Explanation is the process of making something understandable by showing how, why or under what conditions it occurs. If you are asking “what is explanation?”, the useful answer is that an explanation connects a question to a structure that reduces confusion. It may identify causes, mechanisms, rules, relationships, purposes, definitions or evidence. Explanations appear in science, mathematics, history, medicine, engineering, education, law, computing and everyday conversation because people need more than facts: they need to see how the facts fit together.
A complete explanation is not the same as a description, prediction or argument. A description tells what something is like. A prediction says what is likely to happen. An argument gives reasons for accepting a claim. An explanation shows why the phenomenon makes sense. These activities can overlap, but they answer different questions. Saying that a metal rod expanded is a description; predicting that it will expand when heated is a prediction; explaining expansion connects temperature, atomic motion and material structure.
This guide explains explanation from first principles: explanandum and explanans, causes and mechanisms, laws and models, contrastive questions, levels of explanation, mathematical proof, historical explanation, functional explanation, scientific explanation, simplicity, depth, unification, counterfactual dependence, evidence, analogy, examples, diagrams, teaching, common explanatory failures and artificial intelligence. The goal is a clean operating model: identify what needs explaining, determine what kind of answer the question requires, build the causal or logical structure and test whether the explanation improves understanding without hiding important uncertainty.
What Is an Explanation? A Short Definition
An explanation is an account that makes a phenomenon, event, pattern or fact intelligible by connecting it to relevant causes, mechanisms, rules or relationships.
The thing to be explained is often called the explanandum. The information doing the explaining is the explanans. If the explanandum is “the glass cracked”, the explanans might involve rapid temperature change producing uneven thermal expansion and stress.
A strong explanation therefore has direction. It moves from a question or puzzling fact toward a structure that shows why the result occurred, had to occur or was likely under those conditions.
Explanation Versus Description
Description tells us what is present or what happened. Explanation tells us why or how.
“The plant leaves are yellow” is a description. “The leaves yellowed because chlorophyll production fell under nitrogen deficiency” is an explanation if the mechanism and evidence are correct.
Descriptions are still essential because explanations need accurate facts to explain. But adding detail to a description does not automatically transform it into an explanation.
Explanation Versus Prediction
Prediction moves from a model or evidence to an expected outcome. Explanation moves from a phenomenon to a structure that makes it understandable.
A model can predict well without explaining deeply. Machine-learning systems can forecast outcomes using correlations whose mechanisms remain unclear.
Conversely, an explanation can identify a mechanism without producing highly precise predictions if the system contains many interacting variables.
Explanation Versus Argument
An argument aims to justify a conclusion. An explanation normally begins with a fact or phenomenon that is already accepted and asks why it is so.
“The bridge failed because fatigue cracks grew under repeated loading” explains a failure. “The bridge probably failed from fatigue because the fracture surface shows beach marks and the loading history was cyclic” is an argument supporting that explanation.
In real reasoning, explanation and argument often work together: evidence justifies the explanatory model, and the model makes the evidence coherent.
Explanation Versus Definition
A definition tells us what a term means. An explanation tells us why or how the thing behaves as it does.
Defining evaporation as the escape of molecules from a liquid surface does not fully explain why evaporation speeds up at higher temperature. The explanation requires molecular energy distributions and interactions.
Definitions are foundations for explanation because unclear terms make causal or logical relationships difficult to state precisely.
Why Questions
Many explanations answer “why?” questions. Why did the reaction speed up? Why did the student make this mistake? Why did the population decline?
But “why?” is ambiguous. It can ask for cause, purpose, justification, origin or mechanism.
A good explainer first identifies which kind of why-question is being asked. Otherwise the response can be technically true yet unsatisfying.
How Questions
How-questions often ask for mechanism or process. “How does a vaccine generate immunity?” requires a sequence of biological interactions.
Sometimes “how” asks for procedure: “How do I solve this equation?” The appropriate explanation then describes the operations and why they preserve equivalence.
Mechanistic how-explanations are especially useful because they show intermediate steps rather than jumping from input to output.
