You walk into the kitchen.
The floor is wet.
What does that prove?
Almost nothing by itself.
Maybe somebody mopped.
Maybe a pipe leaked.
Maybe a glass broke.
Maybe rain blew through an open window.
The wet floor is an observation.
It becomes evidence only in relation to a question.
Quick Read
Evidence is information that changes the relative plausibility of competing claims, hypotheses or states.
This means evidence is not merely “a fact.”
A fact can be irrelevant to the question.
A useful observation becomes evidence because it is more expected under one hypothesis than another, or because it directly bears on the claim being tested.
One-sentence answer: Evidence is information that should rationally change how much weight we give to one explanation, prediction or claim relative to alternatives.
Evidence Is Relational
A fingerprint on a glass is evidence of what?
That someone touched the glass.
It may support a claim that a person was present.
It does not automatically prove when they touched it, why they were there or what they did afterward.
The same observation can be strong evidence for one proposition and weak evidence for another.
This is why “there is evidence” is incomplete.
Evidence for what?
Observation Is Not Explanation
The student’s marks fell.
Observation.
The student became lazy.
Explanation.
The evidence does not contain the explanation inside it.
Marks can fall because of harder content, poor sleep, illness, missing prerequisites, weak time management, marking differences or random variation.
Evidence constrains explanation.
It does not eliminate the need for explanation.
A Good Piece of Evidence Discriminates
Suppose two hypotheses explain a fever.
If a test result is equally likely under both, it does little to separate them.
If a result is expected under one and unlikely under the other, it is more discriminating.
The same logic applies in education.
Two possible student states:
- does not understand the concept,
- understands but cannot retrieve under time pressure.
Ask a slow, untimed explanation question.
If the student explains accurately but fails a timed item, the evidence shifts weight toward retrieval or execution difficulty.
Good evidence narrows the state space.
Evidence Has Direction
Evidence can support.
Evidence can weaken.
Evidence can leave a belief almost unchanged.
This matters because people often treat every observation connected to a topic as supportive.
A theory predicts rain.
It rains.
Support?
Perhaps.
But if every rival model predicted rain too, the observation does little discriminative work.
Evidence Strength Depends on Alternatives
A footprint outside your house is evidence that somebody walked there.
Is it evidence that a specific neighbour did?
Only if the footprint has features more compatible with that neighbour than alternatives.
Evidence cannot be evaluated responsibly without a comparison class.
This links directly to the Cognitive Art article on comparison.
Evidence says:
This observation fits some models better than others.
Evidence Accumulates
Many decisions are not made from one decisive observation.
Evidence arrives over time.
A blurry shape.
Then movement.
Then colour.
Eventually you decide:
car.
Sequential-sampling models in cognitive science formalise versions of this process by modelling noisy evidence accumulation toward a decision threshold.
A major Annual Review explains how diffusion decision models decompose choices into components such as information quality, response caution and prior bias.
The teaching point is broader than any one model:
decisions can be understood as evidence accumulating until action becomes justified.
Fast Decisions Use Lower Evidence Thresholds
Imagine identifying whether a light is red or green while driving.
You cannot wait for perfect certainty.
Speed matters.
Now imagine deciding whether to sign a thirty-year financial contract.
The evidence threshold should be different.
Decision systems trade speed against accuracy.
Evidence sufficiency depends on stakes.
One Observation Can Be Stronger Than Ten Weak Ones
Quantity of evidence is not the same as quality.
Ten people repeating the same rumour may still represent one original source.
Ten measurements using the same miscalibrated instrument do not create ten independent confirmations.
Evidence strength depends on reliability and independence.
The Independence Problem
Imagine five news sites reporting the same claim.
Looks like five confirmations.
Now discover all five copied one press release.
The evidence count collapses.
Correlated evidence should not be counted as independent support.
This principle matters in journalism, science, intelligence, social media and everyday gossip.
Evidence Has Provenance
Where did the observation come from?
Who measured it?
Using what instrument?
Under what conditions?
Was the record transformed before reaching you?
Evidence without provenance is difficult to calibrate.
A screenshot can be authentic and misleading if cropped.
A quote can be exact and deceptive if context is removed.
A graph can use real numbers and distort through axis choice.
Provenance is part of evidence quality.
Evidence Can Be Direct or Indirect
You see rain falling.
Direct evidence of rain where you are.
You see wet umbrellas entering the building.
Indirect evidence.
Indirect evidence can still be strong.
Much of science studies entities and events that are not directly observed with unaided senses.
The important question is not whether evidence is “direct.”
It is how securely the observation connects to the claim.
Absence Can Be Evidence
You expect smoke if a fire is burning.
No smoke.
Does that weaken the fire hypothesis?
Only if smoke should have been detectable under the conditions.
