are between generals who defend high level notions of theory, the methodological and philosophical concerns with the middle and lower level notions theory suffer. Substantive research, both qualitative and quantitative, becomes simply a question of research techniques. This essay, then, aims to contribute to a grassroots protest, a revolt of the footsoldiers: it aims to defend the view that middle or lower level notions of theory can very well be
382
legitimated in terms of arguments drawn from the philosophy and sociology of scientific knowledge. The pragmatist notions of theory are known in social science, but they are seldom merited attention in discussions on the philosophy of the social sciences. This is clearly the case with explanation-based theory (e.g. grounded theory). As for critical theory, it is more frequently discussed, but only to be treated as an expression of the social philosophical attitude. A main concern in this essay is to see how grounded theory fares under attack from higher level notions of theory. We investigate whether grounded theory deserves to be kept out of the principal discussions. We first use the sixfold typology to define the specific features of grounded theory as a notion of social science theory. Then we evaluate two different criticisms waged against grounded theory.
Grounded theory
The programme of discovering grounded theory (Glaser & Strauss 1967, Strauss 1987) can be specified as a variety of an explanation-based type of theory:2 Grounded theory, we see from Table 2, is less heroic than high-level notions of theory such as the idealizing models of rational choice theory, and the search for general presuppositions among social philosophers. The notion of grounded theory was developed as a reflexion on the practice of participant observation in field work. In field work, the researcher explicitly participates in the object of study.3 That experience makes the researcher aware of the danger of projecting preconceived concepts on to specific social settings. A sensitivity to cases is required, and when explanation of cases is developed in a dialogue with data, sensitivity of cases implies both sensitivity to context and to the interaction orders that develop. Thus, the question of replicability is adressed as follows:
A grounded theory is reproducible in the limited sense that it is verifiable. One could take the propositions that are made explicit or left implicit (whatever the case may be) and test them. However, probably, no theory that deals with a social psychological phenomenon is actually reproducible insofar as finding new situations or other situations whose conditions exactly match those of the original study, though many major conditions may be similar. Unlike a physical phenomenon, it is very difficult to set up experimental or other designs in which one can recreate all of the original conditions and control all the extraneous variables that may impinge upon the phenomenon under investigation. When testing hypotheses derived from the propositions of a grounded theory, the investigator would have to specify the conditions under which
383
Lars Mjset
the hypothesis(es) was being tested and make adjustment in the theory to fit those conditions, (if they did not match those originally specifyied in the theory) (Corbin & Strauss 1990: 424).
Despite the quest for sensitising concepts (Blumer 1969), grounded theory aims to consolidate knowledge that reaches beyond the level of single cases. Theory is discovered in the process of explaining one or a few cases. The dialogue between evidence and ideas (Ragin 1994) may well start from an effort to explain singular cases like historians aim to explain one single event, a therapist relates to a clients case history, or a judge in a court case tries to understand the events leading to a particular act of crime. These explanations can be turned to contextual generalizations in specific fields of research through the selection of additional cases (theoretical sampling). This is why grounded theory can be seen as a notion of theory that is explanation-based. General is not regarded as the opposite of context-specific. New cases are added and studied systematically, but the value of the new cases is not simply to confirm or reject earlier explanations. These may be revised, but comparison may also show that there are distinct differences. The ability to explain single cases is a necessary stage on the way to establish bounded generalizations. Often, the researcher investigates whether findings related to a few cases hold for additional singular cases. If not, one may establish types that are different cases of the same, e.g. different types of welfare states. This implies limits to decontextualization. With systematic inclusion of new cases, bounded generalizations emerge, while context is not lost: typologies and periodizations are major means of combining generalization and context-sensitivity. Systematic specification of the context within which the theory applies is part of the discovery of theory. The scope of the regularity follows immediately. It is required that grounded theory at the same time specifies the conditions under which a phenomenon has been found in this particular data (Corbin & Strauss 1990, 424). Generalization does not take the form of subsumption, but is analogous to learning: a process of contrasting, discovering the extent to which new situations can be understood through earlier knowledge. Grounded theory generates knowledge at the middle or low level, which means that it applies within more or less broadly specified contexts. But it differs from the law-oriented notion of middle range theory by not striving to move towards the high level, it does not see the lawlike regularities with their very extensive scope as a goal at which social science should eventually arrive at in the future. It is a programme of gaining as general knowledge as necessary, given the research problem. If, for instance, the research problem is related to the Norwegian welfare state, the specificities of that welfare state, e.g.
384
in comparison with other Nordic and other European welfare states must be systematically included as context. Further links upwards to a more general theory of industrial society, life world/system or even structure/action is considered unnecessary, given the research problem at hand.4 Grounded theory may be called induction, but induction with specific reference to a research question. If events merit specific description, it is because they are highly significant in terms of the problems that we raise. Grounded theory describes in an analytic way, that is, in terms of the dimensions that must be considered in order to answer the research question (analytic induction). The micro/macro-dualism is less important: theory may be established at several levels. Unlike the variables-oriented techniques of statistical generalization (the law-oriented notion of theory), grounded theory does not aim to generalize to a broader population. The theory must be based on representative concepts. The aim, as Corbin & Strauss (1990: 421) write, is: ultimately to build a theoretical explanation by specifying phenomena, in terms of the conditions that give rise to them, how they are expressed through action/interaction, the consequences that result and the variation of these. Depending on the research question, one may aggregate or disaggregate in many steps. Grounded theory obviously differs from the idealizing notion of theory, where researchers deductions and simulations are thought experiments based on weakly grounded assumptions. Referring to the very lowest level, grounded theory is a more ambitious notion as far as our chances of accumulating empirical knowledge goes than the notion of theory implied by the deconstructionists. The deconstructionist conception takes sociology of knowledge skepticism derived from the uncompromising undermining of any claims to pre-empirical, transcendental knowlegde (the social-philosophical version of high level theory) to its extreme by doubting any knowledge beyond the level of specific cases (cf. Table 2). And even in these cases, systematic knowledge is seen as nothing different from everyday knowledge: it is (at least in Foucaults perspective) an expression of the desire for domination. Given the emphasis on contextualization, grounded theorists must also reflect on their own context in terms of the sociology of knowledge: social research in our part of the world today is institutionalized in research communities operating within research institutions and universities. Social science knowledge clearly participates in social development: a welfare state researcher cannot ignore the fact that the knowledge she commands may influence policy making. At the same time, this knowledge is also critically evaluated within the research community. But since theory is indexed to a specific context, the research frontier within which knowledge is accumulated is always a local one,
385
Lars Mjset
one in which the link back to high level theory is in practice non-existing. Only if a local research frontier does not exist for the question asked, the researcher has to start from scratch. But in most cases, a local research frontier will exist. At such a research frontier, both internal features of the relevant research collective and external features of the social context may influence the questions asked and thus also the knowledge which is accumulated. As for internal features, at least in our kinds of society, the semi-autonomous research collective should work as a guarantee against fads and fashions, simple cheating and overly politisized explanations. Glaser and Strauss (1967: Ch. IV; Strauss 1987: 241ff) distinguished between substantive and formal grounded theory. These may be seen as two different ways to generalize. A substantive theory is an explanation that at the present local research frontier is accepted by the research community as a valid account with reference to a specified context, such as e.g. the Nordic welfare states in the 1990s. A substantive grounded theory consists of grounded generalizations that traces similar patterns among at least a few, and mostly among more cases. It yields typologies, as outlined above (and cf. Rokkans theory, discussed below). It also must contain concepts that specify mechanisms at work. While typologization specifies context, mechanisms specify patterns of interaction between individual or collective actors. Mechanisms turn the various features of the context into explanatory factors. These explanatory mechanisms may be distinct to the substantive field studied, but there is a way to turn them into what Glaser and Strauss dubs formal grounded theory. A formal grounded theory is a module that can apply in various contexts (e.g. the strength of weak ties in network theory), but one that is not explanatory in and of itself. It only becomes explanatory when used in a context. Thus, formal grounded theory is general in a weaker sense than substantive grounded theory. It is general across contexts, but not explanatory independently of a specification in terms of a context defined by research questions asked within established local research frontiers. Formal grounded theories may be useful as components in such an explanation, but formal theory does not explain without context. Thus, Glaser and Strauss (1967: 79) noted that it was difficult to find a grounded formal theory that was not in some way stimulated by a substantive theory. Finally, what distinguishes grounded theory from critical theory? Both can be understood as expressions of a pragmatist or participatory attitude, which relies solely on the experience of doing social research. But while grounded theory may chose any kind of case as its explanatory starting pont, critical theory relates to a specific set of cases: contemporary cases of legitimate claims for social change. In this understanding, critical theory is closely connected to
386
social movements that raise such claims. This puts the social scientist in a crosspressure situation. While a researcher discovering grounded theory will normally remain within the relative autonomy of the social science research collective, a critical theorist may in the extreme case be drawn out of that community, becoming some kind of organic intellectual of the movement involved. There are also many examples of intermediate positions: Social science scholars working in NGOs, psychologists or social workers relating to individual clients, and so on. In fact, even scholars working in close relationship to the welfare state may become involved with weak groups which may very well push them to participate with reference to the ethical standards of critical theory, even though they maintain their institutionalized role as members of the research collective. We now turn to the objections against critical theory. We shall devote most attention to a very clear case of standard criticism, available in John Goldthorpes book of essays, On Sociology (quoted in the following by its year of publication: 2000). The book contains rich, nuanced discussions of a multitude of positions and debates in contemporary social science. But its conclusion is a highly restrictive understanding of sociology and a polarizing rejection of several currents whose work most other sociologists would consider perfectly sensitive empirical sociology. Common to many of these currents is that they pursue a programme of grounded theory.
Goldthorpes programme
Goldthorpes criticism of grounded theory comes as a side effect of his efforts to synthesize variables-oriented empirical studies and rational choice theory. The synthesis is legitimated with reference to the standard practical philosophy of social science. We thus have three elements: First, Goldthorpe invokes selected philosophy of science elements mostly drawn from early postwar philosophy of the natural sciences. We shall soon present the main features of what he calls the single logic of inference informing all scientific methods (2000: 67). One specific feature of this style of argument about the philosophy of the social sciences is the reintroduction of what the logical positivists rejected as metaphysics, the idea that science is not just empiricist, but also concerned with the search for fundamental relations between unobservable entities: scientific realism, what Hacking (1983) dubs realism of both theories and entities. Secondly, Goldthorpe wants to base sociology on the statistical analysis of large data-sets. Goldthorpes main empirical field has been the study of social mobility. In this research tradition, Blau & Duncans (1967) pioneering contriSOSIOLOGISK TIDSSKRIFT
387
Lars Mjset
bution incarnated the programme of moving from correlations to causes by means of sophisticated probabilistic models such as developed by Simon, Wold and Blalock. This view of causation as robust dependence (2000: Ch. 7) was one of the most ambitious expressions of the law-oriented notion of theory: from the middle level of large scale data-bases one could directly establish theories about social processes. But when the programme was taken up by an increasing number of scholars, it was discovered that it was too heroic (Abbott 2001, Ch. 3). Goldthorpe himself contributes a splendid chapter (2000, Ch. 7) demonstrating that statistical methods themselves can never yield explanatory theories about the causal processes the researchers were interested in. Rather than yielding theories, Goldthorpe concludes, the analysis of large scale data-sets provides sociologists with something to explain. The social regularities described should form the basic explananda of sociological analysis: sociological problems are ones that can all in one way or another be expressed in terms of social regularities their formation, continuity, interrelation, change, disruption, and so on (2000: 154). Third, Goldthorpe completes his synthesis by suggesting that sociology should rely on an idealizing notion of theory, as found in what he prefers to call rational action theory. He accepts the brunt of the Popperian criticism of the immunizing strategies of neoclassical economics (2000: 109), but still claims that a leaner version of such theory is what sociology needs. Discussing varieties of such theory (2000: 130, 136) he prefers an intermediate position as to whether strong or weak rationality requirements should be imposed; he locates himself close to the situational conception, doubting the procedural conception (Simons satisfycing); and finally, he prefers to see rational action theory as a theory relevant in specific fields rather than as a general theory. With reference to his three dimensional map (2000: 126, Fig. 6.1), he seems closest to Boudons cognitivist model. These three elements form a synthesis: A foundation in the analysis of largescale data-bases saves rational action theory from the charge of empirical irrelevance, while rational action theory supplies the theoretical aspects that variables-oriented studies proved unable to supply.
The single logic of scientific inference