Contrastive Explanation
Explanations are often contrastive. The real question is not simply “Why did X happen?” but “Why did X happen rather than Y?”
Why did this student improve while another student with the same textbook did not? Why did one component fail while its neighbours survived? The contrast determines which differences are relevant.
A good explanation therefore identifies the comparison class. Without the contrast, the response may include causes that are common to both cases and therefore do not explain the difference.
Causal Explanation
A causal explanation shows how causes produced an effect. It may identify triggers, background conditions, mediators and feedback loops.
Causal explanations are especially valuable for intervention because they suggest what might change the outcome. If poor ventilation caused overheating, improving ventilation is a candidate repair.
But causal explanations require care: correlation, sequence and storytelling are not enough. The proposed cause should survive comparison with alternative causal structures.
Mechanistic Explanation
A mechanistic explanation describes entities, activities and organisation that produce a phenomenon.
Explaining muscle contraction might involve nerve impulses, calcium release, actin, myosin and ATP-driven interactions. The explanation shows how lower-level parts generate higher-level behaviour.
Mechanisms provide depth because they open the black box between cause and effect.
Laws and General Principles
Some explanations show that a phenomenon follows from a general law or principle plus initial conditions.
A projectile’s trajectory can be explained through equations of motion under gravity, while a gas relationship may be explained with thermodynamic principles under stated assumptions.
The power of a law-based explanation comes from connecting one case to a wider regularity.
Models as Explanations
Models simplify reality to expose the relationships relevant to a question. A supply-and-demand graph, circuit diagram or planetary model can explain patterns without reproducing every physical detail.
A model explains when its simplifications preserve the structure that generates the phenomenon.
The best model is not always the most detailed. Extra detail can hide the relationship the explanation is meant to reveal.
Levels of Explanation
The same phenomenon can be explained at several levels. A student’s poor test performance can be explained through missing knowledge, weak retrieval, sleep deprivation, task design or broader institutional conditions.
These explanations are not automatically competitors because they may describe different levels of the same system.
The useful level depends on the question and intervention. A neural explanation may be scientifically interesting but less useful than a curriculum explanation for fixing a classroom problem.
Reduction and Emergence
Reduction explains a higher-level phenomenon through lower-level components. Chemistry can often be connected to atomic and quantum processes.
Emergence emphasises patterns that arise from interactions among many components and may be difficult to understand by studying one component alone.
Good explanations can move between levels: identify the parts, show their interactions and explain the higher-level pattern they collectively produce.
Functional Explanation
Functional explanations describe what a feature or process does within a system. The heart pumps blood; roots anchor plants and absorb water; a checksum helps detect data corruption.
Function should not be confused with conscious purpose. Biological features can have functions shaped by evolution without being designed by an intentional agent.
In engineering, function is often explicitly tied to design goals, so purpose-based explanation is appropriate.
Teleological Explanation
Teleological explanations explain something by its purpose or goal. “The valve opens to release pressure” is natural in designed systems.
In biology, careless purpose language can suggest intentional design where evolutionary function is the better concept.
Teleological language is most useful when goals genuinely organise behaviour, as in human planning, organisations and engineered systems.
Historical Explanation
Historical explanation connects events through chronology, institutions, decisions, constraints and long-term conditions.
A war, economic crisis or political transition rarely has one cause. Historians distinguish structural conditions, enabling factors, triggers and contingent choices.
A strong historical explanation uses primary evidence and comparison rather than merely telling a plausible story after the outcome is known.
Narrative Explanation
Narrative explanations organise events into a sequence that shows how earlier states led to later ones.
Narrative is powerful because human memory and understanding track agents, goals and changes over time.
But a smooth story can create an illusion of causality. Good narrative explanation distinguishes documented links from interpretation and uncertainty.
Mathematical Explanation
Mathematics can explain by showing that a result follows from structure, symmetry, invariance or a theorem.
A proof may do more than establish truth. An explanatory proof reveals why the result holds by highlighting the key relationship rather than relying on opaque calculation.
Two proofs of the same theorem can therefore differ in explanatory value even when both are valid.
Scientific Explanation
Scientific explanations connect phenomena to evidence-supported models, causes, mechanisms or laws.
