How Science Works | Evidence, Models, Testing and Correction

Science works by keeping our explanations answerable to the world.

Science is not mainly a collection of facts, a school subject, a laboratory coat, or a fixed five-step procedure. It is a human system for asking questions about the world, building explanations and models, checking those explanations against evidence, estimating how uncertain we should remain, and changing our minds when reality does not behave as expected.

That makes Science powerful for a very particular reason: a scientific idea is supposed to remain vulnerable to being checked. A claim may be clever, popular, elegant, profitable or supported by an expert, but none of those things removes its obligation to answer to evidence.

Science is the organised human system for building explanations of the world, making those explanations vulnerable to evidence, and correcting them when better evidence requires it.

Quick Read: The Whole Science Loop

A useful high-level map is:

WORLD → NOTICE → QUESTION → OBSERVE → MEASURE → REPRESENT → MODEL → PREDICT → INVESTIGATE → DATA → ANALYSE → ESTIMATE UNCERTAINTY → CRITIQUE → CROSS-CHECK → EXPLAIN → APPLY → WORLD RETURN → REVISE

This is a map, not a compulsory recipe. Real Science can enter the loop at different points. A strange observation may create a question. A model may predict something nobody has measured yet. A new instrument may reveal data that force an old theory to change. A failed replication may expose an overlooked condition. A practical problem may send researchers back to basic Science.

The deeper rule is simpler: the representation must keep returning to the world for checking.

Science Is Not the World

The world exists whether our explanation is good or bad. Science is the human effort to build increasingly useful and reliable representations of that world.

This distinction matters:

Reality is what is actually happening.
An observation is something detected or recorded about it.
A measurement is an observation produced through a defined measurement procedure and expressed with an appropriate quantity, category or scale.
Data are recorded outputs that can be organised and analysed.
A representation is a graph, diagram, equation, image, classification, description or other way of encoding part of the system.
A model is a purposeful simplification used to explain, predict, compare or reason about the system.
A hypothesis is a claim or proposed explanation that can be confronted with evidence.
A theory is a broader explanatory framework that connects many findings and has survived substantial testing; the exact use of the word varies by discipline.
Evidence is information that bears on whether a claim should become more or less credible.
A conclusion is an inference made from evidence under stated assumptions and limitations.

The mistake is to collapse these layers. A graph is not the phenomenon. A measurement is not automatically an explanation. A statistical result is not automatically a causal mechanism. A model is not reality. A published conclusion is not permanent truth.

This is why How Evidence Works, How Reasoning Works, How Knowledge Works and How Information Works sit naturally beside Science. Science depends on all four, but gives them a special constraint: the claim must remain connected to empirical reality.

There Is No Single Scientific Method

School diagrams sometimes make Science look like one universal sequence: question → hypothesis → experiment → conclusion. That sequence can be useful for teaching a controlled investigation, but it is too narrow to describe Science as a whole.

Different scientific questions permit different kinds of evidence. An astronomer cannot move a galaxy into a laboratory. A geologist may reconstruct a past event from rocks and structures that survive today. An ecologist may combine field observation, experiments and models. A physicist may create tightly controlled experiments. A climate scientist may combine observations, physical theory, historical records and simulations. A biologist may compare organisms, manipulate variables, sequence genomes or follow populations through time. A computational scientist may test the consequences of a formal model before those consequences can be measured directly.

Modern philosophy of Science therefore treats scientific method as a family of practices rather than one rigid algorithm. Systematic observation, experimentation, model-building, mathematical and computational reasoning, hypothesis testing, comparison, historical reconstruction, measurement, statistical inference and critical discussion can all be scientific when they are used appropriately for the question.

The common contract is not “everyone performs the same steps.” It is closer to:

make the question clear enough to investigate;
make the evidence traceable;
make the reasoning inspectable;
make uncertainty visible;
make competing explanations possible to consider;
make the claim vulnerable to further checking;
change the claim when the evidence no longer supports it.

The National Academies describes Science as a mode of inquiry in which questions about the world are answered and their degree of certainty assessed through a communal effort intended to keep them well grounded. That is a much better picture of real Science than a single worksheet flowchart.

1. Science Begins With a World That Can Push Back

The first source of scientific discipline is not the scientist. It is the world.

If a child predicts that a larger shadow must always mean a larger object, the world can contradict that idea when distance from the light source changes. If a chemist predicts a reaction will produce a particular spectrum, the instrument may show otherwise. If an astronomical model predicts an orbit incorrectly, repeated observations can expose the mismatch.

