27. Experts, Scientific Literature, and Quantitative Evidence

27.1 Separate the publication from the conclusion

A study's title and abstract can help identify it, but the method, sample, measurement, and limitations determine what it supports. Cite the relevant table, figure, appendix, or passage. If the proposition concerns an adjusted model, a citation to an unadjusted summary table is insufficient even though it is in the right article.

Identify the article's final published version, correction, or retraction status through the publication's own record where possible. A preprint and a peer-reviewed article may differ. Record the version used and avoid describing review status that has not been established.

27.2 Make numbers reproducible

Retain the unit, denominator, time period, population, source version, and any calculation you performed. “Forty percent” is incomplete without identifying forty percent of what. When you derive a number, distinguish your calculation from a number printed by the source.

Fictional example. A dataset contains 120 applications, 30 of which lack a final status. Reporting an approval percentage over all 120 and reporting it over the 90 resolved applications answer different questions. State the denominator and treatment of missing observations before citing the dataset.

27.3 Cite expert reports and underlying materials separately

An expert report can support a claim about the expert's opinion. The underlying study or dataset may support a different claim about observed results. Preserve both when the analysis depends on the connection. Identify report date, author, section or page, and any amended report. A deposition can qualify the report's explanation; use the testimony when that qualification matters.

27.4 Record transformations

For a spreadsheet or script, retain the input files, version, settings, formulas or code, and output. A citation to an agency dataset does not reveal that you excluded rows or changed categories. Describe material transformations in the text or methodology note. An output chart is not self-validating because software produced it.

27.5 Keep citation and admissibility separate

Expert admissibility, authentication, and evidentiary weight depend on applicable law and the record. This chapter's practice makes sources inspectable; it does not certify admissibility. The Federal Rules of Evidence are available from the judiciary, including rule 702 for expert testimony in its proper setting. Official Evidence Rules.

Before delivering an analysis, a reviewer should be able to reproduce every material number or identify the source's reported value. If reproduction fails, narrow the claim or resolve the discrepancy rather than adding a second citation to the same unexplained figure.