A hypothesis is a proposed explanation or relationship that can be examined against evidence.
“Study time improves scores”, “this component causes the failure”, “the decline is driven by reduced demand”, “the signal is produced by sensor drift” and “two variables are unrelated” are all hypotheses. They differ in type, scope, mechanism, testability, evidence burden and status.
Quick answer: how should hypotheses be categorised?
- Type: causal, associative, mechanistic, descriptive, null, alternative?
- Scope: case-specific, population-level, local, general?
- Direction: directional or non-directional?
- Mechanism: does it propose how the effect occurs?
- Testability: what observations could support or challenge it?
- Evidence: what is already known?
- Alternatives: what competing hypotheses exist?
- Confidence: speculative, plausible, supported, strongly supported?
- Status: proposed, testing, supported, rejected, revised, unresolved?
- Provenance: who proposed it and from what observations or model?
This article extends How to Categorise Claims by focusing on claims deliberately structured for investigation.
1. Separate hypothesis from assumption
An assumption is accepted so reasoning can proceed. A hypothesis is proposed so evidence can change our view of it.
2. Separate hypothesis from fact
A hypothesis remains provisional even when evidence is strong.
3. Separate hypothesis from theory
A theory is a broader explanatory framework; individual hypotheses can be derived from it and tested.
4. Separate hypothesis from prediction
A hypothesis proposes a relationship or mechanism. A prediction states what should be observed if the hypothesis is correct under specified conditions.
5. Causal hypotheses propose production
They claim that changing one factor changes another outcome.
6. Associative hypotheses propose relationship
They state that variables move together without necessarily asserting causation.
7. Mechanistic hypotheses propose how
They specify an intermediate process linking cause and outcome.
8. Descriptive hypotheses propose pattern
They may predict distribution, frequency or structure without claiming mechanism.
9. Null hypotheses formalise no effect or difference
They provide a reference proposition against which evidence can be evaluated.
10. Alternative hypotheses propose departure from the null
There may be several competing alternatives rather than one.
11. Directional hypotheses predict direction
They state that one condition increases, decreases, improves or worsens another.
12. Non-directional hypotheses predict difference without direction
They are useful when the existence of an effect is proposed but its sign is uncertain.
13. Simple hypotheses involve few variables
They are easier to isolate and test.
14. Compound hypotheses bundle several propositions
When possible, split them so evidence can support one part without implying support for all parts.
15. Scope controls generalisation
A hypothesis about one classroom does not automatically generalise to every student or school.
16. Population should be explicit
Age, geography, system type, operating mode and sampling frame can change whether a hypothesis holds.
17. Time scope matters
A relationship can be stable in one period and disappear after technology, policy or behaviour changes.
18. Boundary conditions strengthen hypotheses
Stating where a relationship should and should not hold makes testing more informative.
19. Testability requires possible disconfirmation
A hypothesis that can accommodate every possible observation is difficult to learn from.
20. Falsifiability is one form of testability
Empirical hypotheses are stronger when we can identify observations that would count against them.
21. Evidence quality and evidence quantity differ
Many weak observations do not necessarily outweigh one well-controlled decisive observation.
22. Independent replication changes confidence
Evidence from independent sources reduces the risk that one method or dataset created the pattern.
23. Negative evidence can challenge hypotheses
Expected signals that consistently fail to appear can reduce support.
24. Absence of evidence is not always evidence of absence
The test may simply have lacked sensitivity, sample size or correct timing.
25. Competing hypotheses improve diagnosis
Keeping several plausible explanations visible reduces premature closure.
26. Discriminating tests are especially valuable
The strongest next observation is often the one that produces different predictions under rival hypotheses.
27. Hypothesis confidence should change with evidence
Proposed, plausible, supported and strongly supported are better treated as changing states than fixed types.
28. Rejected hypotheses remain historically useful
They show what was considered, why it failed and what evidence changed the conclusion.
29. Revised hypotheses need lineage
A narrower or modified hypothesis should link back to the earlier version it replaced.
30. Exploratory hypotheses differ from confirmatory hypotheses
Hypotheses generated after examining the data should be distinguished from those specified before testing.
31. Data-derived hypotheses need independent testing
Patterns discovered in one dataset can be overfit to that dataset.
32. AI can generate hypothesis candidates
Models can broaden the search space, but generated hypotheses should remain distinct from validated evidence.
33. AI can rank hypotheses imperfectly
Popularity or textual plausibility should not replace domain evidence and discriminating tests.
34. A practical hypothesis record
- hypothesis ID;
- statement;
- type;
- scope;
- population or system;
- mechanism;
- predictions;
- supporting evidence;
- challenging evidence;
- competing hypotheses;
- confidence;
- status;
- proposer;
- version.
35. Hypothesis categories should improve testing
A useful scheme changes which evidence is gathered, which alternatives are compared and what would count as disconfirmation.
36. The deeper idea
A hypothesis is not an answer. It is a structured invitation for reality to disagree.
To categorise a hypothesis well is to preserve what it proposes, where it should hold, what it predicts, how it can fail and how evidence has changed its status.
Final answer
Categorise hypotheses by type, scope, direction, mechanism, testability, evidence, alternatives, confidence, status and provenance. Keep hypotheses separate from assumptions, facts, theories and predictions, and preserve competing explanations until discriminating evidence justifies stronger closure.