Science becomes useful when disagreement between expectation and observation is treated as information rather than embarrassment.

Expected result ≠ observed result is not automatically failure. It may be the beginning of better Science.

2. Observation Is More Difficult Than “Just Looking”

Humans do not receive reality as an untouched stream. We choose what to notice, which instrument to use, what scale to record, what counts as a signal, and how to classify the result. Scientific observation therefore becomes stronger when those choices are explicit and checkable.

Modern Science also observes far beyond the unaided senses. Telescopes convert radiation into data. Microscopes enlarge structures. Sequencers transform biological material into digital records. Detectors register particles that a person cannot directly see. Satellites sample planetary systems. Sensors turn temperature, pressure, acceleration, voltage, light or chemical concentration into records.

This means every instrument introduces another important question: does this device actually measure what we think it measures, with enough accuracy and precision for the claim we want to make?

3. Measurement Is a Modelled Relationship, Not a Magic Number

A measurement result is never improved merely by writing more decimal places. Good measurement requires an appropriate quantity, a defined procedure, calibration, suitable reference standards where relevant, attention to instrument behaviour, and an estimate of uncertainty.

NIST’s measurement guidance treats uncertainty as part of the measurement result rather than an optional apology added afterwards. This matters because instruments have limits, samples vary, environments change, calibration can drift, and the model connecting an instrument reading to the quantity of interest may itself be imperfect.

So “we measured 10” is incomplete Science if the important question is whether the underlying quantity could plausibly be 9.9, 10.0, 10.3 or something much wider.

A useful rule is:

Measurement turns a phenomenon into evidence only when the measurement process is trustworthy enough for the inference being made.

The laboratory version of this discipline is developed further in Controls, Blanks, Standards and Calibration | How Experiments Check Themselves.

4. Models Let Humans Think About What They Cannot Hold in Their Heads

Models are among Science’s most important tools. A model can be physical, verbal, graphical, mathematical or computational. It can represent an atom, a population, a cell membrane, a climate system, a food web, a disease process, a force field or a star.

A useful model is not a miniature copy of everything. It leaves things out deliberately so that selected relationships become easier to reason about.

That creates two permanent questions:

What does this model represent well?
Where does this model stop being reliable?

A particle model may explain some properties of matter brilliantly without showing every feature of an actual material. A population model may reveal growth dynamics while simplifying individual behaviour. A climate model may integrate physical processes at planetary scale while still having uncertainties in particular parameters or local outcomes.

Scientific maturity therefore includes knowing when a model is useful and when it should not be mistaken for the thing itself.

5. Predictions Make Explanations Risk Something

One powerful way to test an explanation is to ask what should happen if it is right.

A model that can accommodate every imaginable outcome after the fact tells us less than a model that makes constrained predictions. Scientific predictions do not have to concern the future. A theory can predict what should already exist in an unexplored layer of rock, what pattern should appear in archived data, or what signature a telescope should detect.

The important structure is:

If this explanation is approximately right under these conditions, what evidence should we expect—and what evidence would make us reduce our confidence?

6. Investigation Creates a Fairer Contest Between Explanations

An experiment is valuable because controlled manipulation can help separate possible causes. But controlled experiments are only one route. Scientific investigations also include longitudinal studies, field studies, natural experiments, comparative studies, observational surveys, archaeological and geological reconstruction, instrument campaigns, computational experiments and many other designs.

Good design asks whether the evidence can genuinely distinguish between explanations. Depending on the field, that may involve controls, comparison groups, randomisation, blinding, repeated measurements, standard procedures, preregistration, independent coding, negative controls, positive controls, calibration checks, sensitivity analyses or multiple independent methods.

Not every method fits every question. The scientific skill is choosing a method whose strengths and limitations match the claim.

7. Data Do Not Interpret Themselves

After data are collected, another scientific layer begins: analysis.

Researchers may classify observations, compare groups, estimate parameters, fit models, calculate uncertainty, test predictions, detect patterns, search for anomalies or combine results from many studies. Mathematics becomes essential here because it provides structures for preserving relationships, quantifying variation and testing whether an inference is warranted.

This is where How Mathematics Works connects directly to Science. Mathematics can tell us what follows from assumptions and data under a formal model. Science must additionally ask whether those assumptions, variables and measurements map the world well enough for the conclusion to travel back into reality.

