Case studies are a completely different research method from statistical analysis. Whereas statistical studies are good for showing you the extent of a phenomenon, they are not good for providing explanation. This is because statistical studies only investigate phenomena at the empirical level, it cannot penetrate to investigate the realm of the real. To investigate causality, it is necessary to do a case study. in a case study you do an in-depth analysis of a particular occurrence of a phenomena and can go deep to investigate the causal factors in the real realm.

So you really have a choice, study a phenomena extensively with statistical methods or study it intensively with a case study. As with any program of study, it is necessary to do multiple studies in order to identify what causal factors may be general to the occurrence of a phenomenon.

I agree with Michael- case studies are very different from statistical analysis, and will yield in-depth data which can be used to develop hypotheses abotu causality. These may support existing literature/ theory, or may suggest gaps in knowledge which need further investigation.

Replication, or conducting a case series is probably the best way to generate data which can be generalised.

A case study, which provides detailed information about the context of the case can enable generalisation to other similar cases. A case series which draws on cases which are different in some ways, but which has similar conclusions strengthen the argument for generalisation.

A well detailed questionnaire analysis, supported by descriptive statistics based on factor and cluster analyses, might be a proper preliminary picture, in order to introduce case studies that can be generalised (if results turn out to be consistent....)

Dear Sir,

thanks very much for your insightful answer. However, I have doubt about cluster analysis. once clusters are sampled one has to employ systematic random sampling since you are not able to interview every one in the cluster. In contrast, case study method is based on theoretical sampling which does not require any form of random sampling.

With this somewhat offensive reply, I want to thank you with gratitude for sending me some of your seminal papers.

Upananda (Sri Lanka)

I think that transferability is a more relevant concept for case study research, rather than generalizability. It concerns the extent to which the results from one study are relevant to other contexts. Presumably what you learn from your case studies is not limited to specific cases you studied, so the idea is to be reflective about other settings where your findings might be relevant.

The original source for the concept of transferability is Lincoln & Guba's 1985 book, Naturalistic Inquiry.

Also, cluster analysis as a data analysis technique is completely different from clustered sampling.

Dear Sir,

Thank you very much for your kind contribution. could you please explain whether we can do cluster analysis without sampling.

According to Wikipedia, "Cluster analysis or clustering is the task of grouping a set of objects in such a way that objects in the same group (called a cluster) are more similar (in some sense or another) to each other than to those in other groups (clusters)."

In other words, a cluster analysis would sort your cases so that sets of them that were similar to each other would be grouped together, while those that were relatively different from each other would be placed in other groupings.

Because cluster analysis is classified as an exploratory procedure, it doesn't involve any assumptions about sampling (which also means that it will not produce tests of significance).

In terms of further information, the bad news is that the Wikipedia article gets quite technical after the definition above. The good news is that a google search for "cluster analysis" will produce several useful sources.

I have little to add to Michael's answer except that there are two different generalization strategies. Quantitative studies achieve what one can call 'empirical generalization': They analyse samples and are able to exten their statements about statistical associations in the samples to the population from which the sample is drawn. Case studies cannot do this but can apply what one can call 'theoretical generalisation'. With case studies, we can identify causal mechanisms, conditions under which they operate (including those triggering them), and effects they produce. We can then say that wherever the identified conditions occur, the mechanisms are likely to operate and to produce the identified effects (and leave the task of identifying the population in which the conditions occur to our colleagues from quantitative research). In this sense quantitative and qualitative research are indeed complementary: Quantitative studies can identify causes, effects and the realm in which the causes produce the effects but cannot explain *how* the causes produce effects. Qualitative studies can explain how the causes produce effects but cannot empirically delineate the realm in which this happens.

Upananda, I believe you have first to clearly define the nature of your study's generalizability. If you are referring to generalizing towards the population of which your case is just a non-random sample, I think it is clear that you cannot make an inference.

However, if you are pertaining to making generalization within your case, it would be different if in that particular case you have identified a distinct population and could get random samples. Your generalization would be towards the "population" within your case, not beyond it. I hope this helps.

Thanks for all (Eddie, Jochen, David, Michael , Vivarelli, and Mark) for clarification.