They are constrained by measurement and testability. An explanation should make contact with observations that could support, refine or challenge it.
Scientific explanation is provisional because models can improve. Provisional does not mean arbitrary; it means conclusions remain responsive to evidence.
Unification
An explanation has unifying power when one principle explains many apparently separate phenomena.
Evolution by natural selection connects adaptation across species. Plate tectonics links earthquakes, volcanoes, mountain formation and continental movement.
Unification is valuable because it reduces the number of independent assumptions needed to understand many observations.
Explanatory Depth
Depth refers to how many important layers of mechanism or dependency an explanation reveals.
“The computer crashed because of a bug” is shallow. Identifying a memory leak caused by unreleased objects and showing how resource exhaustion triggered failure is deeper.
Depth should match the audience and purpose. A deeper explanation is not always a better first explanation if it overwhelms the learner.
Explanatory Power
Explanatory power is the ability of an account to make a wide range of relevant observations intelligible with a coherent structure.
An explanation gains power when it accounts not only for the main result but also for boundary cases, timing, side effects and patterns that rival accounts leave unexplained.
Power should not be confused with vagueness. A theory that can be stretched after the fact to fit every possible outcome explains very little because no observation could genuinely challenge it.
Robustness
A robust explanation survives reasonable changes in assumptions, measurements and modelling choices.
If the explanation disappears when one arbitrary parameter is changed slightly, confidence should be lower. If several independent methods recover the same relationship, the explanation becomes stronger.
Robustness is especially important in complex systems where no single measurement or model captures everything perfectly.
Invariance
Some deep explanations reveal relationships that remain stable across transformations or contexts. Symmetries in physics and conservation laws are powerful partly because they identify what does not change when other features do.
Invariance can also matter outside physics. A reliable algorithm may preserve correctness across different input sizes; a causal mechanism may remain stable across several environments even when surface correlations shift.
Finding what remains invariant helps separate essential structure from accidental detail.
Simplicity
Simple explanations are often preferred when they explain the same evidence with fewer assumptions.
But simplicity is not the same as shortness. A short claim can be simplistic if it ignores important mechanisms or exceptions.
The right goal is economical sufficiency: include enough structure to explain the phenomenon without adding detail that does no explanatory work.
Occam’s Razor
Occam’s razor is the principle that unnecessary assumptions should not be multiplied.
It does not mean “the simplest idea is always true”. A more complex model can be better if the evidence requires the complexity.
The principle helps compare models that explain the same observations equally well: prefer the one that achieves the explanation with fewer unsupported commitments.
Counterfactual Explanation
Counterfactual explanations answer questions such as “What would have needed to be different for the outcome to change?”
If a loan application was rejected because income fell below a threshold, a counterfactual explanation might identify the smallest relevant change that would have produced approval.
Counterfactual explanation is useful in causal reasoning, debugging and interpretable AI because it connects explanation to possible intervention.
Explanatory Relevance
Not every true fact belongs in an explanation. The relevant information is what makes a difference to the explanandum or reduces the specific uncertainty in the question.
A student’s birthday may be true but irrelevant to why an algebra step is invalid.
Good explanation therefore selects. Relevance is one reason concise structure can outperform exhaustive detail.
Background Conditions
Causes operate against background conditions. A spark causes ignition only when fuel and oxygen are present.
Explanations often omit stable background conditions because they are taken for granted. This is efficient until the background changes.
When an explanation fails to transfer to a new setting, hidden background assumptions are often the reason.
Explanation and Evidence
An explanation can be coherent yet false. Evidence determines whether the explanatory model deserves belief.
The strongest evidence discriminates among competing explanations. An observation predicted equally by every hypothesis adds little explanatory support.
Good reasoning therefore asks not only “Does this fit?” but “Would I have expected this evidence if an alternative explanation were true?”
Mechanistic Evidence
Mechanistic evidence shows that the proposed parts and processes actually exist and interact in the way the explanation requires.
In medicine, a statistical treatment effect may be strengthened by evidence that the drug reaches the target, changes the relevant pathway and produces downstream effects predicted by the mechanism.
