Re-conceptualizing generalization : old issues in a new frame

The paper deals with three distinctive and separate issues usually mismatched in methodology: sampling, representativeness and generalisability. 1. Sampling in qualitative research had hard times. On one hand it has been long neglected by many qualitative researchers as a mere positivistic worry; on the other it has been undervalued by quantitative researchers because not probabilistic. The formers forget that sampling is an unavoidable step because it is, first of all, an everyday life activity grounded in thought (e.g. the heuristic of representativeness by Kahneman and Tversky, 1972), in language (e.g. the synecdoche, as pointed out by Becker, 1998, 67) and into practice: the buyer tastes a small helping to choose a wine or a cheese; the teacher asks a student some questions to assess his/her knowledge on the whole syllabus. On the other hand quantitative researchers did not realised yet that in social science probabilistic samples are often a chimera because of (1) the lacking of population list for most of the studies on special groups of people and (2) the phenomenon of nonresponse. Besides experiments (reckoned the best possible example of scientific procedure over the years and also today) are not based on probability samples. 2. Fortunately in contemporary qualitative research the problem of representativeness is a constant and growing concern of many researchers (see Silverman 1997; 2001; Seale 1998). Two main linked questions are usually raised: a. how do we know how representative our cases (sample) are of all members of the population from which the cases was selected? b. can we generalise from case study findings to population without following a purely statistical logic? These questions refer to two analytically separated problems usually mismatched: the representativeness of samples and the generalisability of findings. Offering these issues as two sides of the same coin neglects the existing social space between these two activities. As a matter of fact a researcher can conduct a study on a representative sample but findings cannot be generalised for a number of practical reasons related to different method reliability (interview versus focus group versus ethnography; sloppy data collection; researcher's biases in data analysis; unsuccessful access and relations in the field) and to ecological validity of collected data. 3. This complicate tangle (the non-linear relationship among selecting a sample, its representativeness and generalising findings) has been solved without anxiety by most well-known anthropologists and sociologists, who produced important theories. Good examples are the Lynd’s (1929, 1937), Whyte (1943), Gouldner (1954a and 1954b) observing a small gypsum extraction and refining factory, Dalton’s research (1959) at Milo and Fruhuling, two companies in a highly industrialised area in the US, Geertz (1972) who attended 57 cock fights, De Martino (1961) who observed 21 people suffering from tarantism disease, Becker (1951) who studied many dance musicians, Sudnow (1967) who observed two hospitals, Goffman (1963) who analysed various side involvement and so on. This may lead to the idea that thinking about these complicate matters is wasting time. However defining the sampling unit clearly (which comes before choosing the cases thus picking the sample) is basic to avoid messy and empirically shallow researches. During their analysis of some Finnish researches on “artists”, Mitchell and Karttunen (1991) noticed different findings according to the definition of artist employed by the researchers. The latter included in the category “artist” in some cases (i) those who describe themselves so, in other cases (ii) someone who constantly creates durable works of art, in other cases (iii) someone who is considered an artist by the whole society or (iv) someone who is taken into consideration by artists’ associations. A comparison is therefore impossible. Beginning from the useful but still vague concept of “theoretical sampling” (Glaser and Strauss 1967) the author discusses: • how to practically select a sample according to the conceptualisation of the research topic and questions • some strategies of qualitative sampling (purposive, quota, emblematic case, snowball and so on) and its pitfalls, obstacles, constraints and deviant cases • the concept of representativeness linked with the concept of pervasiveness of the social phenomena. In other words if ethologists, zoologists, astrophysicists, archaeologists, biologists, historians, cognitive scientists and so on do science observing small samples, why not the qualitative researcher? • how generalise from small samples. In other words how to manage the irremediable gap between situation and theory, indexical observations and general statements (see Gubrium and Holstein, 1997).