“No evidence” becomes evidence only when the search had reasonable power to find what the hypothesis predicts.
Absence of evidence is not automatically evidence of absence.
Negative Evidence Requires an Expected Signal
If a planet should cause a measurable wobble in a star and no wobble is observed despite sensitive measurement, that non-observation can constrain the planet hypothesis.
If a student “never asks questions” in a class where asking questions is socially risky, silence is weak evidence of understanding.
The expected signal must be defined before absence becomes informative.
Evidence and Prior Belief
The same observation can rationally produce different final beliefs when prior information differs.
A rare disease test comes back positive.
The meaning of that result depends partly on how common the disease was before the test and how accurate the test is.
This is the intuition behind Bayesian reasoning.
Evidence updates a prior; it does not appear in a vacuum.
But Priors Must Be Revisable
A prior is useful because the world has history.
It becomes dangerous when no amount of evidence can move it.
If every contradiction is reinterpreted as proof of the theory, the representation becomes self-sealing.
Evidence must be allowed to change belief.
Otherwise “evidence” is only decoration around a fixed conclusion.
Belief Updating Should Match Evidence Strength
Weak evidence should move belief a little.
Strong evidence should move belief more.
Converging independent evidence should move it further.
One anecdote should rarely reverse a conclusion supported by thousands of observations.
Likewise, one old belief should not survive overwhelming contradictory evidence merely because it arrived first.
The Vivid Evidence Trap
A personal story is vivid.
A statistical distribution is abstract.
The story can dominate attention even when the statistical evidence is stronger.
This is where salience can distort evidence weighting.
Emotional vividness is not evidential weight.
Anecdote Is Evidence of a Case
“This method worked for my child.”
That is evidence that the method and the outcome co-occurred in one case.
It does not by itself establish:
- causality,
- general effectiveness,
- which component mattered,
- whether the result would have happened anyway.
Anecdotes matter.
They should be assigned the weight their design supports.
Experimental Evidence Creates Better Counterfactuals
A treatment group improves.
What would have happened without treatment?
We cannot observe both worlds for the same person at the same time.
Experimental design creates comparison conditions that approximate the missing counterfactual.
This is why randomisation, controls and blinding can strengthen causal evidence.
Design quality changes evidential meaning.
Evidence Can Be Reliable but Irrelevant
The thermometer is perfectly calibrated.
You are trying to explain why a student failed an essay.
Accurate temperature data may have zero relevance.
Evidence quality has at least two dimensions:
- Is the observation trustworthy?
- Does it bear on the question?
Reliable irrelevance is still irrelevance.
Evidence Can Be Relevant but Unreliable
A witness saw the event.
Highly relevant.
But visibility was poor and the memory was reported weeks later.
Relevant evidence must still be calibrated for reliability.
Evidence and Measurement
Measurements are representations.
A scale reading represents mass under a procedure.
An examination mark represents performance on sampled tasks under conditions.
A survey score represents responses to chosen items.
Measurement error enters evidence through the representation.
A precise-looking number can carry uncertain evidential weight.
Evidence in Mathematics
Mathematics distinguishes example from proof sharply.
Test a conjecture on 1,000 cases.
Strong empirical support.
Not a proof for infinitely many cases.
One counterexample can destroy a universal claim.
This teaches a profound evidential lesson:
the amount of evidence needed depends on the logical form of the claim.
Evidence in English
“The character is selfish.”
What evidence?
A quotation is not enough by itself.
The student must explain why the quoted action or language supports the claim.
Evidence requires a bridge from observation to inference.
That bridge is reasoning.
Evidence in History
A diary is evidence.
Of what?
It may be excellent evidence of what the writer claimed, feared, observed or wanted to record.
It may be weaker evidence of what every person in the society believed.
Source type and question must align.
Evidence in Science
Science strengthens evidence through design, replication, measurement and prediction.
The strongest evidence is rarely one dramatic experiment standing alone.
It is convergence across methods with different failure modes.
Different instruments.
Different laboratories.
Different samples.
Different predictions.
Convergence matters because shared error becomes less likely.
Evidence in Education
Does the student understand?
One correct answer is weak evidence.
Add:
- explain why,
- solve a new version,
- solve after delay,
- identify a wrong solution,
- apply under time pressure.
Now the evidence becomes richer because different tests probe different failure modes.
Understanding is a hidden state.
Assessment collects evidence about it.
The Confirmation Trap
Once we prefer a hypothesis, we notice supporting evidence easily.
Contradictory evidence becomes inconvenient.
A disciplined evidence practice asks:
- What did my hypothesis predict before I observed the result?
- What result would have counted against it?
- Did I search equally hard for alternatives?
Evidence should test a model, not decorate it.
The Moving-Goalpost Problem
A prediction fails.
