Hi, Joseph,

If your goal is to generalize your findings, I think a qualitative methodology may be the wrong way to go. The Journal Article Reporting Standards for qualitative approaches as recently published in American Psychologist identifies the research goals of qualitative methods. "Qualitative designs are used for research goals, including, but not limited to, developing theory, hypotheses, and attuned understandings, examining the development of a social construct , addressing societal injustices , and illuminating social discursive practices—that is, the way interpersonal and public communications are enacted." The article goes on to mention that qualitative methods are particularly useful in giving voice to sub-populations not well represented in the literature or to exploratory studies.

External validity for any study rests primarily on the sampling approach used to select or recruit research participants. Because qualitative methods almost always involve some kind of purposive selection, generalizability is almost always suspect.

I agree with Peter Kindle's answer above. I would also recommend reading Stake's work to understand the differences between single, multiple and collective case studies, as well as when each one is appropriate. There may be newer editions of these books, but this will give you a start:

Stake, R. E. (2005). The art of case study research. Thousand Oaks, CA: Sage.

Stake, R. E. (2006). Multiple case study analysis. New York: The Guildford Press.

Indeed case study is the research platform for complex phenomena that cannot be generalised. The physics is compelling. The complex spaces comprise many connected and moving nodes. The spaces are nonlinear, and often strongly non-ergodic. You might glean more here: Article Public-Private Partnerships (PPP) on Moulding State Structur...

As a source on how to do comparative case studies, I would recommend Yin's book, Case Study Research (6th edition, 2018), Sage.

There is no magic number for how many cases to include for comparative purposes -- two might be enough if they were very well chosen.

Dear Joseph,

Despite the previous answers of my experienced colleagues, I like to generalize from case studies and also have a magic number -- 20, ideally paired or ordered by some obvious parameters. Sometimes you need (have) to work with a smaller number of cases or have to report only a fraction of actually made cases. See enclosed two papers with the synthesis of a rather limited number of cases.

Bests,

Igor Gurkov

True generalization from a sample to a population is well defined by statistical formulas, which will show you that 20 cases is just about as good as 2 cases, in terms of the level of precision (confidence interval) that it produces. Of course, you could just ignore the formulas and claim that generalizability means whatever you choose it to mean, in which case I choose 2 over 20.

I think case study can't be used, neither is it its intention, to generalize. Rather opposite: yoz csn use case study to analyze specific features of an issue.

Case Study: Conclusion – The Power of Questions

A few PhD students are puzzled about concluding their theses given that case study does not seek generalisation. Recall, generalisation is the holy grail of hypothetico-deductive (H-D) work, and the concluding gold-strip of H-D research.

So, if you face the puzzle, simple: Case study illuminates complexity, triggering questions. Frame those questions and you have a powerful concluding message - a pack that will inform kindred phenomena without instructing them.

Dear Joseph

You have received a bunch of answers with rather contradictory opinions. It is you choice to follow (or not follow) some or all of them. However, please take into account what is the nature of data you wish to get from case studies and then intend to generalize. How once J.March noticed , ‘‘variables that can be measured tend to be treated as more ‘real’ than those that cannot, even though the ones that cannot be measured may be more important’’March, J. G. (2006). Ideas as art. Harvard Business Review, 84(10), 82–89. Interview by Diane Coutu. If your task is to generalize on things that cannot be properly measured (and this is likely as you use qualitative methods) try to produce cases from variable objects (subjects). You will find that in many aspects of social and corporate life the level of isomorphism is high and the variability of objects (subjects) is limited to 4-8 dominant types.

It all depends on what you mean by "generalization". If you mean empirically determining a larger population of cases for which your findings hold (what I would call empirical generalization), you cannot do it based on case studies regardless of the number of cases. This kind of generalization is the domain of quantitative studies, which base it on representative samples from a population.

If you mean theoretically describing the conditions under which certain causal mechanisms produce certain effects (what I would call theoretical generalization), you need a sufficient number of cases to capture the relevant variation of causal mechanisms and conditions under which they operate. For this approach, the number of cases depends on your prior assumptions about causal mechanisms and their initial and operating conditions. These might turn out to be wrong, and you might need t add cases later. But the nice thing about qualitative research is that you can do this.

Yes, there are different types of generalizability, most often statitstical generalizabiltiy is what we talk about and refer to for quantitivtive research. Theoretical gen.... is another type and used in qualititave reserach. However, we do not conduct this research for or about "causal mechanisms," and thus is not related to number of cases. Not to repeat what other have said re the initial quetsion, I quote David Morgan (above), "There is no magic number for how many cases to include for comparative purposes -- two might be enough if they were very well chosen. "

It is the process of selecting cases that is most important and helps determine what is "optimal." Each study /project will be different (no set rule for optimal number) so it is the researcher's task to explain his/her detailed process for selecting which cases and be able to justify whys of the selelction and the number.

Dear Mary-Helen

"However, we do not conduct this research for or about "causal mechanisms," and thus is not related to number of cases. "

You might not do this but I do, and many others as well.

Yes, but it's not personal, I am not alone--- many others also do not. Qualitative research for causal mechanisms is in early stages and not accepted by many scholars and researchers, just as it is by many like yourself. The literature is plentiful on both sides of the issue, as is history with many others such issues with 2 povs-- that' the nature of knowledge / new knowledge. So be it , I'm sorry if you were offended.

Hola,

The most common answer --which is also my answer-- is: "It depends". ¿On what? Demands from your institution, research goals (sometimes, for example, you are just testing a research technique or doing a research experience with students), value assigned to statistical sampling, notions of what science (and, mostly, should be or is supposed to be) and so on.

The only depend-option I do not support is that there is some necessary relationship between quantitative methods and representativeness. There is no: it is just easier in that case. Qualitative studies CAN take representative samples... only that you will need a LOT of money and a LOT of time (not to mention patience). Other than that, no obstacles :).

So, it depends. The first question is not that one. I think you should firstly ask yourself: What do I want to do/achieve? Why? What context am I working in? Who is this research thought/done for? What does the people that evaluate my research think about the possibility of objectivity, the goal of (social) science and so on? Somewhere there lies the answer.

Un saludo,

Javier