An explanation is an account of why, how or in what sense something is the case.
“The bridge failed because fatigue accumulated”, “the price rose because supply tightened”, “the ritual matters because it preserves identity”, “the score differs because the groups had different prior preparation” and “the machine stopped because a sensor triggered a safety interlock” are all explanations. They differ in causal depth, mechanism, evidence, scope and purpose.
Quick answer: how should explanations be categorised?
- Type: causal, mechanistic, functional, statistical, historical, interpretive?
- Scope: one case, class of cases, system-wide?
- Depth: surface, intermediate, root-level?
- Mechanism: what sequence or structure produces the result?
- Evidence: what observations and sources support it?
- Alternatives: what competing explanations exist?
- Audience: learner, practitioner, expert, public?
- Uncertainty: what remains unresolved?
- Coherence: does the account fit established knowledge?
- Reach: does it explain related observations or generate predictions?
This article complements How to Categorise Models, How to Categorise Claims and How to Categorise Hypotheses by focusing on explanatory accounts rather than representations or propositions alone.
1. Separate explanation from description
Description states what happened or what something is like. Explanation adds why, how or under what structure it occurred.
2. Separate explanation from prediction
A model can predict accurately without revealing why, while a strong explanation may not forecast every individual case precisely.
3. Separate explanation from justification
Explanation accounts for a state or event. Justification argues that a belief, claim or action is warranted.
4. Causal explanations identify producing factors
They state that one or more factors contributed to bringing about an outcome.
5. Mechanistic explanations identify how causation unfolds
They specify components, interactions and intermediate steps between cause and effect.
6. Functional explanations identify role
They explain what a component or practice does within a larger system.
7. Statistical explanations identify patterned variation
They account for differences through distributions, associations, probabilities or modelled factors.
8. Historical explanations reconstruct pathways
They explain a present state through sequences of earlier events, decisions, institutions and contingencies.
9. Interpretive explanations account for meaning
They explain symbols, texts, actions or practices through context, intention and cultural significance.
10. Structural explanations emphasise arrangement
Network position, institutional design or system architecture can explain outcomes without reducing everything to one local cause.
11. Intentional explanations use goals and beliefs
They explain actions through what an agent intended, believed or valued.
12. Normative explanations need caution
Saying what ought to happen is not the same as explaining why something did happen.
13. Proximate explanations focus on immediate causes
They identify the nearest trigger or mechanism.
14. Distal explanations focus on deeper conditions
Longer-term structure, history or environmental conditions may explain why the proximate trigger had power.
15. Root-cause language should be used carefully
Complex systems often have several necessary or contributing causes rather than one final root.
16. Multi-level explanations can coexist
A behaviour can be explained biologically, psychologically, socially and institutionally at different resolutions.
17. Different levels need not compete
Two explanations can be compatible if they answer different parts of the why-question.
18. Scope should be explicit
An explanation for one failure may not explain every failure of the same type.
19. Case explanations differ from general explanations
One reconstructs why this instance occurred; the other explains a recurring class of phenomena.
20. Evidence should match the explanation type
Causal claims may need experiments or quasi-experiments; historical explanations need source chains; interpretive explanations need contextual textual or cultural evidence.
21. Correlation alone is not causal explanation
Association can motivate hypotheses but does not automatically identify mechanism or direction.
22. Mechanistic plausibility is not proof
A coherent mechanism still needs evidence that the proposed pathway actually operated in the case.
23. Counterfactual reasoning strengthens causal explanation
Ask what would likely have happened if the proposed cause had been absent or different.
24. Alternative explanations should remain visible
Competing accounts reduce the risk that one plausible story becomes accepted too early.
25. Explanatory competition should use discriminating evidence
The strongest evidence is often what one explanation predicts and another does not.
26. Simpler explanations can be preferable
When two explanations fit equally well, the one with fewer unsupported assumptions may be easier to trust and test.
27. Simplicity is not always truth
Real systems can be genuinely complex, and oversimplification can erase necessary structure.
28. Explanatory depth should match audience and task
A Primary-level explanation, professional briefing and research account can differ in depth while preserving the same core mechanism.
29. Educational explanations should preserve direction
Simplification should not teach a false mechanism that later must be discarded entirely.
30. Explanatory reach matters
An explanation gains value when it accounts for several related observations without ad hoc patches.
31. Predictive success can support explanations
If an explanation generates novel predictions that later succeed, confidence can increase.
32. Prediction without mechanism has limited explanatory depth
Forecasting patterns can be useful even when the underlying reason remains unknown.
33. Explanations can change with new evidence
Versioning should preserve how the account evolved and which observations forced revision.
34. AI can generate plausible explanations
Fluency and coherence do not establish truth. Generated explanations should remain linked to evidence and competing alternatives.
35. AI explanations can rationalise after the fact
A model may produce a convincing narrative that was not the actual mechanism behind its output or the real-world event.
36. A practical explanation record
- explanation ID;
- phenomenon explained;
- explanation type;
- scope;
- mechanism or structure;
- causes or conditions;
- supporting evidence;
- challenging evidence;
- alternatives;
- assumptions;
- uncertainty;
- audience level;
- explanatory reach;
- status;
- version.
37. Explanation categories should improve evaluation
A useful scheme tells us what kind of evidence, mechanism and alternative comparison are needed.
38. Explanation is a relationship between question and account
The same facts can support different explanations because different why-questions ask for different levels or kinds of answer.
39. Good explanations make hidden structure visible
They connect observations into a coherent account without pretending uncertainty has disappeared.
40. The deeper idea
An explanation is not merely a story after the fact. It is a structured claim about why the world took one path rather than another.
To categorise an explanation well is to preserve what kind of why-answer it gives, what evidence supports it, what alternatives remain, and how far the explanation can travel before it stops being reliable.
Final answer
Categorise explanations by type, scope, depth, mechanism, evidence, alternatives, audience, uncertainty, coherence and explanatory reach. Keep explanation separate from description, prediction and justification, and preserve competing accounts until evidence discriminates among them.