A draft report contains a tidy sentence: “38% of respondents preferred option A.” A footnote follows it, so the sentence looks finished.
Then a reader follows the footnote. It leads to a news article summarising a survey. There is no questionnaire, fieldwork date, sample description, or explanation of whether 38% describes everyone surveyed or one subgroup. The number may be accurate. The citation still leaves the reader unable to judge what it means.
A useful citation does more than show where a number appeared. It lets a reader find the original evidence, see what was measured, and recognise where your interpretation begins.
Start with the source closest to the number
A statistic can pass through several documents before it reaches your report. A government dataset may appear in an agency bulletin, then in a newspaper article, then in a presentation. When possible, cite the organisation or researcher that produced or released the data.
That source may be:
- the dataset or table itself
- the report containing the table and its method
- a survey's technical report or methodology page
- the original research paper
A news story can be a useful lead. It is rarely the best final citation for a figure. News coverage may shorten a definition, omit a revision, or report an early release. Follow the link trail back to the publisher.
The dataset and the report can have different jobs. Cite the dataset when you downloaded values, filtered records, or performed a calculation. Cite the report when you rely on its published analysis, explanation, or chart. Citing both is often clearest.
For example, suppose a labour agency publishes a monthly employment dataset and a bulletin about changes that month. If you calculate a three-month average from the dataset, cite the dataset and say that the average is your calculation. If you write that the agency links a change to seasonal hiring, cite the bulletin too.
Data publishers often provide a preferred citation. Use it as a starting point, then add the table, release, version, or access date that identifies the material you used. DataCite's guidance on citing data explains the value of persistent identifiers and version information. A dataset can be corrected, expanded, or replaced after its first release.
Name the exact item
“According to the Department of Transport” is attribution. It is not a complete citation for a statistic. It sends a reader to a large website with no clear route to the table you used.
Give enough detail to locate the exact item. Depending on the source, include:
- organisation or author
- title of the dataset, report, or survey
- edition, release date, or dataset version
- table number, figure number, or variable name
- publisher or repository
- DOI or stable URL
- access date, when data are regularly updated
A compact in-text reference might read: “Road traffic rose 4.2% from 2022 to 2023 (National Transport Agency, Annual Traffic Tables 2023, table 4).” Put the full reference in the report's reference list, using the required citation style.
Citation styles differ in punctuation and order. The identifying details should remain. APA Style's dataset reference examples show the information a data reference commonly needs.
Keep the context attached to the statistic
A percentage without a denominator is incomplete. A survey result without dates, population, and question wording can mislead even when copied exactly.
| Detail | Why it changes interpretation |
|---|---|
| Who was measured? | “Adults” might mean all residents, registered voters, customers, or service users. |
| When was it measured? | Views, prices, and conditions may have changed since fieldwork. |
| What was the base? | A share of all respondents differs from a share of people who answered one question. |
| How was it measured? | An opt-in online survey and a probability sample have different limitations. |
| What exactly was asked? | Small changes in wording can produce different answers. |
You do not need to place every detail in every sentence. Put the context that matters most beside the number. Give the full method in a note, caption, or source discussion.
Here is a usable pattern:
In a survey of 1,204 active account holders conducted in May 2026, 38% said they preferred option A when shown two options. The survey did not measure preference among non-customers.
The first sentence identifies the population, sample size, timing, and question setting. The second prevents an unsupported generalisation. Depending on the claim, readers may also need the response rate, weighting method, margin of error, sampling frame, or contact method.
For public-opinion surveys, methodology is not a decorative appendix. The American Association for Public Opinion Research's transparency standards describe disclosures that help readers assess survey evidence, including sampling and fieldwork information.
Do not silently change the measure
Writers often alter a statistic while trying to shorten it. “Revenue grew by 10%” is not the same as “revenue increased by 10 percentage points.” “Median income” is not “average income.” “More than half” does not necessarily mean a majority of the whole population if the result concerns a subgroup.
Keep the source's unit and definition. If you convert units, combine categories, or calculate a percentage, say so. For example: “Calculated from the agency's quarterly table by dividing category B by total valid responses.” This tells readers that the source supplied the counts and that you performed the calculation.
Separate the source's finding from your claim
A source can establish a measured result. It does not automatically establish the explanation placed beside it.
Compare these sentences:
- “The survey found that 38% of respondents preferred option A.”
- “Option A is more effective because 38% of respondents preferred it.”
The first reports a survey result. Cite the survey. The second makes an argument about effectiveness. Preference may be relevant evidence, but it does not by itself demonstrate effectiveness. The claim needs further evidence, a definition of “effective,” or more cautious wording.
Use verbs that show the boundary. “Reported,” “estimated,” “recorded,” and “found” describe a source's result. “Suggests,” “may indicate,” and “is consistent with” signal your inference. “We calculated” identifies a transformation you made from published data.
This also prevents you from assigning a conclusion to an author who did not make it. A table may show a pattern you notice. You can discuss that pattern, but label it as your reading of the data and explain how you reached it.
Make tables, charts, and notes traceable
A table or chart needs a source line even when the surrounding paragraph has citations. Visual material is often copied into slides, shared in meetings, or printed without the paragraph that introduced it.
For a reproduced table, a note might read: “Source: Author, Report Title (year), table 3.” Follow the citation style used in your report, including its rules for permissions and adaptation notices.
For a chart you created from another source's data, state the distinction plainly: “Source: Author's calculations using [dataset title], version 2.1.” If you removed missing values, adjusted figures for inflation, merged sources, or recoded categories, document those decisions in a methods note. A reader should be able to tell what the chart represents and what you changed.
Before submission, read the draft as a skeptical reader would. Can they identify the number? Can they find the original data release or report? Can they tell who was measured and when? Can they separate the published result from your conclusion?
A citation for statistics is a small map of how evidence moved from a dataset, survey, or report into your writing. A citation and similarity review of your own draft can help you find missing source details and check that borrowed evidence remains clearly attributed. Silvertext's features show the kinds of checks available.
