PART   TWO           STARTING    OUT
course of a piece of research. Indeed, one of the strengths of qualitative research design is that it often allows for far greater (theoretically informed) flexibility than in most quantitative research designs, faMason puts it:
Theoretical or purposive sampling is a set oí procedures where the researcher manipulates their analysis, theory, and sampling activities mtfraclively during the research process, to a much greater extent than in statistical sampling. (1996: 100)
Such flexibility may be appropriate in the following cases:
•    As new factors emerge you may want to increase your sample in order to say more about them.
•    You may want to focus on a small part of your sample in early stages, y    using the wider sample for later tests of emerging generalizations.
-»   Unexpected generalizations in the course of data analysis lead you to seek out new deviant cases.
Alasuutari has described this process through using the analogy or an hourglass:
a narrow case-analysis is broadened ... tluough the search (or contrary and parallel cases, into an example of a broader entity. Thus the research process advances, in its final stages, towards a discussion of broader entities. We end up on die bottom of the hourglass. (1995: 156)
Alasuutari (1995: 155) illustrates this hourglass metaphor through his own study of the social consequences of Finnish urbanization in the 1970s. He chose local pubs as a site to observe these effects and eventually focused upon male'regulars'. This led to a second čtudy even more narrowly focused on a group where drinking was heavier and where many of the men were divorced. As he puts it: 'Ethnographic research of this kind is not so much generalization as extrapolation ... the results are related to broader entities' (1995:155).
GENERAUZABILITY AS PRESENT IN A SINGLE CASE
The fourth and final way of thinking about how we generalize in qualitative research is far more radical than our earlier alternatives. According to this approach, since the basic structu res of social order are to be found anywhere, it does not matter where we begin our research. Look at any case and you will find (he same order.
For this linguistically inspired approach, the possibility something exists is enough. As Periikyla suggests:
Social practices that are possible, i.e., possibilities of language use, are the central objects of all conversation analytical eise studies on interaction in particular institutional
loa
1,1                 i.i.i.l.l                 II
SELECTING   A   CASE
settings. The possibility of vanous practices can be considered generalizable even if the practices are not actualized in similar ways across different settings. (1997: 215)
Periikyla illustrates his argument by the example of his own study of AIDS counselling in a London teaching hospital (Periikyla, 1995). This study focused on specific questioning practices used by counsellors and their clients. As he puts it:
As possibilities, the practices that I analyzed are very likely to be generalizable. There is no reason to think that they could not be made possible by any competent inember of (at least any Western) society. In this sense, this study produced generalizable results. The results were not generalizable as descriptions of what other <ounsellors or other professionals do with their clients; but they were generalizable as descriptions of what any counsellor or other professional, with his or her clients, can do, given that he or she lias the same array of interactional competencies as the participants of the AIDS counselling sessions have. (1997:215-16)
As the most cogent proponent of this view once put it: 'tap into whomsoever, wheresoever and we get much the same things' (Sacks, 1984: 22). Sacks had a strategy of working with any data that crossed his path. This clearly conflicts both with the standard approach of quantitative social scientists, who usually work with random samples from particular populations, and with the common defensiveness of their qualitative brethren about the representativeness of the cases that they study.
Sacks's lack of defensiveness on this issue stems from his argument about the obvious pervasiveness of the social forms (or what he calls the 'machinery') with which he is concerned. For example, Sacks notes the ability of a child to learn a culture from very limited contacts and of the sociolinguist Whorf to build a Navajo grammar from talking to just one person (1992, Vol. 1: 485).
The pervasiveness of structures which these examples suggest implies to Sacks that it does not matter what data you select. AS he argues:
Mow if one figures that thafs the way things are... then it really wouldn't matter very much what it is you look at - if you look at it carefully enough. And you may well find that you [have] got an enormous generalizability because things are so arranged that you could get them; given that for a [societal] member encountering a very limited environment, he has to be able to do that, and things are so arranged as to permit him to. (1992, Vol. 1: 4S5)
However, apprentice researchers have to be very cautious about simply parroting Sacks's 'solution' to the problem of the generalizability of research findings. This solution is really only appropriate to the most basic research on social order guided by theoretically sophisticated positions like Sacks's own conversation analytic (CA) approach (or, perhaps, French structuralism). If you are interested in this sort of research, you should now attempt Exercise 8.2.