The first element serves as Goldthorpes basic criterion that demarcates science. It is drawn from the American political science textbook on qualitative analysis by King, Keohane and Verba (quoted in the following by its year of publication: 1994), which aims to list the majority of the important rules of scientific inference (1994: 12, Ch. 1). Their 13 rules relate to four aspects of the research
388
design: research question, theory, data and use of data. Does these rules go smoothly together in a single logic? Two rules concern the research question, which should be important in the real world, and secondly, make a specific contribution to an identifiable scholarly literature by increasing our collective ability to construct verified scientific explanations of some aspect of the world (1994: 15). These statements are quite open, since scholarly literatures exist at many levels of the social sciences, ranging between a general theoretical literature on e.g. rational choice theory, and highly area-specific literatures on e.g. specific aspects of welfare states or on systems of military conscription. Next, there are four rules concerning the role of theory. The first three chose a theory that can be wrong, is falsifiable and as specific as possible applies in any kind of empirical social science. But the last one implies that when relating theory to data, the researchers may extend the theory, but to avoid ad hoc reasoning they should generally not narrow it down in scope (1994: 22), unless new data are collected. This rule may be reasonable in quantitative research based on large-scale data-sets, but less so for qualitative research where theory is developed in a back and forth dialogue with data. As a generalized rule, this extend-only rule reflects the standard attitute. Even in quantitative research, it deflects attention from strategies of clustering populations into types. It also easily leads researchers to discard as ad hoc lower level literatures, such as e.g. area studies or highly topical research areas. Five rules relate to improvement of data quality: some of them clearly apply to any kind of research, but the requirements of reliability, validity and replicability are specified in ways that easily apply to quantitative studies of large-scale data-sets. There is no discussion (except 1994: 26) as to whether they apply, or apply in the same way, to qualitative case-studies. Out of two final rules, one argues that in order to utilize better given datasets, one should analyse across as many as possible data-sets to avoid selection or omitted variables bias (1994: 28). Obviously, this is a rule that only applies to researchers who use large-scale data-sets as their main empirical input. The final rule is a more general one, that empirical social science must strive to explain as much as possible with as little as possible. This, they argue, gives the researcher leverage, which means that complicated effects are reduced to a single or a few causal variables (1994: 29). Again, this is a clearcut preference for striving towards the most general explanation, i.e. aiming to start at a high point in the explanatory hierarchy. This leverage-rule together with the extend only-rule serves to bias Goldthorpes single logic towards quantitative, variables-oriented procedures. These are seen as the clearest expression of the rules of science, an
389
Lars Mjset
unequivocable single logic: The very abstract, and even unrealistic, nature of statistical models is what makes the rules of inference shine through so clearly (1994: 6). Let us call this the variables master example. The application of the rules of the single logic will enable the researcher to enlighten at least those areas of nature which are organized in a fashion that is convenient for the researcher (see 1994: 29)! Here is an element of scientific realism: the convenient system of hopefully just a single causal variable or a few variables (1994: 29) is seen as a representation of real world processes.
Goldthorpes criticism of grounded theory

