How do you read a one to many correspondence row?

A one to many correspondence row in ISIC Rev. 5, UN Statistics Division, means a single source category was split into several target categories, so the old total cannot be carried forward as a single figure. The reader must decide, from the definitions and any available detail, how the old total divides among the new categories, or must accept a broader aggregate that covers all of them. Treating a split as if it were a simple rename is the most common correspondence error.

Why is a split row harder than a renamed row?

A renamed row keeps the same basket of activity under a new label, so the old figure and the new figure describe the same thing and can sit in one series. A split row does the opposite: one basket has been divided into several, and the correspondence table tells you that a division happened without telling you how the contents divided. The old total is a sum of parts that now belong in different places, and without detail at a finer level, or without the definitions that show which activities went where, that sum cannot be reconstructed as separate new figures. Readers who miss this distinction produce trends that look continuous but are built on incompatible baskets. The first skill in reading a split row is therefore restraint: recognise that the row marks a place where a direct comparison stops, and that anything beyond that point needs either better data or a broader aggregate.

What are the legitimate ways to handle a split?

There are three, and the right one depends on what detail is available. If fine detail exists, for example survey data coded at a level below the split, the old total can be divided according to that detail and the division documented. If no such detail exists but all the target categories sit inside one broader category, compare at that broader level instead, and footnote the loss of detail. If neither is possible, present the series with a break: show the old figure up to the revision, show the new figures from the revision onward, and do not join them with a line or a growth rate. What is never legitimate is to split the old total by assumption, for example by dividing it evenly among the targets, or to carry the whole old total into one target and ignore the rest. Both moves invent data. The correspondence table is a map of structure, not a set of weights, and it gives no percentages to apply.

How do you tell a true split from a partial split?

Not every row with several targets is a clean split. In a partial split, part of the source category moves to a new target while the remainder stays in a category that also receives activity from elsewhere, so the targets overlap with other source rows as well. Reading such a row in isolation misleads, because the full picture only emerges when all the rows that feed into each target are read together. Trace each target backward to every source that contributes to it, and each source forward to every target it feeds, before drawing any conclusion about a single row. A target that receives activity from two sources cannot be compared with either source alone, and a source that feeds two targets cannot be carried forward whole into either one. This network view takes longer than reading one row, but it is the only view that shows where comparability survives and where it ends.

Walking a split row in four moves

Move	What to do	What you learn
1	List every target the source row points to	How many new baskets the old basket became
2	Read the explanatory note for each target	Which activities each target now covers
3	Check whether one broader category contains all targets	Whether a broader comparison is still possible
4	Look for finer detail that could divide the old total	Whether a divided series can be justified at all

If move 3 succeeds and move 4 fails, the broader comparison is your answer, with a footnote. If move 3 fails and move 4 fails, a break in series is your answer, with the break labelled. Either outcome is a success: both preserve the meaning of the figures, while a forced detailed comparison would destroy it.

A split row, read this way, stops being a puzzle and becomes a clear instruction. It tells you where detail ends, where a broader view begins, and where honesty requires a break rather than a bridge.

This guide covers statistical classification reading only, and it makes no determination for any individual reader. It does not set a code, a principal activity, or a registration for anyone, and the code applying to a specific record needs to be confirmed with the national statistical institute or business register that holds the record.