Evidence and sources · Lesson 1 of 3

Primary, secondary, tertiary — and grey

Follow every important number back to the place it was first produced. The trail often ends somewhere you would not have cited.

8 min read · Free — no account needed to read

A primary source reports original work by the people who did it: the trial, the survey, the interviews, the dataset, the archival document. It is where the evidence actually enters the world.

A secondary source describes, reviews or analyses primary work: a literature review, a meta-analysis, a textbook chapter arguing a position, a news article about a study. Good secondary sources are enormously useful — they are how you find the primary ones and how you learn the shape of a field. They are not where a specific claim should stop.

A tertiary source compiles and indexes: encyclopedias, handbooks, databases. Excellent for orientation, never a citation for a specific finding.

Grey literature is everything produced outside commercial or academic publishing: government reports, NGO studies, working papers, theses, conference proceedings, internal evaluations. It is often the only source of recent or local data, and skipping it is one of the main ways a literature review quietly becomes a review of whatever was easy to find.

Preprints are primary work posted before peer review. They are current and unfiltered, which is both the benefit and the cost.

The practical rule is not a hierarchy. It is: know which one you are holding, and never let a claim rest on a secondary source describing a primary one you have not looked at.

See it happen

You are writing that "60% of startups fail within three years" and you have four sources for it.

The read that arrives on its own

Four independent sources agree, so the number is well established and you cite the clearest one.

The same situation, with this idea

You follow each one back. Three cite the same consultancy blog post. The blog post cites "our analysis" with no method and no data. The fourth cites a national statistics agency, which turns out to measure business closures of all kinds — including voluntary ones and sales — over five years, not three. You now have one real source that says something related but different, and you write that instead.

After this, you can

count sources by where they originate rather than by how many pages repeat them.

The takeaway

Follow every important number back to the place it was first produced. The trail often ends somewhere you would not have cited.

Easy to take the wrong way

That secondary sources are inferior. A good review saves you months — it is a poor place for a *specific* claim to terminate, not a poor source.

Try this

Take one statistic you have used and trace it to its origin. Record how many hops it took and what the original actually measured.