Goldthorpes criticism of grounded theory mainly relies on the first two elements of his programme. First, since the programme of grounded theory allows successive modifications of the hypotheses formulated at the start of the process of empirical research, Goldthorpe in line with the extend only rule attacks the extreme inductivism (2000: 77) and the adhoccery of grounded theory (2000: 89). Second, grounded theory does not rely on indicators. Instead (as in Blumers (1969) formulation about testing concepts), researchers engage in what Goldthorpe regards as no more than conceptualization (2000: 60), disguished as sensitivity to context. Goldthorpe claims that this makes convergent conceptualization impossible. As a result of these two preferences, grounded theory escapes the testing of theory. There is no clear distinction between context and regularities. Goldthorpe substantiates this criticism in a review of three overlapping literatures: historical sociology, comparative macrosociology, and ethnography (2000, Chs. 2, 3, 4). Historical sociology may be good at analysing long term processes of structural change, but its reliance on secondary sources opens several possibilities of bias. Comparative macro-sociology may be able to contextualize with reference to institutions and historical path-dependencies, but its emphasis on constellations of singular causal forces makes it unable to break with long outdated mechanical models of reasoning. Ethnography may closely analyse actual mechanisms of interaction, but it fails to provide acceptable knowledge about underlying generative processes, since it is unable to deal with variation within and across locales. Let us consider this criticism with reference to Goldthorpes specification of causal analysis as a three-phase sequence (2000: 151). Using King, et al.s terms, phase one is data/use of data, phase two is theory, and the third is testing theory.
390
First phase Establishing the phenomena

Although he accepts the variables master example, Goldthorpe makes clear that the statistical models are not a means of infering causation directly from data, their primary use should rather be seen as descriptive (2000: 12). What then distinguishes such description from the alledged extreme inductivism (2000: 77) of grounded theory? The advanced statistical techniques, Goldthorpe ensures us, allow sophisticated description (2000: 153). He quotes examples from the use of loglinear and sequential logit modelling. Clearly, experience from his own research plays an important role. Erikson/ Goldthorpe (1993) found that both rates and patterns of social fluidity (intergenerational class mobility considered net of all structural influences) have been stable, both across nations and historically within nations. Thus, Goldthorpe concludes:
Such a degree of invariance clearly underlines the need for general theory. For hypotheses on the causal processes capable of producing temporal constancy and cross-national commonality of the kind that our quantitative analysis revealed will have to be derived from a theory of considerable scope: that is, from a theory which is precisely not sensitive to context unlike the theories of national exceptionalism in regard to mobility that our results called into doubt but applicable to societal contexts widely separated over both time and space. (Goldthorpe 2000: 62).
Statistics is the most reliable and versatile means of demonstrating the existence and nature of social regularities, especially regularities that are not readily apparent to the lay members of a society, but only revealed through the analysis of data that have been collected extensively in time or space (2000: 152). These descriptions are sophisticated because they are not sensitive to context. Goldthorpe infers from the scope of the described regularity to the scope of the theory that must be considered to explain it. These descriptions dictate what theory should look like: it cannot be sensitive to context, it must be decontextualized. The grounded theories of ethnography, historical sociology and contemporary macro-sociology, then, represent extreme inductivism since they are sensitive to context, as shown above. Goldthorpe does not discuss whether more unique events require very contextual theories, he rather postulates an unbridgeable gap between history and theory (2000: Ch. 2). This can be reformulated as a gap between context and theory. If this gap is unbridgeable, the pragmatist programme of moving up from historical explanations of single cases to bounded generalizations is
391
Lars Mjset
doomed to fail. If social scientists want to establish general knowledge, they should avoid any procedure that involves sensitivity to context. The problem with this argument is not the wish to generalize, but the postulate of an unbridgeable theory/context-divide. Explicit arguments in favour of ths postulate is lacking. Perhaps Goldthorpes claim that sociology (or social science, since he obviously holds that all social scientists should commit themselves to his single logic) studies social regularities form such an argument? But that is a definition, not an argument, and as such it is both misleading and imprecise. It is misleading since social scientists do not primarily chose their explananda with reference to decontextualized regularities discovered in variables-oriented studies. Rather they mostly want to deal with important real world problems, some of which relate to regularities, while others relate to singulars. Social scientists are often interested in exeptions to regularities. How was it, for instance, that Sweden between 1930 and 1970 was actually able to narrow down the class differences in educational attainment (2000: 180)? If the research problem decides, the sophisticated description must be the kind required by the research question, it could be a sophisticated statistical description, but it could also be an advanced comparison that brings out the unique constellation of one case, or some combination of the two. Even Goldthorpe at times must imply that it is the social importance of the explananda that matters. Although he never explicitly mentions King et. al.s first rule of inference: that the research problem should be important in the real world, he does mention that a chief accomplishment of statistical sophisticated descriptions has been to identify and characterize important social regularities that were hitherto unappreciated (2000: 153). This can be read as a halfhearted admittance of the role played by research questions refering to real world problems. Furthermore, Goldthorpes restriction of sociology to social regularities is also imprecise. We have already mentioned that social regularities may be traced at many levels. The research question crucially influences the level at which we try to generalize: Certainly, there is much research in Sweden concerning regularities that apply in the Swedish context, and if a regularity is found to apply more broadly, the researchers must still discuss what specifically explains its persistence in Sweden. We find no arguments to defend Goldthorpes unbridgeable theory/contextdivide, and must thus reject his claim that variables-based descriptions are more sophisticated than case-based ones. Both are necessary and all descriptions are related to research problems. Let us connect back to the notion of the standard practical philosophy of
392
social science. We have already indicated that in King. et al. and Goldthorpes single logic we find the scientific habitus of moving upwards, to go for the very general relations that are applicable across time and place. A further characteristic of this attitude can be specified with reference to Hackings contrasting of representing and intervening. The mainstream postwar philosophy of the natural sciences regarded science as true representations of the real world (Hacking 1983). With such an attitude, it is irrelevant to reflect on the quest for scientific knowledge in terms of the sociology of scientific knowledge. In contrast, the pragmatist tradition rather discusses the realtionship between the research collective and the world it studies with reference to intervention what we have called participation. Thus, one can reflect on the emergence of research problems (the definition of explanada) in terms of the sociology of scientific knowledge. The pragmatist position also opens up to reflection on processes internal to the research collective. Such collectives are not primarily media through which the essential features of an external reality is represented, they are themselves societies, and this also influences the kind of knowledge they come up with. In his very useful discussion of statistical models, Goldthorpe (2000: Ch. 7) draws a lot on Abbotts seminal paper on the causal devolution, but while Abbott (2001: Ch. 3) in a pragmatist way discusses the rise and fall of causal models in terms of a sociology of professional knowledge, this plays no major role for Goldthorpe. Abbott adds to this perspective a rather internal analysis of how the family of linear statistical models actually work. In that respect, his conclusion is very different from Goldthorpes: he concludes that when these models describe, they decontextualize. Their popularity in the social science research collective around the 1970s led many sociologists to adopt the readymade ontology of a general linear reality (Abbott 2001: Ch. 1), which according to Abbott led sociology on to the wrong track. Only when the limits to these models had become obvious as they were tried out by an increasing number of scholars the pendulum swung back, creating a situation where Abbott (1999: 196) hopes that the pragmatist approach of the Chicago school that one cannot understand social life without understanding the arrangements of particular social actors in particular social times and places may again gain the respect it deserves. This insight was also buildt into Glaser and Strauss notion of grounded theory,.
393
Lars Mjset
Second phase hypothesizing generative processes