Mechanistic evidence alone is not always enough to establish the magnitude of a real-world effect, but it can help distinguish genuine causation from accidental association.
Statistical Explanation
Statistical explanations show how patterns arise from distributions, sampling, randomness or aggregation. A regression toward the mean, for example, can explain why unusually extreme measurements are often followed by less extreme ones.
Statistical explanation is not inferior to mechanism. In some phenomena, the key explanatory structure is probabilistic.
The important point is to avoid turning statistical association into causal mechanism without additional justification.
Explanation and Uncertainty
Explanations should represent uncertainty when the evidence does not uniquely identify one account.
A scientist may say that several mechanisms are plausible, with one better supported than the others. A clinician may explain that symptoms fit a diagnosis but a confirmatory test is still required.
Stating uncertainty does not weaken an explanation. It makes the boundary between evidence and speculation visible.
Inference to the Best Explanation
Abductive reasoning compares candidate explanations and selects the one that best accounts for the evidence.
Criteria can include explanatory scope, consistency with established knowledge, simplicity, causal plausibility and successful new predictions.
The result remains defeasible: a better explanation may appear when new evidence changes the comparison.
Explanation and Prediction Together
A strong explanatory model often generates predictions. If the proposed mechanism is correct, changing a component should produce specified consequences.
Novel successful predictions strengthen confidence because the model succeeded beyond the observations it was designed to fit.
Prediction therefore acts as a stress test for explanation, though some explanatory fields deal with unique historical events where prediction has a different role.
Explanation in Engineering
Engineering explanations often focus on mechanism, failure and design constraints.
Why did the bearing fail? Was lubrication inadequate, alignment poor, material defective or loading excessive? The explanation guides the repair.
Engineering explanation is intervention-oriented: the account is valuable when it identifies changes that prevent recurrence.
Explanation in Medicine
Medical explanation can operate at several levels: symptoms, diagnosis, physiology, cellular mechanism and risk factors.
A diagnosis names a pattern but may not fully explain the mechanism. “Migraine” can classify the condition while leaving detailed biological pathways partly uncertain.
Clinicians also need patient-level explanation: what is happening, what is known, what remains uncertain and why a treatment is recommended.
Explanation in Education
Teaching depends on explanation, but telling is not the same as causing understanding.
A useful educational explanation connects new ideas to prior knowledge, uses examples and contrasts, checks misconceptions and invites the learner to reconstruct the logic.
If students can repeat the explanation but cannot solve a new problem, the explanatory model may not yet be integrated into usable knowledge.
Worked Examples
Worked examples explain procedures by showing each step and the reason for it.
They are especially helpful for novices because they reduce unnecessary search and make expert structure visible.
Examples become more educational when learners compare correct and incorrect solutions, explain transitions and later complete partially worked problems independently.
Analogies
Analogies explain an unfamiliar system by mapping it onto a familiar one.
Electric current may be compared with water flow, and memory may be compared with a network. These analogies highlight useful relationships.
Every analogy breaks somewhere. Strong explainers state the limit so learners do not import features that do not correspond.
Diagrams
Diagrams can reveal spatial, causal or hierarchical relationships more efficiently than prose.
A causal graph shows arrows among variables; a circuit diagram shows components and connections; a flowchart shows sequence and branching.
A diagram explains only if its symbols and relationships are understood. Decorative complexity can make visual explanation worse.
Examples and Non-Examples
Examples show where a concept applies; non-examples show its boundary.
To explain a triangle, showing several triangles is useful, but showing shapes that are almost triangles helps learners identify which properties are essential.
Contrast is explanatory because it prevents the learner from attaching the concept to irrelevant surface features.
Explanation and Misconceptions
Learners often bring existing explanations that conflict with the target model. Simply stating the correct explanation may leave the old model intact.
Effective correction makes the misconception visible, shows where it fails and provides an alternative that explains the evidence better.
Conceptual change is therefore comparative: the learner must see why the new model has greater explanatory power.
Explanation and Cognitive Load
An explanation can fail because it demands too much working memory. Too many new terms, branches and exceptions can overload the learner before the core relationship is understood.