A statistically detectable relationship is also not automatically a causal explanation. Confounding variables, selection effects, measurement error, reverse causation and model assumptions can all create misleading interpretations. Scientific reasoning therefore keeps the chain visible:

data → analysis → inference → claim

Each arrow can fail.

8. Uncertainty Is Part of the Answer

Science rarely delivers absolute certainty about empirical claims. That is not a defect. It is a feature of taking measurement, variability, sampling and incomplete knowledge seriously.

Uncertainty can come from different places: measurement limits, natural variability, limited sample size, incomplete models, unknown parameters, competing explanations, imperfect data or conditions that may not generalise.

Good Science tries to say not only what we think but how strongly the evidence supports it, under what conditions, and what could still change the conclusion.

This is why “scientists are uncertain” does not mean “scientists know nothing.” A forecast can be useful without being exact. A treatment effect can be supported without being identical for every person. A model can be reliable within a range without being perfect outside it.

Uncertainty becomes a problem when it is hidden, misunderstood or used inconsistently—not when it is reported honestly.

9. One Study Is Usually the Beginning, Not the End

Scientific confidence is rarely built by one result alone. It grows through a network of evidence.

The National Academies makes useful distinctions:

Reproducibility can mean obtaining consistent computational results from the same data, code and analysis steps.
Replicability means asking the same scientific question with new data and seeing whether compatible findings emerge.
Generalisability asks whether a result travels to different populations, environments, scales or conditions.

These are related but not interchangeable. A result may be computationally reproducible but fail to generalise. A replication may differ because the original effect was fragile, because the second study is poor, or because a previously unknown condition changes the phenomenon. A single failed replication is therefore not automatically the last word either.

Across many sciences, confidence is better pictured as a web of converging constraints: different instruments, methods, teams, populations, observations and predictions point toward the same underlying explanation.

10. Peer Review Is a Checkpoint, Not a Truth Machine

Scientific communities add another layer of checking because individual researchers have limited knowledge and can make mistakes. Methods are criticised. Statistics are questioned. Alternative explanations are proposed. Data may be reanalysed. Other teams attempt related studies. Reviews combine larger bodies of work.

Peer review is one part of this system. It can improve manuscripts and filter obvious weaknesses, but publication does not certify a result as permanently correct. Reviewers can miss errors. Journals can reward novelty. Fields can develop shared blind spots. Important negative results may be harder to publish. Incentives can distort behaviour.

This is why open methods, data where ethically and legally appropriate, code, preregistration, registered reports, replication, post-publication criticism and correction of the research record can matter. UNESCO’s Recommendation on Open Science explicitly links transparency, scrutiny, critique and reproducibility with stronger and more reliable scientific knowledge.

Publication is not the end of scientific checking. It is the point where checking can widen.

11. Science Is Correction-Capable, Not Magically Self-Correcting

It is common to say that Science is self-correcting. The useful part of that statement is that scientific systems contain mechanisms that can expose error: new measurements, competing theories, replication, better statistics, improved instruments, reanalysis, criticism and accumulated contradictory evidence.

But correction is not automatic. It can be delayed by prestige, incentives, group loyalty, poor methods, weak reporting, publication bias, unavailable data, fraud, inadequate replication, lack of funding, conflicts of interest or simply the absence of technology needed to test an idea properly.

Research integrity therefore matters structurally. Fabrication and falsification directly corrupt the connection between claim and world. Sloppy records make findings hard to inspect. Weak design can produce confident-looking answers to the wrong question. Selective reporting can make an uncertain literature look stronger than it is.

So a more precise sentence is:

Science works best when its institutions keep error visible enough, evidence open enough, incentives healthy enough and criticism safe enough for correction to occur.

12. Scientific Consensus Is Not a Vote on Reality

Scientific consensus is often misunderstood. It does not mean that truth is created by majority vote. It means that, after a body of evidence has accumulated and been scrutinised, many relevant specialists may converge on an explanation because it currently accounts for the evidence better than available alternatives.

Consensus should therefore be read together with the strength, diversity and maturity of the supporting evidence. A mature conclusion supported by multiple independent methods is different from a new interpretation based on a small literature.

Experts matter because expertise changes what a person can notice, measure, compare and infer. But expertise does not abolish the evidence requirement. Authority can guide where a non-specialist should begin; it should not make a scientific claim immune from reality.

13. Science Becomes Stronger When Different Methods Converge

Some of the strongest scientific conclusions are difficult to reduce to one decisive experiment. Instead, many independent lines of evidence converge.