According to Goldthorpe, the descriptions provided by statistics are sophisticated because they invite a decontextualized view of theory. Exercising his polarizing distinction between theory and history, he sees a historical narrative as inherently specific in time and place, in contrast to the applicability across time and place that is required of theory (even if subject to some delimitation of scope). (1997: 128). Before we return to the statement in brackets, we shall consider how Goldthorpe hypothesizes generative processes without any grounding in data. He has been particularly concerned with the absence of change (...) in class differentials under conditions of educational expansion (2000, 171). In two chapters (2000; Ch. 8 and 9) he presents a rational choice model based on the distinction between a working class and a service class. He concludes that in industrial societies class differential spersist because little change occurs in the relativities of cost-benefit evaluations of educational options as these are rationally made in different class situations (2000: 180). There is no space here to discuss the details of this modelling effort, but its underlying idea, writes Goldthorpe, is that despite
non-rational influences, it is rational considerations that are, not the only, but the main common factor at work across indvidual instances, and that will therefore shape patterns of educational choices in aggregate and, in turn, the regularities that constitute our explananda (...). Our model then aims to represent these considerations in an idealized way, so as to capture the key generative processes involved, rather than to represent decision-making as it actually occurs at the level of particular families. (2000: 203; cf. 184).
Prior to testing it is assumed that rational considerations are the main common factor. The idealized representation is the thought experiment style of reasoning, typical of the idealizing notion of theory (Table 1). Goldthorpes strong claim is that this represents the key generative processes at work in the real world. In another essay, he concludes that when
RAT [rational action theory] is used to provide an explanation for probabilistic regularities revealed by QAD [quantitative analysis of large-scale data-sets], it is no longer necessary to suppose that all actors concerned at all times act in an entirely rational manner: only that the tendency to act rationally, in the circumstances that prevail, is the common factor at work, while deviations from rationality are brought about in a variety of ways and with a variety of consequences. (2000: 105)
394
Goldthorpe postulates some kind of homology or elective affinity between the generative processes derived deductively from above (second phase) and the sophisticated descriptions provided from below (first phase). The connection between the world of theory and the real world is bolstered by the concept of probability. Goldthorpe claims that RAT solves the micro/macro-problem: The model relates not to micro decision making at the level of the individual family. Rather, it explains macro patterns, patterns that in the real world emerge out of a stochastic chaos. Since the descriptive statistical models are probabilistic ones, Goldthorpe reasons that the regularities they yield will correspond to patterns derived under RAT assumptions. Since deduction from above and statistical inference from below converge in this way, Goldthorpe seems assured that he has gotten to unobservable key generative processes. In Hackings (1983: 27) terminology, this is a realism of both entities and theories. These are ambitious statements! We should note that Goldthorpe qualifies these with terms such as suppose and should in principle. Still, these are principles which he establishes not as a consequence of testing, but as his conclusion on the two phases of a sequence that precedes the third phase of testing. Goldthorpe states (2000: 152) that the phases are not that readily separable in any particular piece of sociological work, but it seems quite clear that he would not advice any scholar to enter into a dialogue of theory and data, as in grounded theory. In grounded theory, as described earlier, an unbridgeable divide between theory and context is never allowed to emerge, and there is no need for the pre-empirical reasoning about a match of thought experiments from above and data processed by probabilistic models from below. Grounded theory may imply a realism of entities,5 but not a realism of theories. There is no need for the idea that this represents crucial features of reality: rather theory is knowledge buildt from participation in the real world (the researcher has been living his analyses, testing them not only by observation and interview but also by daily living; Glaser & Strauss 1967: 225), which always can be fed back to and thus also challenged by that world. Substantive grounded theories, writes Glaser & Strauss (1967: 239f), will be understandable to laymen, to the people working in the substantive area. But it is also further discussed and evaluated in the research collective. This process of evaluation selects various substantive grounded theories for research frontiers and if resources are continuously allocated to these lines of study there will be growth of knowledge, albeit only at a local level, not at the highest levels. Goldthorpes core arguments are in terms of rational action theory and probabilistic statistical models. It is worth noting, however, that at some points in his essays, he retreats to a more moderate position, where he is willing to
395
Lars Mjset
accept whatever theory of action, as long as the aim is to get at generative mechanisms (2000: 154). Furthermore, he also signals that sophisticated descriptions may follow not just from probabilistic techniques, but also from various non-probabilistic techniques of scaling, clustering, and sequencing that are even more clearly dedicated to descriptive tasks (2000: 153). This allowance for other action theories and non-probabilistic models can only weaken the chain of arguments just recounted. At times, Goldthorpe presents his thin version of rational action theory in terms that seem to demand quite an effort at contextualization. If sociologists wish to understand subjective rationality, the modelling of actors information environment must be of primary concern. Unlike in economics, one must treat the information available to actors as a product of the social relations in which they are involved, and social scientists must recognize that the social structuring of knowledge is a no less important aspect of that situation than the structuring of other resources of, say, a material kind (2000: 132). From a pragmatist vantage point, which is in this respect more skepticist than the standard perspective, one would wonder whether such an analysis of the situation of action could ever yield representations of key generative processes.
The third phase testing

