What is asserted?
Rewrite it plainly.
Remove rhetorical framing until the core statement can be true or false.
Side 29
The Anatomy
of a Claim
A claim-dissection laboratory. Instead of asking whether a statement sounds persuasive, take it apart: proposition, scope, evidence, assumptions, qualifiers, causal language and what would count against it.
statement→scope→warrant→assumptions→test
06claim parts
05claim types
06overclaim traps
29Side
A sentence can hide several propositions.
Before evaluating evidence, identify exactly what is being asserted and which parts of the statement carry the inferential burden.
About whom or what?
Define the population.
Claims often imply a broader group than the evidence actually covers.
What property or relationship?
Describe, compare, cause, predict?
The verb often reveals which evidence standard is required.
How strong is the assertion?
Always, often, may, increases?
Small changes in qualifier can dramatically change what evidence is sufficient.
Under what circumstances?
When and where?
Context can be part of the claim even when it is left unstated.
Why should anyone believe it?
What connects evidence to proposition?
Evidence does not establish a claim without an inferential bridge.
Claim type determines evidence burden.
A descriptive claim, causal claim and normative claim can use the same topic while requiring very different forms of support.
| Claim type | Core question | Typical support | Common mistake |
|---|---|---|---|
| Descriptive | What is or was the case? | Observation, records, measurement | Generalizing beyond observed scope |
| Comparative | How do cases differ? | Comparable measures and sampling | Comparing unlike populations or definitions |
| Causal | What produced the difference? | Identification strategy, mechanism, counterfactual reasoning | Turning association into cause |
| Predictive | What will happen? | Out-of-sample performance, calibration, base rates | Confusing explanation with prediction |
| Normative | What ought to be done? | Values + factual consequences | Hiding value judgments inside factual language |
Most overclaiming happens in the distance between evidence and scope.
Scope determines which people, places, times and conditions the proposition actually covers.
Who is included?
Evidence from volunteers, one industry or one country may not represent a broader population.
When is the claim supposed to hold?
Relationships can change across economic, technological or institutional regimes.
Where?
Legal, cultural and environmental conditions can alter whether a relationship generalizes.
How much?
“Increases” can hide a trivial effect unless magnitude and uncertainty are stated.
Always or sometimes?
Universal claims are far more demanding than probabilistic ones.
Only under which circumstances?
Interactions and thresholds can make a claim true only in particular regimes.
Causal verbs carry heavy obligations.
“Causes,” “drives,” “leads to” and “because” imply a counterfactual: the outcome would have differed had the cause been different.
Temporal order
Did the cause precede the effect?
An outcome cannot normally cause an exposure that happened earlier, though anticipation can complicate timing.
Confounding
Could a third factor explain both?
Shared causes can produce strong association without direct causation.
Mechanism
How could the effect be produced?
Mechanistic plausibility strengthens interpretation but does not by itself identify causal magnitude.
Counterfactual
What would have happened otherwise?
Causal inference tries to construct or estimate the missing comparison.
Intervention
Would changing X change Y?
Some associations are predictive but not useful intervention targets.
A strong claim exposes itself to possible failure.
Testability improves when the statement makes clear what observation would count against it.
What would contradict the claim?
If no imaginable observation can count against it, empirical evaluation becomes difficult.
What follows if the claim is true?
Novel or risky predictions can discriminate between competing explanations.
What else explains the evidence?
A claim gains strength when serious alternatives perform worse on the same observations.
Would the pattern recur?
Repeatability tests whether the result survives new samples, methods or settings.
Where should the claim stop working?
Useful theories often specify their own domain of validity.
What would make you weaken the claim?
Precommitting to revision criteria reduces hindsight protection.
Claim lab.
Take compact statements and identify what is missing before arguing about whether they are true.
“Remote work improves productivity.”
Which workers, productivity measure, time horizon and comparison condition? Is the claim descriptive, causal or policy-prescriptive? What selection effects separate remote workers from others?
“AI reduces fraud.”
Which AI system, fraud type, baseline method and error trade-off? Does it reduce detected fraud, actual fraud losses or investigation time? At what false-positive cost?
“Country X has the best healthcare system.”
“Best” requires an explicit value function: outcomes, access, cost, equity, wait time, choice or some weighted combination. Without that, the statement hides a normative ranking inside a factual form.
“The project failed because leadership was weak.”
Operationalize “weak,” specify the mechanism, identify alternative explanations and ask what evidence would distinguish leadership effects from resources, incentives, timing or external shock.
The Craft of ResearchBooth et al. · claims, reasons and evidence
The Uses of ArgumentStephen Toulmin · claim, warrant and backing
Calling BullshitBergstrom & West · quantitative claim inspection
How to Read a PaperTrisha Greenhalgh · evidence appraisal