Hill’s considerations for causation help organize an argument about whether an observed association may be causal. They are not a scoring system, a proof checklist or a replacement for sound study design. Start by considering chance, measurement, selection and confounding, then ask what the evidence does and does not establish.
Austin Bradford Hill discussed nine viewpoints in his 1965 paper: strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment and analogy. He rejected hard-and-fast rules for declaring cause and effect. The original paper is available as a republication of Hill’s address.
Strength and consistency
Strength concerns the magnitude of an association. Interpret a risk ratio, odds ratio or other measure in its own terms; a large estimate can still reflect bias or confounding, and a small estimate does not rule out a causal effect.
Consistency concerns findings across studies, settings or approaches. Repeated findings can strengthen an argument, but repeated studies can share a measurement problem. Differences across settings may also reflect different populations, exposure levels or effect modification rather than a simple refutation.
Temporality and biological gradient
Temporality asks whether the relevant exposure preceded the outcome’s development. That ordering is logically necessary for a proposed cause, but it is not sufficient: an earlier exposure may have no causal effect. A recorded diagnosis date may also occur after disease began, so consider the timing the research question actually requires.
Biological gradient asks whether outcome patterns vary with exposure level. A dose-response pattern can be informative, but confounding can produce a gradient and genuine relationships need not be linear or monotonic. Absence of a simple gradient does not automatically rule out causation.
For the place of temporality and the other viewpoints in causal assessment, see the Surgeon General’s discussion of causal inference.
The other viewpoints
- Specificity: a distinctive exposure–outcome pattern can be informative. One exposure can nevertheless have several outcomes, and one outcome can have several causes.
- Plausibility: a credible mechanism matters, but current biological knowledge is incomplete.
- Coherence: consider whether the proposed explanation fits what is known from different kinds of evidence.
- Experiment: evidence from changing exposure can strengthen an argument when the intervention and comparison are valid.
- Analogy: related causal knowledge may guide hypotheses, but resemblance is not direct evidence for the new claim.
Assess evidence, not checklist points
Do not label a relationship causal because it meets a fixed number of viewpoints or has a small p-value. Ask whether alternative explanations remain, what comparison the study supports, and how much uncertainty there is.
Randomized trials and observational studies offer different designs and limitations. Random assignment addresses baseline confounding in expectation, not every measurement, missing-data or reporting problem. Hill’s viewpoints can inform interpretation of evidence from either type of design; they are not limited to a post hoc review of observational studies.
Fictional program example: participation doubles after a public-health program begins, while a separate media campaign starts at the same time. The sequence establishes that the program preceded the measured increase. It does not establish how much participation would have changed without the program. A theory of change and a large before–after difference are not substitutes for a credible comparison.
Practice questions
Question 1: identify the stated pattern
In a fictional study, observed lung-cancer risk rises across groups reporting increasing numbers of cigarettes smoked per day. Which Hill consideration is most directly illustrated by this increasing-exposure pattern?
- Consistency
- Temporality
- Biological gradient (dose-response)
- Specificity
Answer and explanation
C) Biological gradient (dose-response). The question states an exposure-level gradient. It does not provide exposure and disease-onset timing, so temporality cannot be inferred from the gradient alone. The pattern supports investigation of a causal explanation but does not prove it.
Question 2: necessary is not sufficient
A prospective study documents an exposure before the outcome develops. Which consideration does that observation address, and does it by itself prove causation?
Answer and explanation
Temporality; no. Correct ordering is necessary for the proposed exposure to cause the outcome, but confounding, selection, measurement and chance still need consideration.
Question 3: match the observation
Match each observation to its most direct consideration: an association estimate far from the null; similar findings across independent settings; a proposed biological mechanism; and changing risk across exposure levels.
Answer and explanation
Strength; consistency; plausibility; biological gradient. These observations address different parts of an argument. None makes an unsupported exposure-before-outcome sequence appear in the data.
Question 4: the program comparison
For the fictional program example, name two alternatives to the claim that the program caused the full increase, and identify a design improvement.
Answer and explanation
The media campaign and an underlying time trend are plausible alternatives. A suitable comparison group or a justified time-series design could strengthen assessment of what would have happened without the program. Their assumptions and other threats still need evaluation.
Question 5: specificity
True or false: a proposed exposure cannot be causal unless it produces only one disease.
Answer and explanation
False. Multiple outcomes do not rule out a causal relationship. Consider the total evidence for the particular exposure, outcome and population instead of treating specificity as a mandatory condition.
Sources and related reading
- Hill (1965), republished: The environment and disease—association or causation?
- The Health Consequences of Smoking: introduction and approach to causal inference
- Causal inference in public health
- Confounding and effect modification
- Randomized controlled trials
Correction — October 11, 2026: the previous first practice question selected temporality even though its scenario did not establish exposure-before-outcome timing. The question now asks specifically about a stated dose-response pattern, with a matching answer and explanation. Other wording was revised to avoid treating Hill’s viewpoints or study designs as proof of causation.
Humanities Moment
The featured image for this article is “‘Bloedberg’ in Antwerp” by Léon-Eugène-Auguste Abry. Léon-Eugène-Auguste Abry was a Belgian painter (1857–1905) best known for realist, often humorous and documentary depictions of military life and portraits rooted in academic training. He contributed to late 19th-century Belgian art by combining careful observation and sketching from actual troop maneuvers with polished academic technique, helping popularize authentic military genre scenes and informing public visual understanding of the army.
If you’ve found this CPH Focus session useful, a share to your friends or colleagues would be greatly appreciated! Likewise, please check out our full CPH Focus and Epi Explained series for more Public Health Knowledge!