First: not sensitive to context. Second: key generative processes. Obviously, Goldthorpe would prefer more than just a pre-empirical claim that the theory really gets at the key processes! The empirical relevance of the theory must be checked. The term for this with associations both to statistics and natural science experiments is testing. The craftwork of ethnographers is to study the situation of action. Thus, in spite of his many critical points, Goldthorpe proves willing to grant sociological ethnography the role as an essential complement to survey research, since it provides intensive, context-embedded studies (2000: 87). This is slightly ironic! When grounded theory was developed as a codification of ethnographic research, it emphasized the logic of discovering theory, against the standard preoccupation with testing. Now Goldthorpe suggests that ethnography should complement survey research in the testing phase, since the hypotheses established should in some way or another be found present in specific locales (2000: 87). But remember that Goldthorpe is out to test hypotheses about key processes that are unobservable. Thus, he is quick to state that if ethnography is to assist survey research in this way, two requirements must be fulfilled:
396
First, ethnographers should be obliged to capture the central tendencies in the locale in question and check whether these corroborate or falsify the hypothesized regularities, which themselves are, as Goldthorpe emphasizes: probabilistic! The variables master example is clearly at work: He notes that etnographers would be helped by the sampling logic of survey research. If not, he claims, they will end up in purely mechanical explanatory models. We shall return to this. Second, ethnographers should also assess variation across locales, providing a basis for generalization across particular ethnographies. A similar outcome, Goldthorpe notes, can be produced by different causal conjunctions. Specifying this point, he hits at grounded theory:
The empirical testing of hypothesized generative processes, whether via ethnography or otherwise, should then be such that this issue of causal heterogeneity can be seriously adressed, rather than being invoked, as it often tends to be in grounded theorizing, merely as an excuse for adhoccery (...). (2000: 89).
One should get to distinct theories that go beyond what Elster calls mere causal models. At this point, Goldthorpe requires proper attention to context!
Testing whether possible causal processes are actually realized in locales that can be reliably positioned within a range of variation is of obvious relevance to what must be the further ambition of explicitly spelling out domains of application: that is, of developing theories that comprise coherent accounts of which of the causal processes that they specify will in fact be most likely to operate in which kinds of context. (2000: 89).
So with respect to testing, context obviously becomes crucial. The scholar who required theory to be not sensitive to context now demands precise attention to context. This is repeated as he embraces the Popperian criticism of economic modelling
if a theory is formulated in such a way that its domain of application remains unspecified, it becomes immune to falsification through empirical findings: every apparent refutation can be countered by the argument that the theory was being applied to the wrong kind of case, and the theory may then be returned, unscathed, to the depository. (2000: 109).
The dilemma was already contained in Goldthorpes brackets, quoted above, theory needs some delimitation of scope. But the standard style of scientific
397
Lars Mjset
argument (Hacking 2002) implies that contextualization is not an art he needs to care about, at least not when developing theory. What, one must ask, is the context he establishes for his own explanation of the constant differential of educational attainment? Strikingly, he fares no better than the social philosophers he is so critical of. He employs vague terms such as advanced industrialism, all modern societies, capitalist societies/industrialism, a class society, advanced industrialism, economically advanced societies, the era of expanding educational provision (2000: 97, 174, 162, 153, 197, 178, 180, 182). Establishing his theory/context-divide, Goldthorpe emphasized that he had nothing against history as an explanatory factor (explanans). His point was that sociologists should refrain from taking singular historical events as their explanandum. Goldthorpe notes that in his own work, he has often recurred to history as the theoretically unexplained explainer (1997: 29). Still, his synthesis between probabilistic statistical methods and rational action thought experiments allows no serious attention to how context and theory can be developed in interaction, so that theory comes out indexed to specified contexts. That is the programme of grounded theory, which, as we have seen, Goldthorpe writes off as ad hoc and reconceptualization. But the fact that Goldthorpe must invoke context when he gets to the testing phase, shows that grounded theory cannot be written off that easily. Testing requires a much more thorough specification of context than just modernity, industrialism or the like: theories can only be considered as competing if they relate to the same specified context. As for testing of the theory on constant differences in educational attainment, Goldthorpe discusses what kind of survey material he would need to test it. He concludes, however, that such a survey would be difficult to conduct, since it would require a direct study of educational decisions, while social sicence has so far mainly developed the tools to study the outcomes of such decisions (2000: 178). But there must be more than just technical reasons why Goldthorpe, after many years of research into social mobility, admits that he has only made a very moderate beginning at developing theories not sensitive to context (2000: 62). In another essay, he states that in sociology, accounts of processes capable of producing observed regularities are not yet for the most part expressed in sufficiently specific and theoretically informed ways to permit substantive models to be developed (2000: 158). He puts his faith in more formalization and in the simulation approach to hypothesis testing. The last thing he would admit is that a programme of discovering grounded theory would be of any help at all. In a footnote, he remarks that while it is surely
398
disappointing, it should not be thought a disgrace for sociologists to admit that they have not been able to develop a theory that will adequately account for their empirical findings (Goldthorpe 2000: 62). Rather than stating why this is so, he turns directly into a rather unreasonable attack on grounded theory. At this point, where a strategy of grounded theory most clearly emerges as an alternative to his theory/context-divide, Goldthorpe can only reply with a counterattack that sinks below his usual level of sophistication. As for his more sophisticated objections to qualitative case-oriented research, we have already noticed his claim that they yield only mechanical explanations (particularly held against Ragins work), which he regards as quite outdated given the probabilistic revolution! What is really at issue here is his implied realism of entities and theories. As boldly as scholars once believed in the mechanical worldview, Goldthorpe states that reality is probabilistic and that RAT can represent the essence of real world processes. But grounded theory roots in the pragmatist skepticism against such heroism: knowledge is empirical, it is systematic, but the statement that it is converging towards a true representation of reality adds nothing (Hacking 1999). Briefly, a main reason why explanations by causal constellations are not mechanistic ones, is the following: The main feature of social science knowledge is that it participates, even though it is institutionalized in relatively autonomous research collectives. It is publicly discussed in the research collective, it is drawn into policymaking, into the strategies of collective actors, etc. Thus these are not mechanistic explanations. But the participation only exceptionally takes the form of an experiment. When theories are related to future cases, they do not participate as theories that must be either corroborated or falsified. The relation to such new cases is not simply to subsume them, but to evalute them comparatively. A suitable illustration is the following: Take unique events such as famines or ethnic cleansing. Goldthorpe would probably argue as he argued with reference to historical sociological studies of revolutions: these are singular events and one should not venture to establish sociological theories about them (2000: 63). But we must consider the researcher as a potential participant, not as a pure observer that just waits for a new revolution, famine or massacre to happen! The researcher may hold a substantive grounded theory, the best knowledge presently accepted at this local research frontier. The theory includes types of such events and mechanisms that have been at work in earlier cases. A future instance of the same is not thought of as a mechanical repetition. And if we possess expert knowledge, we do more than predict, we try to influence, for instance we adress those actors (collective or other) that may be able to prevent a famine. We do not want a famine to happen in order to further corroborate or falsify our theory.
399
Lars Mjset
Rather: what we do is to make an educated guess analysing the context as meticulously as possible, given uncertainty and restraints on information on the extent to which history may be likely to repeat itself! We analyse a new case in comparison with the knowledge accumulated through the analysis of other cases of the same. Social researchers have to manouver vis a vis all these demands, their research questions largely flow from the processes of social change. Grounded theory does not claim to represent a mechanical world: its philosophical realism refers to entities, and entities are responsible actors, individuals or collectives. It is primarily the world that develops the social theories develop with the world! Testing his model to explain constant class differentials in educational attainment, Goldthorpe is well aware that Sweden between 1930 and 1970 is an outlier. He has to admit that such constant class differentials are no automatic consequence of industrial captalism. Defending his theoretical model, he recurs to a rather simplistic philosophy of history, claiming convergence in the long run: the egalitarian tendencies apparent in Sweden (...) will in the end have to be seen as quite specific to and thus limited by the period of a distinctive political conjuncture (...) that has perhaps now reached its close. (2000: 180). Goldthorpe has never been tempted to relate to various subgroups within his large population of industrial capitalist countries, and he does not entertain the idea of systematically developing the periodization between early and late postwar periods for various such subgroups. A viable alternative to Goldthorpes procedure we find in Stein Rokkans contribution to political sociology, which can clearly be presented as grounded theory (Mjset 2000). Here the population of Western European political regimes is stratified into typologies, and there is a historical sensitivity to institutions and institutional complementarities. Rokkans many typological maps reflect his efforts to generalize without loosing context, whereby he was able to proceed quite far in terms of explaining European varities of political modernization. Unlike the social philosophers, who take the existential situation of modernity as their starting point, Rokkan decomposes varieties of modern political transformations. Rokkan used both qualitative and quantitative sources, but avoided Goldthorpes strange implication that certain variables-oriented methods should give more sophisticated descriptions than other (qualitative or quantitative) methods. He gives about the same weight to case-specific monographs as to survey based evidence. Rokkan was not content with postulates about national exceptionalism. His point was rather that while all units may be seen as exceptional, the goal must be to find the dimensions defining all these exceptional cases. He was seeking patterned variety (Ragin 2000). Statistically induced regu400
larities had no priviledged status in terms of influencing the research problem. In fact, the research problem was given more or less before attempts were made to scan the data for general patterns: Rokkan wanted to map and explain the diversity and thus to explain the various 20th century Western European trajectories of state formation and political mobilization through a systematic comparison across the relevant empirical dimensions. Rokkans work shows that contrary to Goldthorpes conclusion, we find much accumulated knowledge in grounded theories discovered within specific fields of study. Similar examples of viable alternatives would be the work of EspingAndersen and Korpi on welfare states and comparative political economy, both of which are best accounted for by a pragmatist attitude leading to explanationbased theory.
A social-philosophical, critical theory-based criticism of grounded theory