The criterion is quietly changed afterward.
Now the model “succeeded.”
This destroys evidential discipline because no observation can falsify the claim.
Where possible, define evidential criteria before the result arrives.
Evidence and Uncertainty
Evidence should move uncertainty.
Sometimes toward one hypothesis.
Sometimes away.
Sometimes by revealing that the model space was incomplete.
Good evidence need not end uncertainty.
It should reshape it.
Evidence and Error
An error is evidence.
If prediction and outcome differ, the difference contains information about the model.
But not every error identifies its cause.
A wrong answer tells us the representation failed somewhere.
Further evidence is needed to locate the failure.
Evidence and Confidence
Confidence should be a summary of evidential support relative to alternatives.
But human confidence can depend on familiarity, fluency, emotion and prior expectations as well as evidence.
This is why confidence must be calibrated against actual outcomes.
The Evidence Audit
- What claim is this evidence meant to bear on?
- What exactly was observed?
- How reliable is the observation?
- What alternatives also predict it?
- How discriminating is it?
- Are multiple pieces genuinely independent?
- What evidence is missing?
- What result would weaken my preferred explanation?
- How much should this observation change belief?
- What next observation would be most informative?
A Practical Exercise: Observation, Inference, Claim
Take one argument.
Make three columns:
- observation,
- inference,
- claim.
Place every sentence where it belongs.
Many arguments become weaker immediately because inference had been disguised as observation.
A Practical Exercise: Find the Discriminating Test
Write two plausible explanations for one problem.
Now ask:
What observation would be likely under one and unlikely under the other?
That is often the best next question.
A Primary-to-Adult Progression in Evidence
Primary: show me why
Children learn that answers should point to observations, text clues, measurements or examples.
Lower secondary: distinguish evidence from explanation
Students should separate data, quotation, inference and conclusion while comparing alternative explanations.
Upper secondary: evaluate evidence quality
Learners should examine source reliability, controls, sample size, measurement error, provenance and independence.
Adulthood: weight evidence rather than count it
Professional reasoning needs calibrated updating across mixed-quality observations rather than simple accumulation of supporting anecdotes.
Five Evidence Failures
1. Evidence as Fact
A fact is treated as evidentially decisive without specifying the claim it supports.
2. Evidence Counting
Ten weak or dependent sources are treated as ten strong independent confirmations.
3. Vividness Weighting
Memorable anecdotes outweigh stronger but less vivid evidence.
4. Missing Counterevidence
The search process retrieves only supporting observations.
5. Self-Sealing Interpretation
Every possible result is reinterpreted as support, making genuine testing impossible.
Frequently Asked Questions
Is evidence the same as data?
No. Data are recorded observations. They become evidence relative to a claim, model or hypothesis when their relationship to alternatives is analysed.
Can one observation be strong evidence?
Yes, if it is highly reliable and strongly discriminates among alternatives. A single counterexample can refute a universal mathematical claim.
Why does independence matter?
Because repeated reports derived from one source do not provide the same confirmation as genuinely independent observations with different failure modes.
Is absence of evidence evidence of absence?
Only when the hypothesis predicts an observable signal and the search had enough power to detect it.
How does prior belief affect evidence?
Prior information affects the starting plausibility of hypotheses. Evidence then updates those plausibilities. Priors should guide interpretation without becoming immune to revision.
What is evidence accumulation?
It is the process of integrating noisy information over time until enough support exists to make a decision. Sequential-sampling models formalise several versions of this process.
Research Notes and Further Reading
For formal models of evidence accumulation, see Forstmann, Ratcliff and Wagenmakers, Sequential Sampling Models in Cognitive Neuroscience. The review explains how noisy evidence can accumulate toward a response threshold and how model parameters distinguish information quality, caution and prior bias.
For human perceptual decision formation, see Neurophysiology of Human Perceptual Decision-Making. For the broader relation between judgement, uncertainty and decision quality, see Judgment and Decision Making.
The Cognitive Art article uses evidence broadly across science, education, reading and everyday inference. Different disciplines formalise evidential support differently; the common question is how an observation should change belief relative to alternatives.
Final Thought: Evidence Is What Moves a Mind That Is Willing to Move
The wet floor does not speak.
It does not announce:
pipe leak.
It enters a competition among explanations.
Then another observation arrives.
The pipe is dripping.
Belief shifts.
That movement is the point.
Evidence is not a trophy collected after we already know the answer.
Evidence is a force acting on uncertainty.
It should narrow some possibilities.
Widen others.
And occasionally force us to admit that the whole question was framed badly.
A mind that never changes after evidence is not strong.
It is sealed.
The real strength is simpler:
see what happened, ask what it means, compare the alternatives, and move belief by the amount the evidence deserves.