Layered explanation solves this by beginning with the central structure, then adding detail progressively.
This is not dumbing down. It is sequencing complexity so each layer has something stable to attach to.
The Audience Determines the First Layer
The same phenomenon may need different explanations for a child, undergraduate, engineer or specialist researcher.
Changing the level does not require changing the truth. It requires selecting the concepts and vocabulary the audience can use while preserving a route toward deeper detail.
A good explanation can therefore be layered: a concise core answer, a mechanism, examples, technical qualifications and source evidence. Different readers can stop at the depth appropriate to their purpose.
The Feynman Technique and Self-Explanation
Self-explanation asks learners to explain an idea in their own words, identify steps and connect them to prior knowledge.
The popular Feynman-style approach—explain simply, find gaps, return to sources and refine—works because generation reveals where knowledge is fragile.
Simplicity should not become inaccuracy. The goal is clear reconstruction, not replacing technical concepts with vague metaphors.
Explanation in Law
Legal explanation connects facts, rules, precedents and reasoning to a conclusion.
A judgment should make the path visible: which facts were accepted, which rule applied and why competing interpretations were rejected.
Transparency matters because legal decisions must often be reviewed by parties who disagree with the outcome.
Explanation in Computing
Computing explanations operate at several levels: source code, algorithms, data structures, operating systems, hardware and network behaviour.
A user may need “the file failed because the format is unsupported”, while an engineer needs the parser state and invalid byte sequence.
Good technical explanation matches level to audience while preserving a route to deeper detail.
Debugging as Explanation
Debugging is explanation under pressure. A system behaves unexpectedly, and the developer must identify the cause.
Logs, tests, reproduction steps and controlled changes discriminate among hypotheses.
The repair is stronger when the explanation predicts both the failure and why the fix should remove it.
Explanation in Science Communication
Science communication must preserve both mechanism and uncertainty while reducing technical barriers.
The temptation is to remove caveats until a result sounds clean. But a simplified explanation becomes misleading if it hides the population, conditions or uncertainty that determine whether the result applies.
Strong public explanation keeps the core model simple while making boundaries visible: what was studied, what the evidence shows, what remains uncertain and what should not be concluded.
Explanation in Artificial Intelligence
AI systems can generate explanations in natural language, but fluent explanation is not automatically faithful to the system’s actual internal process.
Explainable AI includes methods that identify influential features, counterfactual changes, examples or simplified local models.
Different explanation methods serve different audiences: developers may need debugging information, regulators may need accountability and users may need understandable reasons for a decision.
Faithfulness and Plausibility
An explanation can sound plausible without accurately describing the process that produced the result.
Faithfulness asks whether the explanation corresponds to the actual causal or computational mechanism. Plausibility asks whether it sounds reasonable to a human.
For consequential systems, plausibility alone is insufficient. Explanations should be validated against behaviour, interventions or internal evidence where possible.
Local and Global Explanations
A local explanation accounts for one particular result. A global explanation describes how the system behaves across many cases.
Why was this application rejected? is local. What factors generally drive the model’s decisions? is global.
A system can be understandable locally while remaining complex globally, so both levels may be needed.
Explanatory Compression
Good explanations compress many facts into a smaller structure. Instead of memorising hundreds of separate observations, a learner uses one principle to organise them.
Compression is not deletion of truth; it is representation through a model that preserves the relationships needed for understanding.
This is why unifying theories and schemas feel powerful: they reduce cognitive burden while increasing the number of cases that can be explained.
When an Explanation Is Too Shallow
A shallow explanation names a category without revealing mechanism. “The computer is slow because of performance issues” merely restates the problem.
A deeper explanation identifies memory pressure, disk saturation, network latency or another specific bottleneck and shows how it produces the observed symptoms.
The test is whether the explanation helps predict, diagnose or intervene better than the original description.
When an Explanation Is Too Deep
An explanation can also overshoot. A parent asking why a child struggles with fractions may not need a neural-level account of working memory.
Explanatory depth should match the decision. Too much detail can obscure the actionable level.