Michael Burawoy has published programmatic accounts of the extended case method. Like Goldthorpes essays, Burawoys discussions provide rich and deep accounts of contemporary approaches within sociology and their historical roots. Burawoys discussions are focused on the qualitative, field work-based traditions. Goldthorpes criticism of grounded theory is basically that it does not live up to the ideals of the standard attitude. Burawoys criticism is the exact opposite: he rejects grounded theory as positivist and non-reflexive. His criticism combines elements of the social-philosophical and the pragmatist attitude. Critisizing these three social science literatures that are all based on decent craftwork (ethnography, historical, sociology, comparative marcro-sociology), Goldthorpe sees these as no better than social philosophy and post-structuralism. Thus he collapses what we above defined as explanation-based and critical theory with the transcendental and deconstructionist notions of theory (cf. Table 1). From the opposite, anti-positivist side, Burawoy commits the same error, he regards social-philosophical and even sometimes deconstructionist theoretical approaches (2000: 344) as quite in line with his critical theory position. He differs from Goldthorpe only when he tries to establish a polarization between grounded and critical theory. In the following, we shall show that the threefold distinction between researcher attitudes (Table 1) and the treatment of grounded and critical theory as two varieties of the pragmatist attitude lead us to doubt Burawoys criticism of grounded theory. The social philosophical element is the idea that fundamental social science theory must be based on a set of pre-empirical (transcendental) notions about
401
Lars Mjset
action, structure and knowledge. Most social philosophers (take Habermas, Giddens or Luhmann) relate their set of basic notions to the broad historical context of modernity, with particular attention to how this context forms patterns of knowledge and perception. Burawoy pursues a Marxist interpretation in which the underlying social structures (the anatomy of capitalism), relations of production, is what determines knowledge. This critical realism functions in the same way as Goldthorpes scientific realism. It is a realism of entities and theories (Hacking 1983). The main difference is that in Goldthorpes work, rational action theory provides the key generative processes, while in Burawoy, such processes are derived from the Marxist theory of the capitalist mode of production. Furthermore, in Burawoys view, it is obviously not variables-oriented statistics that gives priviliged empirical access to these processes. We shall discuss below what kind of empirical work he promotes. The pragmatist element stems from Burawoys preference for ethnographic and anthropological methods of participant observation, which he interprets in line with the critical theory notion in Table 1 above. Burawoy (1991, Ch. 13) discusses four approaches to the development of theory out of data collected through participant observation. He classifies them by means of the two major challenges that are always pointed out in standard criticisms of case-studies: First, the problem of the significance of a study of singular cases, and second, the question of which level of analysis to start from, macro or micro. This yields the following simple scheme:
Table 3. Four approaches to case studies

Particular Micro Macro Ethnomethodology Extended case method General Grounded theory Interpretive case method
In a later paper (1998), Burawoy distinguishes two models of science. The positive one is what we have termed the standard attitude, while the reflexive one is his preferred synthesis of the social philosophical and pragmatist attitudes. This enables him to suggest another classification of grounded theory, with reference to two different methods of empirical research: interview versus participant observation:
402
Table 4. Empirical research in two models of science

