How is the ISIC version used in a dataset identified?

The ISIC version used in a dataset is identified from the dataset metadata and its reference period, not from the code values in the table. ISIC Rev. 4 was in use since 2008 (FIXED), and ISIC Rev. 5 now replaces it as the internationally accepted standard, with its structure endorsed in March 2023 (FIXED) and its introduction and explanatory notes endorsed in March 2024 (FIXED). During the period in which national classifications and registers move to Rev.

5, datasets on both versions coexist, and a code string that appears in both revisions can mean different things in each. The metadata is the record that distinguishes them, and a dataset cited without its version is an incomplete citation.

Why do Rev. 4 and Rev. 5 datasets coexist?

Datasets coexist on different versions because classifications change on a timetable, and because historical series are compiled on the version in use when the reference period was produced. The United Nations Statistics Division states that the ISIC Rev. 5 publication is currently undergoing copyediting and preparation for publication, and that the final version will be provided when available. National classifications derived from ISIC move on their own timetables, and the datasets compiled from them follow the version of the register and the production cycle behind them. A dataset compiled for a reference period before the move will use the earlier version, even if it is published after the new version is endorsed. A dataset compiled for a later reference period may use the new version, or may be restated on the new basis through backcasting. The coexistence is therefore orderly, and it is managed through metadata, not through a single switch date that applies to every dataset at once.

What does the metadata state?

The metadata for a dataset should state the classification, the revision, the reference period, the statistical unit, and any note about a break, a restatement, or a planned move to a new revision. For an ISIC dataset, the classification is ISIC, the revision is Rev. 4 or Rev. 5, and the reference period is the period the data describe. A reader should record all three, together with the publisher and the publication date, before comparing the dataset with another. The Eurostat transition page describes the same discipline for NACE datasets, with parallel datasets and backcasting methods for the changeover from NACE Rev. 2 to NACE Rev. 2.1, and the discipline applies in the same way to ISIC datasets. A dataset that states its version in a footnote, a code list, or a separate methodology document is still a dataset with stated metadata, provided the footnote or document is cited with the figure. A dataset that states no version anywhere has not been checked, and it should not be compared with a versioned dataset.

How is a code string checked across revisions?

A code string is checked across revisions by opening the category in each revision, reading the title and the note in each, and recording whether the category is unchanged, split, merged, or moved. The United Nations Statistics Division classification detail for ISIC Rev. 5 gives the structure counts for that revision: 22 sections, 87 divisions, 258 groups and 463 classes. A Rev. 4 code string that appears in Rev. 5 with the same digits may have a different title, a different note, or a different place in the hierarchy, and a Rev. 4 category may have been split into two Rev. 5 categories, so that the same code string in Rev. 5 covers only part of the old category. The correspondence between the revisions, where published, shows the mapping type, and the mapping type must be recorded, because a one to many or split mapping does not guarantee a one to one result. A reader who compares code strings without opening the categories in both revisions compares strings, not categories, and the comparison is unreliable.

Check	Record
Classification	ISIC
Revision	Rev. 4 or Rev. 5, as stated in the metadata
Reference period	Period the data describe
Category	Title and note, in the revision stated
Mapping, where compared across revisions	Mapping type, from the correspondence between revisions

The table is a citation checklist for an ISIC dataset. A citation that answers all five rows can be repeated. A citation that gives only the code string cannot, because the same string can belong to two revisions with different meanings.

What happens when no version is stated?

When no version is stated, the dataset should be treated as unchecked, and the version should be sought from the publisher, the methodology document, or the code list that accompanies the dataset. A reader should not assume the latest revision, and should not assume the revision that matches a neighbouring dataset. The assumption that a dataset uses the latest revision is the most common version error in this cluster, because it feels like a safe default, and it is wrong wherever the dataset describes an earlier reference period or was compiled before the move. The honest record, until the version is found, is that the version is not stated in the dataset as consulted, with the date of the check recorded. That record prompts a further check, while an assumed version would close the check down on a guess.

What does the version not decide?

The version of a dataset does not decide the code for any unit, and it does not decide tax, licensing, funding, collective agreement or social security questions. The version is a property of the statistical classification used in the dataset. In the classification version the register uses, the entry for a specific business is confirmed with the national statistical institute or the business register holding the record. The meaning of a category in the dataset is confirmed in the publication for the revision stated in the metadata.

Record the revision, reference period and category note with every ISIC figure, and confirm a business entry with the institute or register that holds it. ”