What does a correspondence table not guarantee?

A correspondence table for ISIC Rev. 5, UN Statistics Division, does not guarantee that the underlying data can be split exactly, and it does not guarantee that a category in one revision means the same thing as its linked category in another. It shows structural relationships between categories, while the actual distribution of activity inside a split category depends on the data, which may not be available at the needed detail. The table is a map, not a conversion of the data itself.

Why does a structural link not prove the data will follow?

A correspondence row records that two categories are related in structure: one was renamed, split, merged, or partly moved into the other. It does not record how much activity moved, because the table is built from definitions, not from survey returns. The distribution inside a split category depends on the real activities that were counted under it, and that distribution can only be known from data coded at a finer level, or from a fresh coding of the underlying records. Without such data, the most the table can support is a statement about structure, and any attempt to turn the row into a set of numbers invents a distribution the table never contained. Readers who treat a link as a conversion factor therefore produce figures that look derived but are in fact assumed, and the assumption is hidden behind the authority of the table.

Does a linked category mean the same thing in both revisions?

Not necessarily, and the difference is often the point of the revision. A category may keep a similar title while its scope changes: activities may have been added from a neighbour, removed to a new category, or redefined by a rewritten note. The correspondence row shows the link, but the explanatory notes on each side show the scope, and only the notes reveal how much meaning survived the move. This matters most for long series, where a reader who assumes that linked categories are identical will read a change in scope as a change in the economy. The safe reading is narrower: the row says these categories are related, and the notes say how. Where the notes show a material change in scope, the comparison needs a break in series or a move to a broader level, even though the correspondence table provides a neat looking link. A link is an invitation to investigate, not a certificate of identity.

What else can a correspondence table not supply?

It cannot supply weights for dividing a split total, it cannot tell you when a country adopted the new revision, and it cannot reconcile a national classification with ISIC where the national structure adds detail or groups activities differently. National adoption timing is a separate fact, and it varies: two countries can both publish Rev. 5 figures while having switched in different years, so their series break in different places. A correspondence table also cannot fix a table that mixes levels, or rescue a figure quoted without a version label. It is one tool among several, and it works only alongside the explanatory notes, the reference periods, and the adoption history. A reader who asks a correspondence table to answer all of those questions will get confident looking answers to questions the table was never designed to answer, and the confidence is the danger.

Three claims to test before you rely on a correspondence row

First, test the claim that the data can be split: is there finer data, or are you assuming a distribution? If you are assuming one, say so, or do not split. Second, test the claim that the linked categories mean the same thing: have you read the notes on both sides, and do they describe the same scope? If they do not, mark a break or move to a broader level. Third, test the claim that the timing lines up: did the countries or periods you are comparing adopt the revision in the same year, and does your series mark the break where adoption actually happened? A correspondence row that passes all three tests can support a careful comparison. A row that fails any one of them can still be useful, but only as a description of structure, with the limitation stated plainly beside the figures that use it.

Read this way, the limits of a correspondence table are not a disappointment. They are the boundary that keeps a structural map from being mistaken for the territory of the data itself.

This guide is general information on statistical classification, and it settles no reader’s individual position. It does not determine a code, a principal activity, or a registration for any reader, and confirmation of the code on a particular record rests with the national statistical institute or business register that holds the record.