Positive model of science Interview Participant observation Survey research Grounded theory Reflexive model of science Clinical research Extended case method
Although, as table shows, grounded theory joins the extended case method in being able to analyse micro/macro-connections, it is critisized as a standard approach. Burawoy claims that grounded theory strives toward greater generality: that is, the inclusion of more phenomena under a single covering law (1991: 8). It is an approach to the analysis of material from participant observation, but it is not reflexive. Let us challenge that claim. Burawoy specifies two objections (see 1991: 303, fn 5). First, he claims that grounded theory, with its preference for discovery of new theory, does not reconstruct theory. Second, he claims that grounded theory cannot reach out, its case-studies are unextended. Both objections must be rejected. Concerning the reconstruction of theory, it is striking that despite his critical stance towards the positive model of science, Burawoy explicitly states that his notion of theory fits with the Popper/Lakatos-tradition. But his emphasis is peculiar: it is on the act of falsification, whereby the theory that was the starting point is reconstructed to account for a particular case that proved puzzling in the light of the original theory.6 This attention to the particularities of cases, however, is not in principle different from the discovery of grounded theory. Burawoy himself (1991: 303, fn) is well aware that Strauss (1987: 13 f, 306 ff) did not imply that theory should be discovered from scratch in every new field work, but required that the earlier theory one would relate to should in itself be grounded. The problem is that Burawoy has a very undifferentiated notion of theory: When he talks about reconstructing theory, theory sometimes means quite specific theories linked to specific local research frontiers (some would label them grounded, others would call them middle range). At other times he implies that theory consists of very broad accounts of modernity, globalization, system/life world. According to our taxonomy (Table1), these would be transcendental social-philosophical theories. As long as the theories Burawoy are reconstructing are low or middle level kind of theories, the difference from grounded theory is hard to detect.7 But if he is thinking of high-level socialphilosophical theories (Habermas system/lifeworld-interpretation of the present is an example he refers to), the problem is that such theories are highly decontextualized, or at least only related to very broad contexts. The link to particular cases becomes very loose. Note that this whole discussion has not made any reference to the extended
403
Lars Mjset
case method, since Burawoy emphasizes that his approach to theory reconstruction does not require extending out, one can confine oneself to the structure and dynamics of the micro situation (1991: 303, n. 5). Concerning the second objection, the extended case method (originating in the Manchester school of anthropology) is a more limited approach, which sees a particular phenomenon as the product of historically specific causes. Burawoy contrasts to grounded theory, which in his view systematically removes particularity (1991: 303, n. 5). This claim is similar to the one quoted above, that grounded theory strives towards covering laws through inductive procedures. This is simply wrong! Burawoy himself notes that there is a distinction between substantive and formal grounded theory, as outlined above. If this distinction is taken seriously, the real difference between grounded theory and the extended case method should be the interpretation of historically specific causes, where substantive grounded theory might yield only a very local causal constellation, whereas the extended case method extends out to the determining macro context. But that would seem to restrict the extended case method to a quite limited set of research questions. Is the macrocontext always the determining one? Is that not a question of the research question posed? Are there not many good pieces of social research that does not relate causal chains to a determining macro context? Furthermore, we have already seen that Burawoy allows us to reconstruct theory with reference solely to the micro-situation. Like in the case of Goldthorpe, the pragmatist link between research question and level of analysis leads to a more realistic assessment of what social scientists actually do. The dilemma that critical social science becomes restricted to a very limited range of reseach problems is not new. That same dilemma runs through the whole of C. Wright Mills argument in his classic The Sociological Imagination (1959). Most of the elements of Burawoys criticism of grounded theory are already contained in Mills book: one only needs to replace the sociological imagination with the extended case method and liberal practicality with grounded theory. By the term liberal practicality Mills refered to the long tradition of social problems studies. Such research, he wrote (1959: 144), has been clogged by undigested facts; and this has led to the idea of a romantic pluralism of causes. At any rate, the values, whether espoused or not, that have been assumed by liberally practical social scientists have now been largely incorporated into the administrative liberalism of the welfare state. Following Mills, quite a few radical scholars today discard grounded theory as kind of an administrative positivism. Mills himself, as the quote shows, held that it was outdated, since it was about to be replaced by abstracted empiricism, his term for Goldthorpes variables-based sophisticated descriptions.
404
Our objections to Mills are the same as to Burawoy, so we return his reasoning: Consider the case of an extended case study that reconstructs social-philosophical theory. The peculiarities of that case would then be related to a determining macrocontext conceived in terms of grand, despecified periodizations such as e.g. globalization, modernity, capitalism or post-modernity. Again, there is a parallell to C. Wright Mills, who (in 1959: 199, 202) was one of the first to search for the peculiarities of the post-modern period. Now, given the social-philosophical roots of such notions, they must be seen as ungrounded. Burawoys problem is strikingly similar to Goldthorpes: striving to base social science in a decontextualized high level kind of theory, they both end up in too crude philosophy of history kinds of periodization! Both would benefit from more concern with the varieties of capitalism. When Glaser and Strauss launched the notion of grounded theory, their programme was never restricted to the micro level. Grounded theory was possible at any level, and in their review of a string of older and more recent research projects (Glaser/Strauss 1967: Ch. VI), they assessed several macrostudies. The best works of comparative historical sociology shows how one can pursue grounded strategies at the macro level. A focus on the macro-context is in no way in opposition to doing grounded theory. In fact quite the opposite: In order to establish a specification of the determining macrocontext that is relevant to the specific research questions asked, Burawoy needs a grounded theory at the macro level!8 Also in this case, grounded theory can solve some of the problems of its critics.
Some important implications