109
PART   TWO           STARTING   OUT
Within CA, following Sacks: i
the baseline assumption is that the results ate or should be ganerallzable to the whole domain of ordinary conversations, and to a certain extent even across linguistic and cultural boundaries. (Perakyla, 1995: 214)
However, Perakyla notes that even this depends on (he type of CA research:
Even though the most primordial conversational practices and structures —such as hirn-taking or adjacency pairs - are almost universal, there are others, such as openings of telephone calls (see Schegloff, 1986; Houtkoop-Steenstra, 1991;
Undsrrom, 1994), which show considerable variation in different cultures. This variation can only be tackled through gradual accumulation of studies on ordinary conversation in different cultures and social milieux. (1995: 214)
H
Perakylä's observation about the need for comparative work shows that even the most potentially radical approach, like CA, has to take seriously the issue of the empirical genera I izabi lily of its findings. Sometimes, an appeal lo 'possibilities' will be sufficient. Often, however, other examples will be required.
CONCLUDING REMARKS
In this chapter, I have set out various strategies which you can use to defend your research against the charge that it 'merely' depends upon a single case. My overall message is that there is usually no need to be defensive about the claims of qualitative research. As Howard Becker argues:
Sampling is a major problem for any kind ot research. We can't study every case ot whatever we're interested in, nor should we want to. Every scientific enterprise tries to find out something thai will apply to a/en/lhing of a certain kind by studying a few example*, (he results of the study being, as we say 'gcnerahzable' to all members of that class of stuff. We need the sample to persuade people that we know something about the whole class. (1998: 67)
Following Becker, sampling troubles quantitative researchers too. Indeed, as we have seen, the relative flexibility of qualitative research can improve the gcneralizability of our findings by allowing us to include new cases after initial findings are achieved.
The crucial issue here seems to be thinking through one's theoretical priorities. Providing that you have done that and can demonstrate a research design driven by those priorities, nobody should have cause for complaint.
So the secret seems to be to substitute theoretical cogency for the statistical language of quantitative research. In this sense, as Alasuutari has suggested, perhaps 'generalizability' is the wrong word to describe what we attempt to achieve in qualitative research. As he puts it:
110
SELECTING   A   CASE
Generalization is ... la| word ... that should be reserved for surveys only. What-can be analyzed instead is how the researcher demonstrates that the analysis relates to tilings beyond the material at hand ... extrapolation better captures the typical procedure in qualitative research. {1995: 156-7)
SUMMARY
In this chapter, I have discussed four positive answers to the question of how we can generalize from qualitative data:
•    combining qualitative research with quantitative measures of populations
•     purposive sampling guided by time and resources
•     theoretical sampling
•     using an analytic model which assumes that generalizability is present in the existence of any case.
NOTE
1 As Clive Scale (personal correspondence) has pointed out, theoretical sampling may have more to do with generating theories than with empirical generalization. 1 take up Scale's point at the end of this chapter in relation to Alasuutari's argument that the idea of empirical generalization 'should be reserved for survevs only' (1995:156).
Further reading
The most thorough book on this topic is Clive Seale's The Quality of Qualitative Research (Sage, 1999).
Other useful discussions are: Jennifer Mason, Qualitative Researching (Sage, 1996), Chapters 5-7; Pertti Alasuutari, Researching Culture (Sage, 1995), Chapter 12 'Generalization'; and Howard Becker, Tricks of the Trade (University of Chicago Press, 1998) Chapter 3 'Sampling'.
Robert Stake's chapter 'Case studies' is a good account of the conventional qualitative methods position on generalizability (in N. Denzin and Y. Lincoln (eds). Handbook of Qualitative Research. Sage, 1994). Anssi Peräkylä's chapter 'Reliability and validity in research based upon transcripts' is an excellent, more specialist treatment (in David Silverman (ed.). Qualitative Research, Sage, 1997).
Ill