Publish the Research Trail, Not Just the Headline Statistic

“Seventy-three percent of leaders say…” may travel through social posts, sales decks, search results, and generated answers long after its questionnaire, sample, and field dates disappear. The communication asset earns trust when a reader can find the question wording, population, recruitment, exclusions, weighting, missing data, and the condition attached to the claim.

Write the research brief before collecting data. Name the decision the work should inform, the unit of analysis, primary measures, comparison, likely sources of bias, and minimum useful sample. If the data cannot answer the glamorous question, narrow the question rather than decorating the result.

Store the raw input, cleaned dataset, transformation code or query, analysis output, chart data, figure, and publication as distinct versions. W3C PROV models entities, activities, and responsible agents and can describe when one artifact was derived from another. A team does not need a complex ontology to borrow the discipline: record who produced what, from which input, through which step.

Use immutable snapshots for published results and a dated release note for change. A dashboard that updates every morning can be useful, but it is a poor citation target unless the page shows the observation period and lets a reader recover the version behind an earlier claim.

Open with the finding and its boundary, then provide the method, sample flow, definitions, limitations, tables, downloadable values, author or reviewer, publication date, and contact for corrections. Put source notes beside the figure they govern. Do not send a reader through a lead form to learn the denominator.

Give useful figures descriptive alt text and a nearby table or data download. Provide citation guidance that includes title, organization, version, and date. DataCite’s metadata work is built around making research outputs discoverable and citable; a stable identifier is more useful when its landing page still explains the object.

Prepare a short finding, a chart, a methodology note, a subject-matter interview, and a newsroom contact. Every derivative should link back to the research page and carry the relevant period, population, and qualifier. Let journalists and analysts reach the person who understands the data rather than a spokesperson holding a headline alone.

Measure qualified citations, source-page engagement, dataset use, corrections, expert inquiries, relevant coverage, and commercial influence separately. A large pickup count can reflect a catchy statistic; it does not prove the method was understood or that the work changed a useful decision.