skills/scientific-thinking/hypothesis-generation
Hypothesis Generation
Hypothesis generation is the bridge between observation and experimentation. A good hypothesis transforms vague curiosity into a testable prediction, making explicit what you expect to happen, why, and under what conditions. This skill provides frameworks for converting research gaps, contradictory findings, and cross-domain insights into falsifiable statements that drive experimental design and scientific progress.
When to use
- Identifying testable predictions from literature gaps or contradictions
- Converting brainstorming ideas into experimentally tractable questions
- Formulating hypotheses from clinical or observational patterns
- Developing grant-specific aims with clear predictions
- Designing experiments to distinguish between competing theories
- Applying cross-domain analogies to generate novel predictions
- Moving from exploratory analysis to confirmatory testing
When NOT to use
- Early-stage exploration without clear direction (use brainstorming)
- Evaluating evidence quality of existing hypotheses (use critical thinking)
- Confirming established knowledge (that's verification, not hypothesis generation)
- When a phenomenon is already well-explained and no gap exists
- For purely descriptive research without predictive component
Prerequisites
- Sufficient domain knowledge to understand the current state of research
- Familiarity with relevant literature and open questions
- Basic understanding of experimental design principles
- Access to data or observations to build upon
- Awareness of competing theories or explanations in the field
Core workflow
1. Identify the hypothesis source
Determine where the hypothesis idea originates:
Knowledge gaps in literature
- What questions remain unanswered?
- What has been overlooked or insufficiently studied?
- What contradictions exist between studies?
Contradictory findings
- Studies that report opposing results
- Inconsistent effect sizes across populations
- Conflicting mechanistic explanations
Cross-domain analogies
- Insights from other fields that might apply
- Biological parallels to engineering solutions
- Methodological advances from adjacent areas
Methodological advances
- New techniques that enable previously impossible tests
- More precise measurements or manipulations
- Scale or resolution improvements
Clinical or observational patterns
- Unexpected observations in practice
- Patient/subject patterns noticed but unexplained
- Natural experiments or quasi-experiments
Theoretical predictions
- Predictions from models that need testing
- Implications of established theories
- Boundary conditions not yet explored
2. Specify the hypothesis anatomy
A well-formed hypothesis has five components:
1. Variables
- Independent variable (what you manipulate or vary)
- Dependent variable (what you measure)
- Control variables (what you hold constant)
2. Directional prediction
- What you expect to happen
- Specific direction (increase/decrease, positive/negative)
- Magnitude or effect size expectations
3. Mechanism (if causal)
- Why you expect this to happen
- Underlying causal chain
- Theoretical basis for the prediction
4. Scope/conditions
- When the hypothesis applies
- Boundary conditions
- Populations, settings, or contexts
5. Operationalization
- How each variable will be measured
- Cutoffs or thresholds for categorization
- Specific experimental protocols
Template:"
"We hypothesize that [IV] will [directionally affect] [DV] in [population/samples], because [mechanism], and this will be measured by [operationalization]."
3. Apply the falsifiability test
Following Popper's criterion, a hypothesis must be falsifiable:
Questions to ask:
- Could the hypothesis be proven wrong by observation?
- Is there a conceivable result that would contradict it?
- Are the variables operacionais in a way that allows rejection?
Types of unfalsifiable hypotheses to avoid:
- No clear prediction (too vague)
- Invulnerable to disconfirmation (always interprets结果是"支持")
- Tautological (true by definition)
- Untestable with available methods
Making hypotheses falsifiable:
- Specify exact predictions, not just direction
- State magnitude expectations
- Define rejection criteria in advance
- Identify alternative explanations
4. Design a test
How would you test this hypothesis?
Requirements:
- An experiment or observation that could yield conflicting results
- A way to measure the dependent variable
- Appropriate controls
- Sufficient sample/power
Design considerations:
- What design can differentiate this from alternatives?
- What would confirm? What would falsify?
- What are plausible confounds?
- How will you handle ambiguity in results?
5. Specify alternatives and boundary conditions
A strong hypothesis acknowledges:
Competing hypotheses
- What else could explain the results?
- What are the leading alternatives?
- How does this hypothesis differ?
Boundary conditions
- When would you expect this not to hold?
- What are the limits of the prediction?
- What contextual factors matter?
Effect size expectations
- What magnitude of effect makes the hypothesis "true"?
- What magnitude suggests falsification?
- Is the effect clinically/practically significant?
Code patterns
Hypothesis specification template
HYPOTHESIS: [Number]
Variables:
- Independent variable: [precise definition]
- Dependent variable: [precise definition]
- Control variables: [list]
Prediction:
- We expect [IV] to [increase/decrease] [DV] by approximately [magnitude/percentage].
