How to choose a study design
Last reviewed 16 October 2026
The design should follow from the question, not the other way round. Start with what the question asks, then check what is feasible, ethical and affordable.
Open the study design chooserStart from the question
Does an intervention change an outcome? Use a randomised trial where it is ethical and feasible, otherwise a cohort or a quasi-experimental design. How common is something? A cross-sectional survey. How accurate is a test? A diagnostic accuracy study. Who will develop an outcome? A prognostic cohort. How do people experience something? Qualitative methods. What does the existing evidence say? A systematic or scoping review.
Randomised or not
Randomisation is the best protection against confounding, but it is not always possible. When groups cannot be randomised, a cohort study with careful adjustment, or a method such as propensity scores or interrupted time series, can answer the question with stated limits.
Rare outcomes and common exposures
For a rare outcome, a case-control study is efficient because it starts from the cases. For a rare exposure, a cohort study starts from the exposed. Each has typical biases to plan for.
Individuals or groups
If an intervention acts on whole clinics or villages, or contamination between participants is likely, randomise clusters and plan the larger sample that follows.
Report it properly
Each design has a reporting guideline: CONSORT for trials, STROBE for observational studies, STARD for diagnostic accuracy, PRISMA for systematic reviews, COREQ or SRQR for qualitative work. Journals and ethics committees expect them.
Common questions
- Is a randomised trial always the best design?
- It is the strongest design for the effect of an intervention, but only for that question. It is not suited to measuring prevalence, or to exposures that cannot be assigned.
- When is a cross-sectional study enough?
- When the question is how common something is at one point in time. It cannot show which came first, so it does not prove cause.
- Can I combine designs?
- Yes. Mixed-methods studies pair, for example, a trial with interviews that explain why the intervention worked or failed.
Keep reading
General methods guidance, not a substitute for advice from a statistician on your specific study.