A strong communicator can move between layers: short answer first, mechanism next, technical depth on demand.
Explanatory Gaps
An explanatory gap occurs when a proposed account leaves an important transition unexplained.
“The policy changed behaviour because incentives changed” leaves a gap if it never explains how the incentive affected choices or which population responded.
Finding gaps is easier when the explanation is written as a chain of explicit steps rather than a smooth paragraph.
Common Explanatory Failures
- Restatement: repeating the phenomenon using different words without adding structure.
- Label substitution: naming a category as though the label itself explains the mechanism.
- Single-cause oversimplification: ignoring interacting causes in complex systems.
- Correlation-as-cause: treating association as causal explanation.
- Storytelling after the fact: producing a coherent narrative that was not tested against alternatives.
- Excessive jargon: replacing understanding with technical vocabulary.
- Analogy overreach: extending a comparison beyond the relationships it genuinely shares.
- False certainty: presenting one explanation as settled when evidence leaves important alternatives open.
A Practical Explanation Checklist
First define the explanandum: exactly what needs to be explained? Then identify the contrast: why this rather than what alternative?
Choose the explanatory type—causal, mechanistic, mathematical, functional, historical or another appropriate form. State assumptions and relevant background conditions.
Finally, test whether the explanation distinguishes alternatives, fits the evidence, supports useful predictions or interventions and is pitched at the right level for the audience.
How to Explain Clearly
Begin with the core relationship in one or two sentences. Define unfamiliar terms. Show the mechanism or logic in ordered steps.
Use one strong example, then a contrasting example. Add a diagram when relationships are spatial or sequential.
End by checking transfer: can the reader use the explanation to understand a new case? If not, the explanation may have produced recognition without genuine understanding.
How to Learn Through Explanation
After studying a concept, close the source and explain it from memory. Identify the phenomenon, mechanism, evidence and boundary conditions.
Notice where you use vague phrases such as “somehow”, “basically” or “it just works”. Those phrases often mark missing structure.
Return to the source, repair the gap and explain again. Repeated self-explanation turns passive familiarity into organised knowledge.
Frequently Asked Questions About Explanation
What is the difference between an explanation and a description?
A description states what something is like or what occurred. An explanation shows why or how it occurred.
What makes an explanation scientific?
Scientific explanations connect phenomena to evidence-supported models, mechanisms, laws or causal relationships and remain open to testing and revision.
Does a good explanation need a cause?
Not always. Mathematical, functional and structural explanations can be explanatory without having the same causal form as physical mechanisms.
Can two explanations both be correct?
Yes. They may operate at different levels or explain different contrasts. A biological and social explanation can both contribute to understanding one outcome.
What is a mechanism?
A mechanism is an organised set of entities and activities that produces a phenomenon or connects a cause to an effect.
Why are analogies useful?
They map familiar relationships onto unfamiliar ones, reducing the amount of new structure a learner must build at once.
Why can an explanation sound good but be wrong?
Coherence and fluency can make a story persuasive even when evidence does not support its causal or logical structure.
How do I know whether I understand an explanation?
Try to reconstruct it without notes, identify assumptions, predict what would change under a different condition and apply the model to a new case.
Authoritative Starting Points
For deeper study, use philosophy-of-science texts and the Stanford Encyclopedia of Philosophy entry on scientific explanation. University resources in causal inference, modelling, history and mathematics show how explanatory standards differ by domain while retaining common demands for relevance, structure and evidence.
What Is Explanation? The Complete Idea
Explanation is the structured reduction of puzzlement. It connects a phenomenon to causes, mechanisms, laws, models, purposes or logical relationships that make the result intelligible.
The strongest understanding keeps several distinctions clear: description is not explanation; prediction is not explanation; plausibility is not evidence; a label is not a mechanism; one explanation need not exclude every other level; and more detail does not automatically produce more understanding.
Once those distinctions are stable, explanation becomes a practical tool for thinking and teaching. Define what needs explaining, identify the contrast, choose the right level, expose the mechanism, connect it to evidence and test whether the account supports new understanding. A good explanation does more than answer a question. It changes what the learner can see next.
