How does a merge differ from a split?

A merge and a split are opposite row patterns with opposite consequences for reading data. In a merge, several old classes point to one new class, so the new heading is broader than any single source. In a split, one old class points to several new classes, so each new heading is narrower than the source. The merge hides detail that the old structure showed separately. The split reveals detail that the old structure hid.

A reader who confuses the two will combine series that should be separated, or separate series that should be combined, and will misread every comparison that crosses the change.

What changes for a dataset user?

For a dataset user, a merge means aggregation. Figures for the old headings must be added together, with care for any activities that moved elsewhere, before they can be compared with the new heading. A split means disaggregation. A figure for the old heading cannot be divided between the new headings from the table alone, because the table does not state the shares. Splits therefore break simple comparisons more sharply than merges do. Eurostat states that classes for which some activities move are presented in the correspondence tables, and that backcasting and double reporting plans exist so that statistics can be presented under both versions for a period. Those plans exist precisely because neither pattern can be undone by reading the table alone. The table shows the link. The published backcast or double reported series shows the numbers on a comparable basis.

Pattern	Old side	New side	Effect on detail
Merge, many to one	Several classes	One class	Detail hidden, headings combined
Split, one to many	One class	Several classes	Detail revealed, heading divided
Unchanged, one to one	One class	One class	Detail preserved
Rearranged, many to many	Several classes	Several classes	Detail both hidden and revealed

The table above describes patterns only. It names no codes, because the pattern for a given heading is found in the published correspondence table for that heading, and a general table cannot substitute for that check.

What changes for a register entry?

For a register entry, a merge and a split raise different questions. A merge asks whether the combined note still describes the activity as recorded, since the new heading covers more than the old one did. A split asks which of the new headings describes the activity, since the old heading no longer exists as a single destination. The split question is harder, because the table offers more than one destination and the note for each destination must be read. The merge question is easier to state but still needs the note, because a combined heading can exclude an activity that one of its sources included. In both cases, the classification does not determine the entry. The national statistical institute or the business register that holds the record confirms the entry, and this guide does not determine any reader’s code.

How are the two patterns told apart in practice?

The patterns are told apart by counting destinations. One old heading with one destination is a continuation or a rename. One old heading with several destinations is a split. Several old headings with one destination is a merge. Several old headings with several destinations is a rearrangement. The count is taken from the correspondence table, not from titles, and not from memory of the old structure. Titles can stay similar across a split and can change across a continuation. Counts in the table are the reliable signal. After the count, the explanatory notes for every destination must be read, because a split destination that looks right in title can exclude the activity in its note, and a merge destination that looks too broad in title can include it in its note.

What is the shared limit?

The shared limit is that a correspondence row, in either pattern, is a category link, not a guarantee for a specific unit. The EUR-Lex summary states that the legislation only applies to statistical purposes. A merge does not make two businesses alike outside statistics, and a split does not divide a business. The classification groups activities for comparison. It does not decide licensing, tax, social security, funding or collective agreement questions. Those questions are answered by the authority or register that applies its own rules to the entry it holds. The table, the note and the institute record together give a complete check. Any one of them alone gives an incomplete one.

Merges combine and splits divide, so count the destinations first, read every destination note next, and confirm the entry with the institute or register that holds it.