Evidence

Evidence is what makes a statement indubitable, through direct observation or through compelling inference. It is the ground on which a belief may be held. Everything else on this page concerns how that ground is tested.

Epistemology examines the conditions of knowledge, how it is acquired and where its limits lie. It is the field that asks what would have to be true for a claim to count as known. It is also the field that search systems now operate in without noticing.

The theory of knowledge concerns the foundations, limits and validity of human knowing. It supplies the standards the rest of this page applies. It is older than every measurement described here.

Justification is the epistemic ground that turns a mere belief into knowledge. Without it a true belief is an accident. It is the component that evidence supplies.

A belief is a cognitive state of a person, whose truth is decided by reasons and evidence rather than by conviction. Holding something firmly has no bearing on whether it is so. The distinction is the whole of the subject.

Certainty is a state produced by immediate evidence or compelling proof. It is a property of the knower rather than of the claim. Certainty and correctness come apart regularly.

Truth is a necessary condition of knowledge and a predicate that marks statements. A false statement cannot be known, however well justified. That is the constraint everything else works under.

Empirical evidence rests on systematic observation and data collection. It is the form of evidence that can be gathered rather than seen. Most of what is disputed publicly is disputed at this level.

Direct observation serves in philosophy as a source of immediate insight. It appears to need no argument. It is also where theory-ladenness enters, which is treated below.

Sense experience is, for empiricist positions, the basic instance that turns the world into knowledge. It is the entry point of evidence. Its reliability is exactly what scepticism questions.

Scepticism questions the underlying system of evidence and doubts that the standards hold. It is the permanent opposition in this field. It is also the reason the standards were formalised at all.

Compelling inference produces mediated evidence from existing statements. It extends what is known beyond what is observed. Its validity is the subject of the science-theory pages in this field.

A randomised controlled trial is a design that assigns participants by chance, in order to remove systematic differences between groups. It is treated as the most reliable single study type. It is the instrument that made evidence rankable.

A systematic review gathers the available external evidence on a clearly bounded question, by an explicit procedure. It is a study of studies. It is the step at which individual results become a state of knowledge.

Clinical expertise is the judgment acquired through experience. It is the component that applies general evidence to a particular case. It is also where evidence-based practice is most often misunderstood as its opposite.

A gold standard is the most reliable available benchmark for quality. It sets the level against which weaker evidence is measured. It exists because somebody built a ranking and defended it.

It treats evidence as a positive condition: evidence justifies, justification makes knowledge. It leaves out everything that defeats, degrades or distorts evidence, everything probabilistic, and the entire institutional apparatus that decides in practice which evidence counts.

A defeater is a consideration that removes a justification without disproving the claim. Texts cover only the positive condition. Half of practical reasoning consists of defeaters, and none of it is represented.

Pragmatic encroachment is the view that how much evidence you need depends on what turns on being wrong. Texts treat evidential requirements as fixed. In practice everybody already applies it and no account of it is offered.

Epistemic luck is arriving at a true belief by accident. Texts focus on the correlation of evidence and belief. Without the concept there is no way to separate a reliable method from a fortunate one.

Degree of belief treats conviction as a quantity rather than as present or absent. Texts model belief in binary form. Every real judgment is graded, and the binary model cannot represent it.

Epistemic risk weighs the cost of error against the cost of caution. Sources treat evidence as static proof without evaluating what being wrong costs. The cost is what decides how much evidence is enough.

Peer disagreement is the case of two equally competent people reaching different conclusions from the same evidence. Literature handles evidence for a single subject in isolation. The disagreement case is the one that arises in public.

Bayesian epistemology models belief as probability and updates it with evidence. Analysis follows qualitative models of justification. The formal treatment is the one that actually says how much a piece of evidence should move you.

A prior probability is what you believed before the observation. Sources treat evidence separately from background assumptions. The same evidence moves two people to different conclusions because their priors differ, and that is not an error.

Underdetermination is the situation where the same data support several incompatible theories. Texts assume evidence maps onto one explanation. Where it does not, further data does not settle the matter.

Falsifiability is the property of a claim that some observation could refute it. The literature examined focuses on verification and confirmation. A claim that nothing could refute is not supported by the evidence that appears to support it.

The Duhem-Quine thesis holds that hypotheses are never tested alone, but always together with auxiliary assumptions. Texts test hypotheses in isolation. When a test fails, what failed is not identified by the test.

A likelihood ratio expresses how much more probable the observed data are under one hypothesis than under another. Evidence is handled verbally, without the number. The number is what distinguishes strong evidence from a suggestive anecdote.

Preregistration fixes the hypothesis and the analysis before the data are seen. Sources describe collection methods without the timing. Without it, a confirmed prediction and a post-hoc explanation are indistinguishable in the report.

Publication bias is the tendency for positive findings to be published and null findings to remain unpublished. Texts equate the available literature with the whole of the research. What is visible is therefore not a sample of what was done.

P-hacking is the selective analysis that produces a significant result from data that do not contain one. Sources treat analysis as a neutral step. It is a set of decisions, each defensible, that together manufacture a finding.

A replication study repeats an investigation independently. Texts consider primary evidence generation in single studies. A result that has never been repeated is a hypothesis with a number attached.

An evidence hierarchy ranks study designs by how much confidence they warrant. Sources name study types without ordering them. The ordering is what makes evidence usable by somebody who is not a specialist.

A confounder is a third factor that produces an apparent relation between two others. Sources assume a direct causal relation between observation and conclusion. Most spurious findings are confounding rather than fraud.

Triangulation combines several independent methods on the same question. Texts rest on single sources of evidence. Agreement across methods that fail differently is worth more than repetition within one.

A meta-analysis pools results from several studies quantitatively. Texts name narrative and qualitative syntheses. The quantitative version produces an estimate with a stated uncertainty, which narrative synthesis cannot.

GRADE is a standardised scheme for rating the certainty of evidence and the strength of recommendations. Sources assess evidence without such a scheme. It exists precisely so that assessment does not depend on who is assessing.

Cochrane is the international body that produces systematic reviews to a fixed methodology. Texts focus on theoretical concepts without reference to the institutions. The institutions are what turned evidence-based practice from a principle into a routine.

A data monitoring committee reviews accumulating trial data independently of the investigators. Analysis considers how evidence arises without the organisational check. The check exists because investigators cannot judge their own trial mid-flight.

CONSORT is the reporting standard for randomised trials, specifying what must appear in the publication. Texts mention study formats without reporting standards. What is not reported cannot be assessed, whatever the study did.

Medicine built a way to rank evidence, a scheme for rating certainty, institutions that apply it, and reporting standards that make assessment possible. The web ranks by links and mentions. Every element above exists, is documented, and has never been carried across.

State what a claim rests on, name the kind of evidence it is, and say what would change your mind. That is the portable core of the whole apparatus. It costs nothing and it is what distinguishes a source that can be checked from one that can only be believed.