Skip to content

Bradford Hill criteria

Nine epidemiologic viewpoints used to organize evidence about whether an observed association may be causal, without treating any fixed subset as a necessary, sufficient, or mechanically scored proof of causation.

Version
v1 · 2026-09-28 · History
Domain-specific #
8260
Domain group
Applied Sciences & Engineering
Origin domain
Medicine & Healthcare
Subdomains
Epidemiology, Causal Inference → Medicine & Healthcare

Core Idea

The Bradford Hill criteria, more accurately Hill's nine viewpoints, are a framework for asking whether an observed epidemiologic association may reflect causation. The viewpoints are strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy. They organize heterogeneous observations rather than replace research design or causal reasoning.

The framework is often misused as a checklist. Hill did not require every viewpoint, and most are neither individually necessary nor sufficient. Temporality has a special logical role: a cause must precede its effect. The evidential force of the others depends on context; weak associations can be causal, specificity can fail in multifactorial systems, dose-response patterns can be nonmonotonic, and lack of a known mechanism can reflect incomplete knowledge. A defensible assessment also examines confounding, bias, measurement, selection, reverse causation, intervention evidence, and uncertainty. The output is a reasoned synthesis, not a numerical verdict.

How would you explain it like I'm…

Nine Clue Questions

When doctors notice that two things often happen together, like a habit and a sickness, they wonder: does one actually cause the other? The Bradford Hill criteria are a set of nine questions that help them think it through, like 'Did the habit come first?' and 'Does more of the habit mean more sickness?' They are thinking helpers, not a score sheet — the answer comes from putting all the clues together carefully.

Nine Clues for Causes

The Bradford Hill criteria, also called Hill's nine viewpoints, help scientists who study health in populations decide whether a link between two things might really be cause and effect. The nine are: strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy. Only one is a must: the cause has to come before the effect, which is temporality. The others are clues that can help or not depending on the situation — a weak link can still be causal, for example. People sometimes treat them like a checklist to tick off, but Hill didn't mean them that way. Scientists also have to look for other explanations, like mistakes in how data were gathered.

Hill's Viewpoints on Causation

The Bradford Hill criteria — more accurately Hill's nine viewpoints — are a framework for judging whether an observed epidemiologic association may reflect causation: strength, consistency, specificity, temporality, biological gradient (dose-response), plausibility, coherence, experiment and analogy. They organize different kinds of evidence; they do not replace good study design or causal reasoning. They are often misused as a checklist, but Hill did not require all of them, and most are neither necessary nor sufficient on their own. Temporality is special because a cause must come before its effect. The others depend on context: weak associations can be causal, specificity often fails when diseases have many causes, dose-response can be non-monotonic, and a missing mechanism may just reflect incomplete knowledge. A sound assessment also considers confounding, bias and reverse causation, and ends in a reasoned judgment rather than a score.

 

The Bradford Hill criteria, more accurately Hill's nine viewpoints, are a framework for assessing whether an observed epidemiologic association may reflect causation: strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experiment, and analogy. They organize heterogeneous observations rather than substitute for research design or formal causal reasoning. A frequent misuse is treating them as a checklist; Hill did not require every viewpoint, and most are neither individually necessary nor sufficient. Temporality has a distinct logical role, since a cause must precede its effect. The weight of the others is context-dependent: weak associations can be causal, specificity often fails in multifactorial systems, dose-response relationships can be nonmonotonic, and absence of a known mechanism may reflect incomplete knowledge. A defensible assessment also addresses confounding, bias, measurement error, selection, reverse causation, interventional evidence, and uncertainty, and yields a reasoned synthesis rather than a numerical score.

Structural Signature

Sig role-phrases:

  • observed exposure-outcome association. Defines the empirical relation whose causal status is being assessed. Constitutive starting point. If altered: The framework does not create an association from no data.
  • temporal ordering. Tests whether the proposed cause precedes the outcome. Constitutive causal requirement. If altered: Reverse temporal order defeats that causal direction.
  • nine viewpoint pattern. Organizes strength, consistency, specificity, gradient, plausibility, coherence, experiment, and analogy around temporality. Identity-bearing heuristic set. If altered: The viewpoints can be informative without all being present.
  • bias confounding and alternative explanations. Assesses selection, measurement, confounding, reverse causation, and competing mechanisms. Necessary evidential discipline. If altered: Viewpoint agreement cannot rescue a badly biased design.
  • integrated causal judgment. Combines study designs, background knowledge, uncertainty, and consequences into a qualified conclusion. Constitutive output. If altered: A point score is not the framework's warranted conclusion.

What It Is Not

  • Causal checklist. Are the nine viewpoints being mechanically counted?
  • Statistical significance. Is association strength being mistaken for causation?
  • Evidence hierarchy. Are study types ranked instead of integrated?
  • Causal proof. Is uncertainty falsely eliminated?

Scope of Application

Use Bradford Hill criteria with exposure, outcome, population, time order, study designs, each relevant viewpoint, bias and confounding analysis, causal alternatives, and uncertainty stated.

  • Epidemiology. Assesses exposure-outcome causation.
  • Public health. Synthesizes policy evidence.
  • Occupational health. Evaluates workplace hazards.
  • Environmental health. Assesses distributed exposures.
  • Evidence assessment. Organizes heterogeneous studies.

Clarity

Converging viewpoints can strengthen a causal interpretation, but dependence among studies and shared measurement errors can create false consistency. Plausibility and coherence are also historically contingent on current science.

