Revisions

Published figures are not final figures

A statistical bulletin carries a publication date, but the numbers inside it sometimes do not stay fixed. Revisions — corrections or updates to previously published figures — are a routine part of official statistics, and understanding when and why they occur matters for anyone who cites disability data over time.

Revisions happen for several reasons. The most common is late data: a survey respondent whose record arrives after the initial processing run, or a weighting adjustment applied once the full sample is complete. Administrative sources are revised when a department reconciles its caseload figures against audited records. Census-derived population denominators are updated between releases as methodology improves, which ripples through every rate calculated from them.

The Office for Statistics Regulation expects producers of official statistics to publish a clear revisions policy, stating in advance what kinds of revision are routine and what would constitute a more substantial methodological correction. When a revision is made, it should be flagged in the bulletin itself — typically in a "revisions" or "quality and methodology" section — with a brief explanation of what changed and why. Producers are also expected to distinguish between a scheduled revision (anticipated, within the normal processing cycle) and an unscheduled revision (unexpected, often prompted by an error identified after publication).

For disability figures specifically, the stakes are raised by the fact that different series are already hard to compare across sources. A revision to the Annual Population Survey estimates of the disability employment gap, for instance, changes a figure that is already sensitive to question wording and reference period. If the revision is not prominently signalled, a reader comparing this year's bulletin to last year's may attribute genuine movement to what is in fact a data correction.

The standard mechanism for signalling is a "what's changed" note at the start of the bulletin, combined with revised figures in the data tables marked with an annotation — often a footnote code or a column flag. Some producers maintain a separate revisions log as a standalone document. The responsible approach, when citing a figure, is to check whether the table or dataset has been superseded since the date you first retrieved it.

Revisions do not undermine trust in published statistics; handled transparently, they are evidence that quality control is working. What erodes trust is a revision discovered by a third party rather than declared by the producer.