Sampling in qualitative research

Quantitative research is concerned with quantities and frequencies of occurrence in the general population, so the study sample must be representative. In qualitative research it can often “be impossible to map the population from which a random sample might be taken” (p416), making probability sampling infeasible. Qualitative researchers may well be familiar with statistical techniques but that is hardly the point. It is true that certain types of qualitative research emphasize the importance of developing research questions during research, adding to the sample from time to time. Sampling remains important for all types of research, however.
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 408

Qualitative researchers often use purposive samples rather than random, probability samples, in order to develop a grounded theory. This typically involves selecting additional participants on the basis of the ideas and concepts that emerge as the project progresses, and it ensures that the researcher gathers data that is relevant to their research questions. Consequently, it is better to start out by choosing members of a sample “purposely”, deliberately, in other words, rather than randomly.
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 410-412

The big problem for many researchers is that very little can be generalized from findings emerging from purposive samples. This approach to sampling is done, after all, because a probability sample may not be feasible or appropriate. So generalizability, or the lack of it, is bound into the definition of sampling type. Onwuegbuzie and Leech (2010, cited on page 418) point out, however, that it is possible to generalize from one case to another that is broadly similar. We are cautioned not to make claims for our data that cannot be supported by our research methods.
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 416-418

Ethnographic research tends to rely on convenience or snowball sampling, because the ethnographer can only glean information from whoever is prepared to talk to them. The shifting population of such groups also makes it difficult to map out the sampling frame from which a probability sample could be selected. However, as a qualitative research design, it is generally seen as more important for this technique to lead to interpretive understanding than to statistical generalisation.
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 418

The very nature of purposive sampling makes it relatively free from fixed constraints but we use the terms “fixed” and “contingent” to refer to sampling on a once-off basis or conducted sequentially. Similarly, “a priori’ and “contingent” refer to the terms we use for criteria for sample selection – being determined in advance, or changing as the research progresses. Bryman uses the label of ‘generic purposive sampling’ to refer to sampling done purposively but not (necessarily) in order to generate theory. This means that generic purposive sampling remains ‘purposive’ but will mostly be fixed rather than sequential and the criteria will be set ‘a priori’ (at the outset) rather than being contingent on data gathered.
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 412-414

Bryman draws from the work of Patton (1990) and Palys (2008) to produce a list of prominent types of purposive sampling, including answers (b), (c) and (d). The whole point of purposive sampling is that it is not probability sampling. The aim is to recruit people to the sample because of particular characteristics they have, rather than for their (possibly unlikely) ability to represent the entire population.
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 409, Key concept 18.2

This term relates to Glaser & Strauss’s (1967, cited on page 410) grounded theory, in which the aim is to allow concepts to emerge gradually from the data. Theoretical sampling involves collecting more and more data to refine one’s theory until no more new ideas emerge; this is called the theoretical saturation point. Strauss & Corbin (1998, cited in Key concept 18.3 on page 411) show how saturation can be reached at the levels of concept, category and relationships between categories.
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 410-412, Key concept 18.4

Trying to figure out the ‘right’ number of people to interview is almost impossible in theoretical sampling – who can say in advance when the point of theoretical saturation will be reached? On the other hand, if the objective of the research is to make comparisons, then a minimum number should be established. Warren (2002, cited on page 416) says that publishers (and we add: dissertation assessors!) are unlikely to be impressed with a number less than twenty. Bryman (p418) recommends concentrating on your reasons for using a particular sampling strategy and then explaining why you feel a particular sample size is justified. The very nature of qualitative research obliges us to give answer (d) as correct, even though it may not be very helpful!
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 416-418

Bryman reports (p409) on the study by Savage et al. (2005) in his discussion on sampling levels. The main point is that we may not get the information we want from a single attempt at sampling. The study’s authors picked Manchester as a typical case, then four residential areas within the Manchester area. Next they used a quasi-random method for selecting households within each area. Answer (d) is therefore correct. They did use the electoral register as a sampling frame, so there was an element of statistics and randomness in their approach. Were answers (a) and (c) inverted (to purposive and random, for example), they would not be totally inaccurate.
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 409

The principles of purposive sampling can be applied to documents in the sense that we will select for our sample those that satisfy certain criteria relevant to our research questions. Because of the practical problems encountered in ethnography, there is a need to sample for observations at different time periods. Equally, observation studies need to be clear about how behaviour may change in different circumstances. Since this is not probability sampling, we are free to pick and choose, as it were, between contextual issues. We just need to stay consistent with our research questions.
Reference: Bryman: Social Research Methods: 5th Edition Page(s) 418