The Percentage Is Correct. Is the Claim?

The spreadsheet says 40 percent last quarter and 50 percent this quarter. The draft says performance “improved by 10 percent.” The subtraction is neat, the sentence is plausible and the claim may still be wrong. The change is 10 percentage points, or a 25 percent increase relative to the earlier rate. The Office for National Statistics style guide↗ draws the same line: a percentage point is the difference between percentages, so 10 percent falling by one point becomes 9 percent, while a 1 percent fall gives 9.9 percent¹. Neither description explains what the rate measures. So check the denominator, the comparison, the period and the verb before you check the arithmetic, and write the claim no wider than the source.

Percentage errors often survive review because the visible number receives more attention than the nouns and verbs around it. A reviewer needs to reconstruct the full assertion: which population, what event, during which period, compared with what baseline and according to whose method.

The numerator and denominator, who was counted and who was excluded, the period and the method, all from the original table rather than a copied cell.

Open the original table or analysis rather than relying on a copied cell. Record numerator and denominator, exclusions, unit, period and sample definition, and read the footnotes and chart notes where exclusions tend to hide. Check whether the two periods use the same method. A percentage of survey respondents cannot quietly become a percentage of customers. A rate from completed service calls cannot represent every booking.

Reproduce the calculation in the review note. If 20 of 50 eligible cases met a condition, the rate is 40 percent. If the later result is 25 of 50, the rate is 50 percent. Show those counts when they may change how readers interpret the claim. A larger percentage from a small or selected group can be mathematically correct and editorially unsuitable for a broad statement.

Record whether the figure is final, provisional, modeled or forecast. The source date and the period described need separate fields. A report uploaded today might contain a result from last year, and a newly calculated forecast does not establish a result that has already occurred.

The comparison word, the population noun, the verb and any present-tense promise; each can make a correct number say something the source does not.

QuoteIt checks claims this way. Each fact from a source is stored with its subject, value, scope and period beside the passage it came from, and each claim in the draft is compared with that passage. A sentence that widens the population, drops a qualifier or turns an association into a cause is flagged for a person, even when the number itself is right.

Restate exactly what was measured, for whom and over which period, and drop any cause or general quality the source does not establish.

Use a fictional training example. A trial records on-time completion for 40 of 100 participating jobs in January and 50 of 100 in February. The draft says: “Our new process makes deliveries 25 percent more reliable.” The relative increase from 40 to 50 percent is 25 percent, but the source establishes a change in one defined completion rate across two months. It does not establish cause, all deliveries or a general concept of reliability.

A narrower proposal could be: “In the internal trial, the share of participating jobs completed on time rose from 40 percent in January to 50 percent in February.” Before publication, confirm the counts, definition of on time, trial boundary, source status and permission to disclose. If seasonal conditions or sample changes make the comparison unsuitable, remove it despite the correct arithmetic.

The stronger sentence may attract more attention, while the narrower one may need more explanation. That is an editorial tradeoff for the communications owner, not a reason to let the percentage carry implications the analysis cannot support.

The proposed wording, the counts, the calculation, the population, the period, the source page, the fact owner, the public-use decision and the approval state.

Keep a compact card beside the draft: exact proposed wording; numerator and denominator; comparison values; calculation; population; period; method notes; source page; fact owner; public-use decision; and approval state. Add any material qualifier the sentence must retain. If a reviewer changes the claim, ask them to update the card or return it for another evidence check. The BBC’s editorial guidelines↗ ask journalists for the same habit: statistics should be explained, put into context and, where appropriate, challenged, with important caveats and limitations explained³.

Finally, read the sentence without the number. “Our process makes deliveries more reliable” remains a factual assertion. Review that meaning with the same care as the calculation. The finished card should let another reviewer reproduce the arithmetic and explain why every surrounding noun, verb and qualifier fits the source. It slots into the wider checklist for fact-checking a press release as the step for numeric claims.