Defining the case and its boundary, choosing between the main traditions, and answering the generalisation question before it is asked.
Case study research investigates something in depth within its real-world setting, using several sources of evidence, particularly where the boundary between the thing and its context is unclear. It is a research design, not a sample size: studying twelve people is not a case study, and studying one hospital’s handling of a policy over two years is. Its findings generalise to theory rather than to a population, and saying so confidently is what settles the question examiners always ask.
The most common weakness in case study chapters is that nobody ever defined the case. A case is a bounded system: something with an edge you can point to, examined as a whole.
A case might be an organisation, a programme, a policy implementation, a team, an event, a decision, a community, or one individual’s trajectory. What it cannot be is a topic. “A case study of teacher motivation” names a subject, not a case. “A case study of how one secondary school responded to a new inspection framework across an academic year” names a case, because it has an edge in space and time.
Two questions have to be answered explicitly in the methodology chapter:
Being unable to say where the case ends is a strong signal that a case study is not the right design for the question.
Case study methodology developed along distinct lines, and they disagree about what a case study is for. Naming yours is expected.
The most structured and most cited approach, closer to a post-positivist position. Emphasises a clear research design set out in advance, theoretical propositions that guide what gets collected, multiple sources of evidence, a case study database, and an explicit chain of evidence linking questions to data to conclusions.
Multiple cases are treated through replication logic: cases are chosen because theory predicts they will show similar results, or contrasting results for predictable reasons. This is a deliberately different idea from sampling, and confusing the two is a recognisable error.
More interpretive and more flexible. The case is studied for its own particularity and complexity, and the researcher’s interpretive role is acknowledged rather than minimised.
Stake’s distinction between kinds of case study is widely used: intrinsic, where this case is of interest in itself; instrumental, where the case is a means to understand something more general; and collective, where several cases serve that purpose together.
Developed largely in education research, and sits between the two. Emphasises the case as a bounded system and is strongly practice-oriented, which makes it a common choice in applied and professional doctorates.
Two design choices sit on top of the tradition you pick.
Single or multiple cases. A single case is entirely defensible, and needs a stated reason: the case is critical for testing a well-formed theory, extreme or unique, revelatory because access was not previously possible, or longitudinal. “It was the organisation that agreed” is a real constraint but not a rationale, and where access drove the choice the honest move is to say so and then argue what the case is nonetheless good for.
Multiple cases allow comparison and generally produce more robust theoretical claims, at the cost of depth in each. The number is decided by replication logic rather than by sampling adequacy — you add cases while each is still expected to confirm or extend the developing explanation.
Holistic or embedded. A holistic design treats the case as a single whole. An embedded design examines defined sub-units within it — departments within a hospital, classes within a school — while still analysing the case as a case. The risk in embedded designs is well known: the analysis drifts down into the sub-units and never returns to the case level, at which point it has become something other than a case study.
Case study is defined partly by drawing on several kinds of evidence: interviews, documents, observation, archival records, artefacts. This is not thoroughness for its own sake. Converging evidence from different sources is what makes a case study claim credible, because each source has different weaknesses.
A study calling itself a case study while resting entirely on interviews is vulnerable, and the objection is straightforward: what you have is an interview study of people in one organisation. If documents and observation were genuinely unavailable, say so and account for what that costs the design.
Two practices from the Yin tradition are worth adopting whichever tradition you follow. Keep a case study database — your evidence, organised and separable from your interpretation of it. And maintain a chain of evidence, so a reader can trace a conclusion back through the analysis to the specific material it rests on. Both are straightforward to do as you go and effectively impossible to reconstruct at the end.
Every case study viva includes some version of “but how can you generalise from one case?” It is not a trap, and a confident answer settles it.
The answer is that case studies do not generalise statistically, and are not trying to. A case is not a sample of one drawn from a population; a statistical claim was never on offer. What a case study supports is analytic generalisation: the findings speak to theory, extending, refining or challenging it, and that theory is what travels to other settings.
Two further moves strengthen the answer. Provide description rich enough that readers can judge for themselves whether their own setting resembles yours closely enough for the findings to inform it — transferability is properly the reader’s judgement, made possible by your detail. And state plainly what your case cannot tell anyone, which reads as command of the design rather than defensiveness.
What weakens the answer is quietly generalising anyway — a conclusion that starts describing “this trust” and ends describing “the NHS”.
An in-depth investigation of a bounded system — an organisation, programme, event, decision or individual — within its real-world context, drawing on several sources of evidence. It suits questions where the phenomenon cannot sensibly be separated from its setting. It is a research design rather than a sample size, so a small number of interviews is not by itself a case study.
State what the case is, where its boundary falls in space and time, and what your unit of analysis is. A case has an edge you can point to: not “teacher motivation” but “how one school’s staff responded to a new inspection framework over one academic year”. Everything outside the boundary can still matter as context; it is just not what you are studying. If you cannot describe the edge, a case study is probably the wrong design.
Yes, and many strong doctorates are single-case. It needs a stated rationale: the case is critical for testing a theory, extreme or unique, revelatory because the access was not previously available, or longitudinal. If access was in fact what determined the choice, say so honestly and then argue what this case is nonetheless well placed to show — that is more convincing than a rationale invented afterwards.
Not statistically, and you should say so directly. Case studies support analytic generalisation: the findings speak to theory, and it is the theory that travels to other settings. You strengthen that by providing description rich enough for readers to judge whether their own context resembles yours, and by stating clearly what the case cannot tell anyone. Confidence on this point closes the question; hedging invites more of it.
It is decided by replication logic rather than sampling adequacy. Cases are added while each is still expected either to confirm the developing explanation or to contrast with it for reasons theory predicts. You stop when further cases would not extend or test the explanation. Any number quoted as standard is a convention, and what matters in the viva is the reasoning for the cases you chose.
Yin is more structured and closer to a post-positivist position: designs set out in advance, theoretical propositions guiding collection, a case study database, and an explicit chain of evidence. Stake is more interpretive and flexible, treats the case’s particularity as the point, and acknowledges the researcher’s interpretive role. Both are respectable; what is not defensible is following neither while citing both.
An undefined case is the problem that surfaces at examination, long after it can be fixed. Describe what you intend to study and where you would draw the edge, and a PhD in your field will work through it with you.
Leave an email or a WhatsApp number — whichever you prefer — and tick how we should reach you. We reach out within 30 minutes.