Mechanism:
- [IV] affects [DV] through [mechanism], based on [theory/prior work].
Scope:
- This prediction applies to [population/context].
- We expect this in [conditions], but not when [conditions].
Operationalization:
- [IV] will be operationalized as: [specific measurement]
- [DV] will be operationalized as: [specific measurement]
Rejection criteria:
- If [observed result], we reject the hypothesis.
Alternative explanations:
1. [Competing hypothesis 1]
2. [Competing hypothesis 2]
Example: From gap to hypothesis
Source: Two studies on mitochondrial function in aging report conflicting results - one shows decline, one shows no change.
Gap: Why the inconsistency? One uses tissue homogenates, one uses single cells.
Hypothesis Generation:
HYPOTHESIS 1: Tissue-level measurements obscure cell-type heterogeneity
in mitochondrial function decline during aging.
Variables:
- Independent variable: Measurement scale (tissue homogenate vs. single cell)
- Dependent variable: Mitochondrial function (ATP production rate)
- Control variables: Age, species, tissue type
Prediction: Single-cell measurements will reveal significant
age-related decline that tissue homogenates obscure by averaging
across cell subpopulations with different function trajectories.
Mechanism: Cell-to-cell variation increases with age; averaging
masks decline in high-function cells by diluting with low-function cells.
Scope: Applicable to post-mitotic tissues (neurons, muscle).
Not applicable to proliferating tissues with ongoing stem cell input.
Rejection criteria: If single-cell measurements show no greater
age-related decline than tissue homogenates, reject hypothesis.
Contradictory findings resolution framework
CONFLICT RESOLUTION: [Citation A] vs. [Citation B]
Both studies claim:
- Study A: [Claim]
- Study B: [Opposing claim]
Potential explanations:
1. Population differences (species, age, sex)
2. Methodological differences (assay, timing)
3. Context differences (in vivo vs. in vitro)
4. Statistical artifacts (power, analysis)
HYPOTHESIS to resolve:
We hypothesize that [variable X] explains the discrepancy,
specifically [prediction].
From clinical observation to hypothesis
CLINICAL OBSERVATION:
[Describe unexpected pattern noticed in practice]
PATTERN TO EXPLAIN:
[What was observed]
LEADING HYPOTHESES:
1. [Potential explanation 1]
2. [Potential explanation 2]
HYPOTHESIS TO TEST:
We hypothesize that [specific mechanism], based on
[observations from basic science / analogous conditions].
Testable prediction:
[What would we expect to see if this is correct?]
Common pitfalls
- Vague predictions: "X affects Y" is not a hypothesis; "X increases Y by 20-30%" is.
- Unfalsifiable wording: Avoid "may," "could," or "might" when making predictions.
- Mechanical application: Don't force every observation into hypothesis form when it's better described as an exploratory finding.
- Missing mechanism: Without a plausible mechanism, predictions are arbitrary.
- Ignoring alternatives: Strong hypotheses specify what's being distinguished from.
- Unrealistic scope: Start with narrowly testable hypotheses before combining.
- Shoehorning: Don't try to find evidence FOR your hypothesis; design tests that could falsify it.
- Neglecting boundary conditions: Specify when your hypothesis should NOT hold.
Validation
How to know your hypothesis generation was successful:
- The hypothesis is stated as a specific, testable prediction
- Variables are clearly defined and operacionalized
- The hypothesis could be falsified by a plausible outcome
- There's a clear experimental design to test it
- Alternative explanations are acknowledged
- The scope and boundary conditions are specified
- The mechanism or theoretical basis is provided
- Effect size expectations are stated
- The hypothesis addresses a genuine gap or contradiction
References
-
Related ors- skills:*
- ors-scientific-thinking-brainstorming (for initial exploration)
- ors-scientific-thinking-critical-thinking (for evaluation)
- ors-scientific-thinking-perspective-tour (for multi-perspective framing)
- ors-scientific-thinking-failure-handling (if results are negative)
- ors-research-grants-specific-aims (for hypothesis in grant context)
-
External resources:
- Popper, K.R.. The Logic of Scientific Discovery
- Lakatos, I.. Falsification and the Methodology of Scientific Research Programmes
- CONSORT Statement - hypothesis reporting in trials
- STROBE Statement - hypothesis reporting in observational studies
Changelog
- 1.0.0 (2026-06-10): Initial adaptation by Pradyumna Jayaram, integrating Popper's falsifiability criterion, Lakatos's research programmes, and structured frameworks for converting knowledge gaps and contradictions into testable hypotheses.
