Judging quality in each family · Lesson 2 of 3

Generalisability and transferability

Generalisability is about how the sample was drawn. Transferability is about how well the context was described.

8 min read · Free — no account needed to read

Generalisability is the quantitative goal: does what I found in this sample hold in the population it came from? It depends far more on how the sample was drawn than on how big it is. A random sample of 1,000 from a defined population supports generalisation to that population. A self-selected sample of 100,000 people who clicked a link supports generalisation to people who click that kind of link.

This is why "big data" does not automatically solve it. A large biased sample is a precisely measured wrong answer, and the precision makes it more persuasive rather than less.

Transferability is the qualitative counterpart and works differently. Qualitative work usually does not claim its findings hold elsewhere. It provides enough description of the setting, the participants and the context that a *reader* can judge whether it transfers to theirs. The responsibility deliberately shifts to the reader, which is why thick description is treated as a quality marker rather than padding.

The parallel criteria run right through. Quantitative reliability corresponds to qualitative dependability: is the process documented well enough to be followed? Internal validity corresponds to credibility: are the interpretations recognisable to participants and grounded in the data? Objectivity corresponds to confirmability: can a reader trace a finding back to the material rather than to the researcher's expectations?

They are not translations. Qualitative work is not attempting a worse version of generalisation — it is answering a different question, and it has built criteria that suit it.

See it happen

A study of 18 nurses in two hospitals describes how handover errors happen, and a colleague dismisses it as unrepresentative.

The read that arrives on its own

18 people in 2 hospitals cannot tell you about nursing in general, so it cannot support a policy change.

The same situation, with this idea

You ask what it was for. The study is not claiming a rate; it is describing a mechanism — the specific sequence by which information gets lost. A mechanism found in 18 cases is a real mechanism that exists. Whether it operates in your hospital is a transferability judgement you can make from the description, and if it does, you now know what to look for. It answers "how does this happen", which no survey of 10,000 would have produced.

After this, you can

ask what claim a study is making before judging its sample.

The takeaway

Generalisability is about how the sample was drawn. Transferability is about how well the context was described.

Easy to take the wrong way

That transferability is a softer standard. It demands a great deal of description and honesty about context — a thin qualitative study fails it plainly.

Try this

Take a study you rely on and write what it can support: a rate, a mechanism, or an experience. Then check whether you have been using it for that.