Goldthorpe is very eager to commit researchers to one logic of inference, but our discussion shows that there are at least three logics. One is surely the logic of deduction and testing. But such a logic works only within a context. If we want to conduct empirical social science, we cannot rely on the rational choice solution of establishing context in the form of armchair assumptions. The deductive logic is often compressed or substituted by mechanisms (or formal grounded theory). For these two reasons, we must also be aware of two other logics: The second is the logic of context, a logic of types and comparisons. How thoroughly we need to engage in periodization and typologization, depends on the research question. Mechanisms is the third logic: here very complicated chains of action, interaction, including complicated aspects such as feelings, limits on knowledge, etc. are compressed into stylized processes of interaction. We have seen that when he got to the phase of testing, Goldthorpe had to
405
Lars Mjset
require specification of the scope within which the formal patterns emerge as lawlike. But only substantive grounded theory gives meaning to a notion of a scope or context within which the theory applies. Importantly, this is not a question of testing, since it is only within a set of scope conditions that one can actually test competing (explanatory) theories. Neither can scope conditions be deduced from more general theories, they must be established pragmatically, with reference to the participation of social science knowledge-production in society. Scope must be a function of what we want to know that is our knowledge interests. We realise the difference from the law-oriented notion of theory and the standard attitude towards generalization: we rather have explanation by singular causal factors, they establish the context within which we may enrich the analysis by our knowledge of formal grounded theories (mechanisms). Pragmatist philosophy always opposed dualisms, and throughout our discussion we have questioned the generalizing/specifying dualism. Generalization is not the opposite of explaining particulars (as both Goldthorpe and Burawoy are inclined to think), it is a careful accumulation of knowledge from explained cases chosen according to a sampling procedure reflecting the logic of context. That is not a logic of testing, but of creating the necessary context given the research problem asked. The understanding/explanation-dichotomy is here irrelevant. The analysis of causal constellations are at the core of the more pragmatic sciences such as psychology (as therapy), law and history. Formal grounded theory is the kind of grounded theory that is most similar to transcendental theory of action and the kind of generalized theory required by scholars with a standard attitude! But the pragmatist position is that none of these are explanatory outside of a context! Thus the pragmatist notion of general theory is not linked to formal theory, but to substantive theory. This implies that general theory is not converging. There may be several general theories, uniting various local research frontiers. It follows that formal theory can not be regarded as the most general theory, contrary to what both Goldthorpe and Burawoy seem to assume. Rather, general explanatory theory must be general summaries of contextual factors that may be of value to several local research frontiers. Good typologies can in this respect be seen as the most general knowledge we can gain, relevant to a number of local research frontiers! Such general theory is not necessarily high level theory! It is general in terms of providing context for a large number of specialized studies (local research frontiers), but not for all. There may be several, non-convergent general substantive theories. Still, they are more general than the multitude of hypenathed sociologies.
406
Rokkans maps and Korpi/Esping-Andersens families of welfare states are exactly general in this sense, they aid the researchers in the complicated task of establishing context, whether the research is on industrial relations, on welfare states, on political development, of ethnic mobilization, etc. If social theory is about interpreting the present, then these local research frontiers are indispensable. As we have seen, there are even reasons to regard them as among the most general theories that are of relevance to anybody who try to interpret the present within which they are themselves situated. This implication is probably most relevant for studies at the macro-level. Here most of the generalizations will be substantive ones. At the micro-level, it would seem, the notion of formal grounded theory will prove to be relatively more important. Grounded theory solves some of the problems of its critics. Certainly, it does not solve all problems of empirical social science. But it deserves to be treated as an approach that survives a systematic assessment in line with contemporary philosophy and sociology of scientific knowledge.
Noter
1 A first draft of this essay was presented at the 37th World Congress of the International Institute of Sociology, Stockholm July 59, 2005. 2 We use a terminology that is specified in Mjset, forthcoming. For simplicity, we shall in the following use the term grounded theory without specifying the difference to other varieties of explanation-based theory (e.g. network theory, ethomethodology, etc.) 3 In historical sociology, this participation is obviously indirect, cf. Mjset 2003. 4 One may even at the philosophical level involve a skepticism which argues that theory at such a level does not really aid the accumulation of knowledge at all. Cf. the discussion of elevator words in Hacking 1999: 2224. 5 Cf. Hacking 1999 which develops his entities realism (Hacking 1983) into a discussion of the classifications of social science as interactive kinds. 6 For a discussion of the generalizing/specifyingambivalence in Poppers understanding of social science, see Mjset 2001: 15642. 7 Burawoy might argue that the difference lies in the partipatory relationship to claims for legitimate rights, as in our definition of critical theory. That would fit with the emphasis on the peculiarities of single cases. He notes (Burawoy 1988: 25) that grounded theory brackets involvement as bias, and that it accepts the role of social scientists as outsiders. But these features concern the relationship of participation between the social science research collective and the collectives they relate to in their research. In our taxonomy (above), this is what defines the difference between grounded theory and critical theory, namely that grounded theory involves participation that does not affect the relative autonomy of the research collective, while the way critical theory participates, actually makes that relative autonomy problematic. 8 This point is parallell to our counterposing (above) of Rokkans grounded approach to Goldthorpes ungrounded approach!
407
Lars Mjset
References
Abbott, Andrew (1999): Department and Discipline. Chicago: University of Chicago Press. Abbott, Andrew (2001): Time Matters. Chicago: University of Chicago Press. Blau, P. M. & O. D. Duncan (1967): The American Occupational Structure. New York: Free Press. Blumer, Herbert (1969): Symbolic Interactionism. Englewood Cliffs: Prentice Hall. Burawoy, Michael (1991): Ethnography Unbound. Berkeley: University of California Press. Burawoy, Michael (1988): The Extended Case Method. Sociological Theory,16:1, 433. Burawoy, Michael (2000): Global Ethnography. Berkeley: University of California Press. Corbin, Juliet & Anselm Strauss (1990): Grounded theory research: procedures, canons and evaluative criteria. Zeitschrift fr Soziologie (19), 6: 418427. Erikson, Robert & John H. Goldthorpe (1993): The Contant Flux. A Study of Class Mobility in Industrial Societies. Oxford: Oxford U.P. Glaser, Barney G. & Anselm L. Strauss (1967): The Discovery of Grounded Theory. New York: Aldine de Gruyter. Goldthorpe, John H (1997): A Response to the Commentaries, Comparative Social Research (16): 121 132. Goldthorpe, John H. (2000): On Sociology. Oxford: Oxford U. P. Hacking, Ian (1983): Representing and Intervening. Cambridge: Cambridge U.P. Hacking, Ian (1999): The Social Construction of What? Cambridge, Mass.: Harvard U.P. Hacking, Ian (2002): Style for Historians and Philosophers, in Hacking, Historical Ontology, Cambridge, Mass.: Harvard U.P. King, Gary, Robert O. Keohane and Sidney Verba (1994): Designing Social Inquiry. Princeton: Princeton U. P. Mills, C. Wright (1959): The Sociological Imagination. Harmondsworth: Penguin. Mjset, Lars (2000): Stein Rokkans Thick Comparisons. Acta Sociologica (43), 4: 381398. Mjset, Lars (2001): Theory: Conceptions in the Social Sciences, in N. J. Smelser & P. B. Bates, ed.: The International Encyclopedia of the Social and Behavioral Sciences. Amsterdam: Elsevier. Mjset, Lars (2003): Versuch ber de Grundlagen der vergleichenden historischen Sozialwissenschaft, in H. Kaelble & J. Schriewer, ed.: Vergleich und Transfer Zum Stand der Komparatistik in den Geistes- und Sozialwissenschaften. Frankfurt & New York: Campus Verlag. Mjset, Lars (2006): Six notions of theory in the social sciences. Forthcoming. Ragin, Charles C. (1994): Constructing Social Research, Thousand Oaks: Pine Forge Press. Ragin, Charles C. (2000): Fuzzy-set Social Science, Chicago: University of Chicago Press. Strauss, Anselm L. (1987): Qualitative Analysis For Social Scientists: Cambridge: Cambridge U. P.
408

Goldthorpe+Mjøset 2005 CanGroundedTheorySolveTheProblemsOf ItsCritics

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Goldthorpe+Mjøset 2005 CanGroundedTheorySolveTheProblemsOf ItsCritics

Uploaded by

Copyright:

Available Formats

Sosiologisk Can Grounded 13, 379408 Universitetsforlaget 2005 tidsskrift: VOL Theory Solve the Problems of its Critics?

CAN GROUNDED THEORY SOLVE THE PROBLEMS OF ITS CRITICS?1

A typology of notions of theory

Can Grounded Theory Solve the Problems of its Critics?

Table 1. Six notions of theory

Social science Explanation- based Critical

Table 2. Notions of theory and levels of analysis

Can Grounded Theory Solve the Problems of its Critics?

Can Grounded Theory Solve the Problems of its Critics?

Can Grounded Theory Solve the Problems of its Critics?

The single logic of scientific inference

Can Grounded Theory Solve the Problems of its Critics?

Goldthorpes criticism of grounded theory

Can Grounded Theory Solve the Problems of its Critics?

First phase Establishing the phenomena

Can Grounded Theory Solve the Problems of its Critics?

Second phase hypothesizing generative processes

Can Grounded Theory Solve the Problems of its Critics?

The third phase testing

Can Grounded Theory Solve the Problems of its Critics?

Can Grounded Theory Solve the Problems of its Critics?

Can Grounded Theory Solve the Problems of its Critics?

A social-philosophical, critical theory-based criticism of grounded theory

Table 3. Four approaches to case studies

Can Grounded Theory Solve the Problems of its Critics?

Table 4. Empirical research in two models of science

Can Grounded Theory Solve the Problems of its Critics?

Some important implications

Can Grounded Theory Solve the Problems of its Critics?

You might also like