How do I correct a published statistic when the denominator was wrong?
Recreate the claim from the original source, identify who was counted, and update every place that depends on the result. Let a reader see what changed and whether the story’s conclusion still holds.
Auf Deutsch lesenStep-by-step walkthrough
Save the published claim and the version of the source used at the time.
Write a plain-language definition for the numerator and denominator. Check both with the data owner when the labels are unclear.
Have another editor rebuild the calculation independently.
Update the story and any related chart, newsletter correction, or homepage label that repeats the claim.
Read the correction as a reader who did not see the original reporting.
Free practice example
A fictional library story reports that 80% of all 100 branches extended hours. The survey actually covered 40 responding branches, of which 32 extended hours. The defensible claim is 32 of 40 responding branches, or 80%. The share among all 100 is unknown. Rewrite the headline using only what the survey supports.
How to check your result
A reviewer should be able to identify the response pool, reproduce the fraction, and explain why it cannot be generalized to every branch.
Questions you can work through
How can the arithmetic be right while the headline is wrong?
A percentage can accurately describe respondents while saying nothing definitive about nonrespondents. Check who could enter the denominator before revising the headline.
Where do I find the original universe of a survey?
Read the methodology, questionnaire and table notes together. If the document distinguishes invitations from completed responses, record both counts and which one each table uses.
What if the story calls households people?
Check the unit attached to every record. A household response counts as one observation unless the source provides a defensible person-level measure.
Can I fill missing responses with zero to fix the table?
Only if the definition explicitly establishes that a missing response represents zero. Otherwise keep missing values separate and explain the unknown part of the result.
Do I need a new chart when the bar height stays the same?
Change the population label, note and alt text. A bar labeled all branches communicates a different finding from one labeled responding branches, even at the same height.
How should a newsletter correction refer back to an old edition?
Identify the edition and claim clearly. Provide the corrected population and link to the updated story so subscribers can locate the complete explanation.
What if the corrected figure reverses the ranking?
Recalculate every comparison using the same methodology. Revise the conclusion and explain any remaining cases whose data cannot be compared.
Should I attribute an error to the source?
Check whether the error was in the source, the newsroom’s calculation or its wording. If the origin is uncertain, describe the observable change without assigning blame.
How do I avoid a second rounding error?
Calculate with the source counts and round only for publication. A count alongside the displayed percentage gives readers a way to check your work.
How do I audit the correction after publication?
Read the public article, chart and displayed teaser. Compare their population labels and results with the dated worksheet, and note any external copies your newsroom cannot edit.
Frequently asked questions
Can the exercise be done offline?
Yes. Write the two groups, 100 invited and 40 responding, on paper and calculate 32 divided by 40.
What if the story is still developing?
Separate corrections to an earlier mistake from new facts learned later. Time-stamp the change that readers need to understand.
Does this replace an editorial policy?
Use your publication’s correction process for approval and distribution. This exercise addresses the reasoning behind a numerical correction.
Sources
Checked
Continue with the masterprompt
Original product files are in German. EUR 37.50 one-time.
Related guides
- How do I audit an AI search journey before rewriting my website?
- How do I write a forum reply that actually helps?
- How do I find the questions people ask before hiring a service?
- How do I edit a bylined article without losing the author's voice?
- How can I tell whether a student got a hinge question right for the right reason?
- How do I write an oral history when two relatives remember different dates?
- How do I keep a mystery scene fair about what the viewpoint character knows?
- How can I test a training video storyboard before booking a shoot?
- How do I find open sales deals with no confirmed next step?
- How can I tell whether an original study actually supports my claim?
- How do I audit affiliate links that send readers to the wrong product or market?