An explanation may be supported by laboratory results, field observations, mathematical models, different instruments, comparative evidence and successful predictions. Each method has different weaknesses. When methods with different failure modes point toward the same explanation, confidence can grow because one hidden defect is less likely to account for everything.

This is also why systematic reviews and evidence synthesis matter. Rather than selecting one dramatic study, they use explicit methods to identify, appraise and combine a larger body of research. In suitable cases, meta-analysis can estimate an overall effect across multiple studies. The synthesis is still only as good as its included evidence and methods, but it helps move Science from isolated papers toward cumulative knowledge.

14. Science, Mathematics, Engineering and Technology Are Connected but Not Identical

Mathematics can establish what follows necessarily inside a formal system from definitions, assumptions and valid transformations. Science asks whether a representation or explanation remains adequate to empirical reality. Engineering uses knowledge to design systems that meet goals under constraints. Technology extends human capability through tools, processes and systems.

In practice they overlap constantly. Science depends on mathematics, instruments and engineering. Engineering depends on scientific knowledge. New technology creates new observations. New observations create new Science. The telescope changed astronomy; microscopy changed biology; computation changed almost every scientific discipline.

For the larger eduKateSG mechanism map, continue through How Mathematics Works, How Technology Works and How Innovation Works.

15. Science Cannot Decide Every Human Question

Science can tell us a great deal about what happens, what has happened, what mechanisms are plausible, what an intervention tends to cause under specified conditions, what risks are measurable, and what consequences are likely.

But evidence alone cannot decide every question about what humans ought to do.

Choosing a policy can involve values, rights, fairness, cost, risk tolerance, dignity, culture, law, feasibility and competing legitimate interests. WHO’s work on evidence-informed decision-making and research ethics makes this boundary explicit: evidence should inform decisions, but value judgements and ethical obligations remain part of the decision process.

Science can estimate the likely consequences of different actions. It cannot, by itself, determine whose interests must take priority or what a society should regard as just.

This boundary protects Science from being asked to supply moral authority it does not possess, while also protecting ethical and policy decisions from pretending that evidence does not matter.

16. Science Has a Culture Because Humans Do It Together

Science is a knowledge system, but it is also a human institution. People must decide which questions are worth funding, which measurements count as adequate, how results are communicated, who receives credit, which data can be shared, how risks to participants are controlled, and how disagreements are handled.

Culture therefore affects Science without turning scientific claims into mere cultural preference. Norms of honesty, criticism, recordkeeping, openness, intellectual humility and willingness to revise help the evidence-correction system work. Norms that reward only spectacular positive results can distort it.

The wider social mechanism is explored in How Culture Works. The language required to name distinctions, communicate methods and preserve meaning connects to How Vocabulary Really Works.

How Science Grows From Primary School to Research

School Science is one training corridor into the much larger scientific enterprise. The underlying relationship with the world stays recognisable, but the resolution increases.

Stage	Main development	Typical Science capability
Primary	Notice patterns and explain observable phenomena	Observe → model → explain → test → transfer
Secondary	Use stronger representations, measurement and disciplinary models	Observe → model → represent → measure → predict → test → evaluate → transfer
Junior College	Formalise, quantify and integrate Biology, Chemistry and Physics reasoning	Observe → model → formalise → quantify → predict → investigate → analyse → evaluate → integrate → transfer
University / Research	Work at the edge of established knowledge	Specialise → formulate questions → choose methods → generate evidence → analyse uncertainty → critique → publish → replicate → synthesise → revise
Adult / Citizen / Professional	Use scientific evidence responsibly in real decisions	Identify claim → inspect evidence → judge confidence → apply within limits → observe consequences → update

These are not separate definitions of Science. They are changing levels of resolution inside the same evidence-correction relationship.

Continue through the current eduKate Science corridor:

What Parents, Tutors and Teachers Should Protect

If a child learns Science only as a vocabulary-and-answer system, the student may score on familiar questions yet struggle when the surface changes. The deeper educational job is to preserve the chain between world, evidence, model and explanation.

A useful teaching conversation repeatedly asks:

What exactly did we observe?
What are we inferring rather than directly observing?
What model are we using?
What evidence supports this explanation?
What else could produce the same observation?
What variable or condition matters?
What would we expect if the explanation were right?
How could we test or check it?
How certain should we be?
Would the conclusion still hold in a changed context?

Vocabulary still matters enormously because scientific reasoning needs precise language. Mathematics matters because quantity, change, probability and uncertainty need structure. Practical work matters because students learn that evidence has to be produced carefully, not merely quoted. But all three should serve the same central job: help the learner build an explanation that remains answerable to evidence.

