13 Research methods, statistics, critical review and evidence-based practice13.1 Introductory concepts
13.1.6 Assessing cause and effect
2 min read1 figure
Written by Resident Doctor Committee · Published
New in the 2027 syllabus
The College added this topic for exams from January 2027.
Association and causation
An association means two things occur together. Causation asks how the outcome would differ if the exposure were changed, compared with an otherwise comparable alternative.
Temporality: the cause must come before the effect. If exposure and outcome are measured at the same time, their order is often unclear.
Reverse causation: the outcome, as it develops, changes the exposure. For example, early symptoms may prompt treatment.
Competing explanations
Before accepting that an association is causal, consider other explanations.
Explanation | How it works |
|---|---|
Confounding | A third factor causes both exposure and outcome |
Selection bias | Entry into the analysed sample depends on exposure and outcome |
Measurement bias | An outcome is more likely to be detected in one exposure group than another |
Chance | Sampling variation produces an apparent association |
Causal effect | Changing the exposure changes the outcome |
Mediator: a variable on the causal pathway. Adjusting for it can remove part of the total effect you are studying.
Collider: a variable caused by both of two other variables. Conditioning on it can create an association between those two.
Causal diagram: a drawing of your assumptions about how variables relate. Drawing the arrows does not show those assumptions are correct.
Bradford Hill considerations
Consideration | What it adds to the argument |
|---|---|
Strength and consistency | A large, replicated association may be harder to explain by a single bias |
Biological gradient | A dose-response pattern (more exposure, more outcome) may support a causal explanation |
Plausibility and coherence | The explanation fits other scientific knowledge |
Experiment | Changing the exposure under controlled conditions changes the outcome |
Specificity and analogy | A focused relationship, or a similar established mechanism, may support the interpretation |
Hill's considerations are prompts for judgement, not necessary and sufficient criteria. Only temporality is required. A real cause can produce an association that is neither strong nor specific.
Treat dose-response with care. Confounding can also produce it, and a genuine effect may have a threshold or a curved relationship.
Strengthening an inference
Randomisation separates allocation from prognosis, on average. After the design stage, allocation concealment, valid measurement and follow-up protect the comparison.
Triangulation compares results from approaches with different likely biases. Repeating the same biased design does not count as independent confirmation.
Sensitivity analyses test how far plausible unmeasured confounding, or other assumptions, could change the conclusion.
Residual uncertainty should stay visible. If the data cannot tell the causal claim apart from its alternatives, say so.