This clarification explore an important aspect of empirical studies.

Majority of researchers select populations from their territories/countries where economies are stable and uncertainty of the institutional environment (political, economic, sociology-cultural, and technological) is minimized. Researchers in the developing countries , very often, attempt to apply those findings without making room for adjustments to those findings in the West. As a result most of management theories do not work equally well in the East.

I thinks case studies play an important role in adopting/adaptations of the empirical findings in the West (with gratitude).

Upananda

Thanks for this question!

Upananda,

You are quite correct. Findings obtained in the first world may not apply to the third world. There is a need to repeat these studies in the third world since as you mention the situations (stability and certainty) are often not found there.

This also creates a great opportunity for contrastive case study. You could study similar organizations, one in the first world and another in the third and then compare and contrast. This will yield a fascinating view into the phenomenon under study.

Either way, replication of an existing study or contrastive case study will yield a wonderful contribution to the field. I wish you great success in finding your research subjects.

Mike

Dear Sir (Mike)

Thanks for your kind words and insightful suggestions.

Upananada

Having read this thread, I think the underlying issue is not one of quantitative methods or qualitative methods and empirical generalisation or theoretical generalisation which all have been explained well by previous contributors. The question is more how assumptions are being treated in empirical research. What we usually do is partial research, i.e. we assume away some influences which we expect to be the same (or randomly distributed) across our sample. The benefit of qualitative research is maybe that it needs fewer assumptions and is better suited to include contextual influences, but most certainly some untested assumptions will always remain. No matter whether you want to infer from qualitative of quantitative studies conducted in a first world context, you will always have to evaluate: Are the assumptions also applicable in the thirld world context and, if not, how would it affect the results if we modified them?

Good question! I'd like to add a note to Jochen Glaser's very useful answer. Blatter & Haverland's "Designing Case Studies: Explanatory Approaches in Small-N Research" (2012) differentiates *three* types of generalization. As I understand it (and please correct me if I'm off base):

(1) "Statistical generalization" is similar to Jochen's description of 'empirical generalization' above, and is based on a knowing the results of a statistically significant sample of a population. So if you know the results of a representative sample of cases, we can predict the results of additional cases, on average. We can't predict a single case with absolutely certainty, but we posit that, on average, predictions will be correct more often that not. That phrase "on average" is the key to understanding how statistical generalization works.

(2) "Possibilitistic generalization" is similar to what Jochen labelled 'theoretical generalization' above. The term "possibilistic" seems to refer to the possibility, or likelihood, that a causal mechanisms will result in similar effects given similar conditions. Since no two cases are perfectly identical, we don't know for sure how a particular case will turn out. The higher our confidence that similar conditions will produce similar results, the stronger our ability to generalize.

(3) Blatter & Haverland reserve the term "theoretical generalization" for research designs which seek to understand the power of different theories for explaining a particular case. Here the strength of our generalization depends on our ability to draw conclusions from the relevant theories for more or less 'crucial' cases. The main difference between this research design and the prior one is that the focus is on using a "known" case to provide insights on a new case in the first, or using a "known" theory to provide insights on a new case in the second.

This third approach to case study design and generalization is often overlooked, but has been employed to great success in cases such as Graham Allison's analysis of the Cuban Missile Crisis.

My brief look at Blatter & Haverland suggests that their concept of "possibilitistic generalization" is similar to what Lincoln & Guba call "transferability in their 1985 book, Naturalistic Inquiry.

With transferability, the goal is to describe other situations where the results of a case study might also be relevant. Thus, there is nothing about statistics that determines whether a developed-world study is relevant for the developing world, but there might be things about a case study that make it meaningful across a variety of settings.

I like to think of transferability in terms of concentric circles. The circle nearest to your own case will consist of similar issues in similar circumstances, so the transferability is relatively obvious. Further out, at the periphery, there will be equally obvious differences that limit transferability. So the real interest is in the middle ground, where you can make arguments about whether your results could indeed have relevance for (be transferable to) other kinds of cases.

Yin has some great advice on this in: Yin, R. (2014). Case Study Design and Methods (Fifth Edition (Kindle) ed.). Thousand Oaks: Sage Publications.

Thanks to u all