Statistics and research: statement on revisions, errors and postponements

Planned revisions can be due to methodological changes, re-basing or the availability of final data.

When considering revisions, the timeliness gained from issuing provisional data will always be balanced with the confusion and damage to users’ confidence that revisions can cause. Section heads and the ‘Head of the statistical policy and standards team’ should jointly decide whether to change or withdraw information, with the final authority resting with the Chief Statistician.

All outputs that include statistics subject to planned revisions will include a revisions statement, state whether the statistics are provisional or final and highlight any revisions. If there are any revisions, it will state:

• what they are
• why they have been made (a general statement rather than a description of each individual change)
• how they affect any previously published commentary or interpretation

Our release practices protocol will be followed when planned revisions are published. For example the output that will contain the revised statistics will be pre-announced and published at 9:30am on a weekday. The date will be chosen to avoid creating the perception that revisions are being released for any reason other than to simply correct the data.

In cases where a release or bulletin is revised to cover the same data and time periods as the original, the same reference number but with a suffix of (R) will be used. In cases where a release or bulletin updates data for previous time periods, but includes data for a new time period, a new reference number will be used. The revised data for previous time periods will be clearly marked as such.

In the case of planned revisions, for the sake of transparency, it will be made clear on the web page the date that updates have been made and what those changes were, and the output will clarify the detail of the change.

As far as possible the revised statistics will be released to the same audiences and via the same media. The same supporting information will be provided.

Non-trivial methodological changes, and their likely effect, will be announced before publication of any resulting revisions, to give users time to absorb the methodological issues first. Revisions will be monitored over time for systematic bias.

Continuing efforts will be made to provide users with access to the most up-to-date time series. Where appropriate a full time series will accompany each output. Regular releases will not be relied upon to allow users to compile their own time series, as this can lead to errors, especially if statistics are subject to revisions.

When revisions are made as much consistent back data as practicable will be provided, subject to avoiding any further delay in the release of final statistics.

The checklist attached at Annex A should be used whenever planned revisions are made.

Subject to the final authority of the Chief Statistician, section heads and the ‘Head of the statistical policy and standards team’ should jointly decide whether an error is significant enough to warrant changing or withdrawing the information. If so, the following steps should be taken. If the error is not significant, for example just a typographical error, the document should simply be corrected and re-posted to the internet.

If substantial or potentially harmful errors are found in an output a notification will be issued to alert users while the revised statistics are being prepared.

The revised statistics will be released as soon as practicable. In most cases unplanned revisions will be handled by producing a revised output covering the same data and time periods as the original, and so for releases or bulletins the same reference number but with a suffix of (R) will be used. In the few cases where a release or bulletin corrects errors in data for previous time periods, but also includes data for a new time period, a new reference number will be used. The revised data for previous time periods will be clearly marked as such.

As far as possible the revised statistics will be released to the same audiences and via the same media. The same supporting information will be provided.

In cases where an entire table is provisional, or is a revised version of a previously published table, the term “provisional or “revised” will be included in the table title, and no symbols will be used in the body of the table.

The following symbols will be used to denote provisional and revised data. No other symbols will be used.

• ‘p’ for ‘provisional’.
• ‘r’ for ‘revised’ (to apply to all corrections, whether planned or unplanned)

Revisions will generally be denoted at a cell level using the symbol ‘r’ in any table containing revisions. In cases where the revision applies to the majority of cells in an entire row or column of data, an alternative approach will be to mark the row or column heading with the symbol ‘r’, which will be separate from any other footnote applied. The symbol ‘r’ will be removed if the same data are unchanged when published in a subsequent output.

Provisional status will generally apply to a subset of data within a table, as given in an entire row or column. The relevant row or column headings will be marked with the symbol ‘p’, which will be separate from any other footnote applied. The symbol ‘p’ will be removed when data are finalised.

Both symbols will be clearly explained, either by means of a footnote to a table making use of them, or in a ‘notes’ section describing all the symbols used in the output.

The accepted Welsh translations for these symbols and meanings are as follows.

English: provisional (p); revised (r)

Welsh: dros dro (p); diwygiedig (r)
The symbols are not translated to avoid the need for multiple symbols in bi-lingual tables.

Postponements are where releases or bulletins are published later than the first pre-announced date.

Postponed outputs will be treated differently from revisions. A note to the relevant minister will be sent explaining the postponement. This will also be distributed to all internal staff that would usually be notified when data has been released. A new date for the output will be pre-announced. Our release practices protocol will be followed, including the normal restrictions on privileged early access.

• Follow the release practices protocol when publishing revisions.
• Publish revisions only when they significantly improve the ability of the statistics to inform good decisions. (This does not apply to planned revisions which will be published whatever their magnitude).
• Highlight the existence of revised data: add a statement to the web page of the relevant output.
• Announce substantial methodological changes before the release of the affected statistics.
• Consider issuing an experimental dataset based on past data to illustrate the changes.
• Correct errors as soon as practicable.
• Alert affected users to errors.
• Document the reasons for, and effects of, any substantial revisions.
• Include back data where practicable when revising a series.
• Ensure revisions are carried through to all relevant outputs, including for example any tables on StatsWales.