How sources are assessed

What counts as a usable source

• Peer-reviewed articles and academic monographs, read in the original rather than through a summary.
• Official statistical releases where the survey instrument and the sampling design are published alongside the results.
• Research reports from universities and research institutes where the method section is complete enough to judge the sample.
• Historical scholarship that identifies its archival sources, including work that disputes an earlier account.

What is excluded

• Press coverage used as a source for a finding. News articles may point to a study; the study itself is then the source.
• Surveys run by companies with a commercial interest in the result, and surveys with no published method.
• Figures that appear repeatedly online without an identifiable origin. A number that cannot be traced to a data collection is not printed.
• Preprints in place of published work, unless the article says explicitly that the paper has not been through review.

How disputed findings are handled

Most of the research covered here is contested in some way, and the disagreement is treated as part of the finding rather than as a caveat added afterwards. When a result has been challenged in the literature, the challenge appears in the same article, attributed to the people who made it. When two credible estimates disagree, both are given with the samples and periods that produced them, and the article says which questions the disagreement leaves open.

Where the honest answer is that the evidence does not settle the matter, the article says so plainly instead of choosing the more quotable side.

How findings are described

A finding is reported at the scope the study supports. An association observed in one cross-section is described as an association observed in that cross-section, not as a general rule. A sample drawn from one city in one year is named as such. Effect sizes are reported as the authors reported them, without conversion into more dramatic units.

No finding is extended into guidance. Research about how a group of workers was hired in 2004 says nothing about what any individual reader should do, and articles do not pretend otherwise.