Agreement
Independent evidence supports the same narrow claim.
MODULE 5 · RELIABILITY & CORROBORATION · 40–60 MINUTES
Reliable for what claim?
“Is this source reliable?” is often too blunt to be useful. A stronger question is: which claim am I asking this source to support, and what makes it strong or weak evidence for that claim?
Claim-specific reliability
An eyewitness may be excellent evidence that they experienced fear, confusion or coercion. They may be much weaker evidence for the exact number of people present. An official dataset may be precise about what an agency counted, but say nothing about experiences the agency never measured.
A source can be strong for one claim and weak for another at the same time.
Make the claim narrow before you corroborate it
“The policy was bad” is too vague to corroborate. “Recorded unemployment fell by X measure during Y period” and “participants reported Z effect” are different claims requiring different evidence.
Before asking whether sources agree, write the smallest claim they are actually capable of testing. Otherwise apparent agreement may hide that each source is talking about something different.
Corroboration is relational
Corroboration means checking a claim against other evidence. Strong corroboration is not simply “more sources”. You need to know whether the sources are independent and whether they actually test the same claim.
Conflict
Sources disagree in ways that need explanation rather than averaging away.
Different scale
A personal account and a national dataset may answer different parts of the question.
Different method
Documents, interviews, statistics and observations may test a claim through different evidentiary routes.
Different perspective
Sources may describe the same event from positions that reveal different information.
False independence
Imagine five websites report the same statistic. Website 2 copied a newspaper. Websites 3 and 4 copied Website 2. Website 5 was generated by AI from those pages. You may have five carriers but only one upstream evidentiary chain.
Likewise, two academic articles written by different authors may both analyse the same dataset. They are independent authors, but not independent measurements of the underlying phenomenon.
Count evidence chains, not browser tabs or author names.
Practice 1: what would corroborate this?
For each claim, first rewrite it narrowly enough to test. Then identify two genuinely different forms of evidence that could bear on it.
- “Residents strongly opposed the redevelopment.”
- “The policy reduced unemployment.”
- “A newspaper campaign shaped public attitudes.”
- “Students experienced the rule as discriminatory.”
Then ask: are the evidence types truly independent, partly dependent, or testing different aspects of the claim?
Practice 2: draw the dependency
Report A publishes an official statistic.
Newspaper B quotes Report A.
Researcher C independently analyses the underlying administrative dataset.
AI answer D summarises Newspaper B and Researcher C.
Survey E collects new participant data about a related lived experience.
- Which carriers share an upstream chain?
- Does Researcher C count as fully independent of Report A? What depends on whether both use the same raw dataset?
- Does Survey E corroborate the official statistic, or answer a different claim?
- What would you draw as one chain, and what would you draw as a genuinely separate branch?
Practice 3: disagreement is evidence too
Source A: an official statement says an operation was orderly and necessary.
Source B: participants describe fear, humiliation and confusion.
Source C: a later historian argues that the official account understated the operation's social impact.
- Do these sources necessarily contradict one another on every claim?
- Which claim could A establish well?
- Which claim could B establish well?
- What extra evidence would C need for a broader causal interpretation?
- What does the disagreement itself reveal about power, purpose or perspective?
Practice 4: confidence ladder
Choose one important claim from your current subject. Build three levels:
- Low confidence: what is the minimum evidence currently supporting it?
- Medium confidence: what genuinely independent evidence or method would improve it?
- High confidence: what combination of provenance, method, independence and corroboration would make the claim difficult to dismiss?
Do not use “high confidence” to mean certainty. Historical and social-scientific conclusions can remain provisional even when well supported.
Year 13 extension: methodological independence
Two studies can be written by different researchers yet remain partly dependent if both use the same dataset, archive, survey instrument, coding scheme or theoretical assumption. Independence therefore has degrees.
Extension: compare “independent authors”, “independent datasets”, “independent methods” and “independent archives”. Which kind of independence matters most for your research question, and why?
Mastery check
Take one claim and write: “I would rely on ___ for this claim because ___. It shares / does not share an upstream evidence chain with ___. The strongest genuinely different check would be ___.”