C.5. Revision policy
20.42. Recognizing that revisions of data on the international supply of services are often necessary, and given that dissemination of the revised data is essential for users, the present Guide encourages compilers to develop a well-designed, carefully managed and transparent revision policy that is well coordinated with other areas of statistics, thereby allowing users to cope with revisions in a systematic manner. A proper revision policy entails the following aspects:
(a) A detailed description of the revision policy on the responsible agency’s website, including the data to be revised, the frequency at which data is to be revised and any reasoning behind such revisions;
(b) A reasonable stability of timing of the revisions, clearly reflected in the data-release calendar;
(c) An early notification whenever a revision requires changes in the time series, going as far back as the beginning of the series, to retain methodological consistency, explaining the reasons for the revision and providing information on its possible impact on the data;
(d) Easy access to sufficiently long time series of revised data;
(e) The dissemination of all revised monthly, quarterly and annual data to ensure the consistency of all data available to users, including seasonally adjusted data and indices;
(f) Adequate documentation of revisions in the statistical publications and databases;
(g) Coordination of revision policy with data providers that might be the origin of large revisions;
(h) Establishment of an archival database to measure the size of revisions and generate quality indicators.
20.43. Data sources are often continually updated, and some data sources are received with a lag, making it necessary to revise data. Any revision has the potential to impact multiple sections and teams, all operating on different time periods and publications, so consistent treatment surrounding the revision process is necessary to maintain quality, coherence and the usability of the published data.
20.44. Revision policies should balance accuracy with usability. On one hand, it is desirable to account for each and every possible change in the data as soon as possible. On the other hand, frequent changes to a data set may make it difficult for users to get a clear picture of the underlying series and makes version control difficult. The policy should be practical to implement, and should itself be revised if necessary.