Where Science Commonly Breaks

Failure	What goes wrong	Repair question
Observation becomes assumption	Inference is reported as if directly observed	What did we actually detect or record?
Measurement without calibration	A number is trusted without checking the instrument or method	How was this measurement validated and what is its uncertainty?
Model becomes reality	A useful simplification is treated as complete	What does the model omit and where does it fail?
Correlation becomes cause	Association is mistaken for mechanism	What alternative causes or confounders remain?
One study becomes certainty	A preliminary result is overgeneralised	What independent evidence supports it?
Peer review becomes proof	Publication is treated as permanent certification	Has the claim survived later scrutiny and related work?
Uncertainty becomes ignorance	Qualified knowledge is dismissed because it is not absolute	How much is known, how confidently, and within what range?
Expertise becomes immunity	Authority replaces evidence	What evidence and reasoning support the expert conclusion?
Incentive becomes evidence	Prestige, funding or popularity distort interpretation	Would the evidence still support the claim without the status signal?
Science becomes policy by itself	Empirical findings are treated as if they settle value choices	Which parts are evidence, and which parts are ethical or social judgement?

A Practical Test for Any Scientific Claim

When you encounter “Science says…” in a textbook, article, video, advertisement, policy debate or social-media post, do not begin by asking whether you agree. Begin by locating the claim inside the scientific chain.

Claim: What exactly is being asserted?
Question: What question was the research actually designed to answer?
Evidence: What observations or data bear on the claim?
Measurement: How were the important variables measured?
Method: What kind of study or investigation produced the evidence?
Inference: What reasoning connects the data to the conclusion?
Uncertainty: What variation, limitations or confidence ranges remain?
Alternatives: What competing explanations were considered?
Cross-check: Do independent methods, studies or data point the same way?
Boundary: To which people, places, scales or conditions does the conclusion apply?
Integrity: Are methods and reporting transparent enough to inspect?
World return: When the claim is used, do real outcomes behave as expected?

You do not need to be a specialist to ask these questions. You may still need expert help to evaluate the answers, but the structure protects you from confusing confidence with certainty, publication with proof, or data with explanation.

The Civilisation-Scale Science Loop

At civilisation scale, Science becomes a distributed memory-and-correction system. Observations are recorded. Instruments are standardised. Methods are taught. Results are archived. Scientific communities criticise claims. Engineering turns some knowledge into tools and infrastructure. Medicine, agriculture, computing, environmental monitoring and many other fields apply the resulting knowledge. The consequences of those applications then create new evidence.

That gives us a larger loop:

WORLD → EVIDENCE → MODEL → TEST → SHARED KNOWLEDGE → APPLICATION → CONSEQUENCE → NEW EVIDENCE → REVISION

Science therefore does more than accumulate answers. At its best, it improves a civilisation’s ability to notice when its understanding is wrong before the cost of being wrong becomes too large.

The Most Important Scientific Habit

The most important habit is not scepticism about everything. Endless doubt can become as unscientific as unquestioning belief.

The stronger habit is calibrated confidence: believe in proportion to the quality and quantity of evidence, use the best available explanation when action is needed, remain clear about uncertainty, and keep enough of the pathway visible that new evidence can change the conclusion.

Science works when confidence can rise with evidence, fall with contradiction, and remain correctable by the world.

Observable Mastery Test

Take any scientific claim—something simple such as “light is needed for photosynthesis” or something advanced from medicine, climate science, astronomy or materials science.

You understand how Science works if you can trace:

phenomenon → question → observation → measurement → model → prediction → method → data → analysis → uncertainty → alternative explanations → independent checking → conclusion → boundary → application → return evidence → revision

If one link is missing, you have found the next useful question.

Evidence Base and Further Reading

This article uses a mainstream baseline drawn from scientific-practice, measurement, research-integrity and Science-education sources. Useful starting points include:

Where to Go Next

If you want the world-facing Science library rather than this mechanism map, continue to Science World. If you are teaching a learner, enter through the Primary, Secondary or JC Science routes above. If you want the neighbouring mechanisms, continue through How Mathematics Works, How Evidence Works, How Reasoning Works and How Vocabulary Really Works.

Science does not work because humans never make mistakes. Science works when the system makes mistakes increasingly discoverable, explanations increasingly testable, and knowledge increasingly correctable by the world.

How Science Works | How Humans Build, Test and Correct Knowledge About the World