Manages Complexity

Policy may require action before causal certainty, while scientific statements should preserve uncertainty. The framework helps expose reasoning but does not choose the decision threshold.

Abstract Reasoning

  1. Define exposure, outcome, population, and causal direction.
  2. Verify that exposure precedes outcome.
  3. Review all nine viewpoints without checklist scoring.
  4. Test bias, confounding, selection, measurement, and alternatives.
  5. Integrate designs and uncertainty into a qualified causal judgment.

Knowledge Transfer

Multi-viewpoint evidential synthesis transfers to other causal sciences, but the named nine viewpoints and epidemiologic association context delimit this framework. The nearest stopping boundary is explicit: A generic causal checklist is closest: it may ask similar questions, but the Bradford Hill framework is the historically specific nine-viewpoint set and explicitly resists mechanical necessary-and-sufficient scoring. The inclusion test remains: An assessment uses the Bradford Hill framework when it examines an exposure-outcome association through the nine named viewpoints as aids to a reasoned, uncertainty-aware causal judgment rather than as automatic proof. The structure no longer applies when the case exits when temporal order is false, alternatives are not examined, or viewpoint counts are treated as conclusive proof.

Examples

Canonical

An occupational study program evaluates whether an exposure causes disease by establishing that exposure precedes onset, comparing results across populations and designs, examining gradient and mechanism, seeking experimental or quasi-experimental evidence, and testing confounding and bias before reaching a qualified conclusion.

Mapped back: observed exposure-outcome association → workplace exposure and disease; temporal ordering → exposure before onset; nine viewpoint pattern → multiple named viewpoints assessed; bias confounding and alternative explanations → explicit design review; integrated causal judgment → qualified synthesis.

Applied / In Practice

A report declares causation because six of nine boxes are checked, although exposure was measured after diagnosis and selection bias was not assessed. The vocabulary is Hill-derived, but the procedure is not a valid Bradford Hill synthesis.

Mapped back: observed exposure-outcome association → reported correlation; temporal ordering → uncertain or reversed; nine viewpoint pattern → box count; bias confounding and alternative explanations → omitted; integrated causal judgment → unwarranted certainty.

Structural Tensions

T1: structured review vs. false algorithm. Named viewpoints improve transparency but invite mechanical scoring. Diagnostic: How is contextual judgment preserved?

T2: plausibility vs. knowledge limits. Mechanistic fit can support causation while penalizing novel effects. Diagnostic: Is implausibility evidence or ignorance?

Structural–Framed Character

Description turns on observed exposure-outcome association, temporal ordering, nine viewpoint pattern, bias confounding and alternative explanations, integrated causal judgment. Skeletal core. Multiple heterogeneous tests interrogate a proposed causal relation and feed a qualified synthesis rather than a threshold score. Domain-bound accent. Exposure, outcome, temporality, gradient, plausibility, coherence, experiment, confounding, and epidemiologic design define the framework. Transfer remains bounded because Why not prime. Triangulated causal assessment is portable; this is Hill's historically named epidemiologic heuristic. The negative boundary is concrete: Any correlation, randomized-trial result, causal diagram, risk-factor list, statistical significance test, mechanism claim, evidence hierarchy, or checklist score is not automatically a Bradford Hill assessment. The Bradford Hill framework is heuristic-evidential: nine viewpoints structure causal interpretation without constituting an algorithm. Its character: disciplined triangulation around an association, with temporality and alternative explanations kept visible.

Structural Core vs. Domain Accent

Skeletal core. Multiple heterogeneous tests interrogate a proposed causal relation and feed a qualified synthesis rather than a threshold score.

Domain-bound accent. Exposure, outcome, temporality, gradient, plausibility, coherence, experiment, confounding, and epidemiologic design define the framework.

Why not prime. Triangulated causal assessment is portable; this is Hill's historically named epidemiologic heuristic.

  • Causal inference. It is the broader analytic task.
  • Evidence synthesis. It supplies the integrative activity.
  • No strict parent is asserted.

Neighborhood in Abstraction Space

Bradford Hill criteria sits in a moderately populated region (45th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Empirical Measurement & Statistical Inference Methods (50 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With

  • Causal checklist. Tell: Are the nine viewpoints being mechanically counted?
  • Statistical significance. Tell: Is association strength being mistaken for causation?
  • Evidence hierarchy. Tell: Are study types ranked instead of integrated?
  • Causal proof. Tell: Is uncertainty falsely eliminated?

References

  • Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Bradford_Hill_criteria (revision 1365729422).
  • Preserved source candidate: https://pmc.ncbi.nlm.nih.gov/articles/PMC4291332/
  • Preserved source candidate: https://priceonomics.com/why-the-father-of-modern-statistics-didnt-believe/
  • Preserved source candidate: https://web.archive.org/web/20251206211553/https://priceonomics.com/why-the-father-of-modern-statistics-didnt-believe/
  • Preserved source candidate: http://journals.lww.com/00008480-200302000-00017
  • Preserved source candidate: http://qualitysafety.bmj.com/content/early/2017/08/02/bmjqs-2017-006756.full
  • Preserved source candidate: http://www.aicr.org/continuous-update-project/

The frozen Wikipedia revision is discovery provenance. The retained source set was reviewed for identity, formal or operational relation, and scope. The encyclopedia's structural synthesis is bounded to those claims; a thin authority surface is recorded as a nonblocking source-strengthening repair rather than concealed.