A correction to your study should identify the error, revised results, affected conclusions and methodology version. Preserve the earlier state and change history so readers can establish which version they are citing.
Separate a correction from a new study
A correction repairs an error in published material. New data or another measurement period may constitute a later study edition. Do not present a corrected old result as a new trend.
Give the study a stable identifier and version labels. Readers should distinguish the original, corrected version and later edition. A page-update date alone does not explain what changed.
Record the error and its reach
Before editing, preserve previous tables, methodology and data needed to reproduce the result within what can be shared. Record where the error arose, its cause and every dependent result.
Check table, conclusions, chart, download and translations. Correcting one number does not automatically update related materials.
Keep a methodology-change register
Record old and new versions, correction date, reason, changed elements and effect on conclusions. Distinguish classification changes, calculation fixes and a changed definition of the measured phenomenon.
If an indicator definition changes, explain whether comparison with earlier editions remains valid. Keeping the name does not ensure comparability. Do not combine differently measured results without explanation.
Worked correction example
This example is fictional. Study R contained 200 observations. Version 1.0 classified 24 into one category and reported 12%. Review found two misclassified observations. The corrected numerator is 22 while the denominator remains 200, producing 11%.
Version 1.1 should explain that two classifications changed, not sample size or period. The movement from 12% to 11% is a correction, not an observed change in audience behaviour.
Recheck conclusions depending on that value. If they remain valid, record why. Avoid “no impact” without checking dependent elements.
Put the notice where readers encounter the result
On the current page show the corrected version and link to history. Label downloads with versions so old and new files are distinguishable.
Mark earlier versions as archived and identify the replacement. Preserve context needed to understand earlier citations. If individual records cannot be shared, publish an adequate change description and impact scope without private data.
Keep translations consistent
Apply the same correction to every language. Numerators, denominators, period, version and reason remain identical. Translate the explanation, not the result.
If a language version is pending, state its actual condition rather than claiming full synchronisation. Check attachments and charts that may still contain old values.
Verify reproducibility
A reviewer should reproduce the percentage from its numerator and denominator and identify why it changed. Compare correction notice, table and conclusions. Retain approval ownership.
When testing study questions in AI, assess current version and correction reason. An old value can be appropriate in a historical context; the problem is presenting it as current without that context.
