How do you handle a dataset that still uses Rev. 4?

A dataset that still uses Rev. 4 should be handled as a Rev. 4 dataset, with its version label kept visible, even when the reader’s current work uses ISIC Rev. 5, UN Statistics Division. The practical options are to report the Rev. 4 figures separately, to convert them with a correspondence table where the data allow it, or to aggregate to a level that remains comparable. Silently relabelling Rev. 4 codes as Rev. 5 codes is the error to avoid, because it creates false precision.

Why is keeping the old label an act of honesty?

A dataset built on Rev. 4 does not become a Rev. 5 dataset because the reader wishes it were one, or because Rev. 5 is now the current standard. Its codes were assigned under Rev. 4 definitions, its categories carry Rev. 4 scope, and its totals add up on a Rev. 4 basis. Changing the label without changing that basis would claim a precision the data do not possess. Keeping the Rev. 4 label visible, as the opening paragraph insists, is therefore not pedantry. It is the difference between a dataset that can be used carefully and one that misleads everyone who touches it next.

Labels travel badly. A file copied, quoted, or merged into a larger dataset often sheds its documentation, and a label that was clear in the original release may vanish by the third reuse. Writing the version into file names, table titles, and chart footnotes, rather than leaving it in a methods document nobody reopens, is cheap insurance against that slow loss. The label should survive copying, because copying is what datasets are for.

When is reporting separately the right choice?

Of the three options in the opening paragraph, separate reporting is the simplest and often the wisest. Present the Rev. 4 figures as Rev. 4 figures, alongside any Rev. 5 figures, with the break between them marked. This suits short comparisons, descriptive work, and any case where the audience mainly needs to see the shape of each series on its own terms. Separate reporting refuses to manufacture comparability where the classifications do not support it, and that refusal is a strength, not a limitation.

Separate reporting also keeps the reader’s expectations realistic. A chart with a visible break invites the reader to ask what changed, and the footnote can answer. A chart that smooths over the break invites the reader to interpret a classification change as an economic event, which is the exact error the whole revision apparatus exists to prevent. A visible break is a feature of honest presentation, not an embarrassment to be designed away.

How does conversion with a correspondence table work?

Conversion is the second option, and it works only where the data allow it. The correspondence table maps Rev. 4 categories to Rev. 5 categories, and where a mapping is one to one, conversion is straightforward. Where a Rev. 4 category splits across several Rev. 5 categories, or several Rev. 4 categories merge into one, conversion needs more than the table. It needs underlying detail that shows how the old total divides, or a defensible rule for allocating it, and that detail is often unavailable in published aggregates. A conversion attempted without it produces figures that look detailed and are in fact invented in their fine grain.

A prudent converter therefore works in stages. First, map every category and mark each mapping as direct, split, merged, or retired. Second, convert only the categories whose mapping the data genuinely support, and aggregate the rest. Third, document every allocation rule and every category left unconverted, so a later reader can see where conversion stopped and judgement began. Conversion done this way is slower than relabelling, and unlike relabelling, it can be checked.

When does aggregating upward save a comparison?

Aggregation is the third option, and it is the fallback that rescues comparisons conversion cannot. By moving to a broader level, such as a section or a division total where the old and new scopes align closely enough, a reader can compare series that cannot be compared in fine detail. The cost is resolution: the comparison says less, but what it says is sounder. A broad comparison honestly described beats a fine comparison quietly fabricated, every time.

Aggregation also has a discipline of its own. The broader level must be chosen because its scope genuinely aligns, not because it is the first level that makes an awkward jump disappear. Check the correspondence rows at the broader level too, and footnote any remaining misalignment. If no broader level aligns well enough, the honest answer may be that the two series cannot be compared at all, and that the comparison should wait for a recoded series or be abandoned. That answer disappoints, but it does not mislead.

This is general guidance about handling statistical material only, not a ruling on any reader’s own code, principal activity, or registration. For a decision about a particular record, confirmation rests with the national statistical institute or business register that holds that record.