Field entry, published 2026-11-03T00:00:00.000Z, updated 2026-11-03T00:00:00.000Z. Group: From ISCO-88 to ISCO-08, and revision

What does a revision mean for a long time series?

A revision means that a long time series gains a break at the switch year, because figures before and after that year were produced with different group definitions, and any change measured across the break mixes real employment change with reclassification. That mixing is the central fact to grasp. A series that runs smoothly through a revision year without a note is almost certainly hiding something, since at least some groups will have split, merged, or moved. The task for a reader is not to erase the break, but to understand its size, to bridge it where bridging is sound, and to describe it honestly where it is not.

Why does a break appear even when employment itself is steady?

Imagine a workforce that does not change at all between two surveys, while the classification changes between them. The same people, doing the same tasks, will still be counted differently if a group they belonged to was split, or if their work was moved to a neighbouring group. Totals at the highest level may stay roughly stable, because splits and merges often cancel out in aggregate, but detailed groups can jump sharply. That pattern, stable totals with jumping details, is a classic signature of reclassification. It appears because classification change redistributes counts among groups without creating or destroying jobs. A reader who looks only at totals may miss the break entirely, while a reader who looks at detail without knowing about the revision may invent an economic story for a purely administrative redistribution.

What bridging methods can a series builder use?

Series builders have several bridging methods, each with a trade off. Dual coding, where a sample is coded in both editions, gives the strongest bridge, because it shows directly how the same jobs map from old to new. Published correspondence tables give a weaker but still useful bridge, especially where mappings are one to one or where a split can be allocated from a sample. Backcasting applies the new structure to older microdata, where task detail survives, and can produce a long consistent series, but only where the older records contain enough detail to recode. Modelling can fill remaining gaps, but modelled portions must be flagged as estimates. No method restores detail that was never recorded, and a series builder who claims otherwise is overstating the bridge. The choice among methods should be documented group by group, since a method that works for one field may fail for another.

How large can the distortion be, and where is it largest?

Distortion is largest in detailed groups and in fields that changed most, such as technology, service, and specialised professional work, and smallest in broad totals and in fields that changed least. That gradient matters for how a series should be published. A publisher who offers only detailed groups across a revision, with no break note and no bridge, invites misreading. A publisher who offers broad groups with a break note, plus bridged detail where the bridge is sound, gives readers a usable series. Readers should therefore prefer publishers who show their working: a break marker in charts, a methods note naming the editions, and a table showing which groups were bridged and how.

What questions reveal whether a long series can be trusted?

Five questions reveal most of what a reader needs. Which editions were used, and in which years did the switch occur. Was any sample coded in both editions, and can its results be seen. Which groups have exact correspondence, which have partial correspondence, and which have none. Are bridged figures labelled as estimates where estimation was used. Is the break marked in charts and tables, or smoothed away. A series that answers those questions clearly can be trusted within its stated limits, even if some detail remains unbridgeable. A series that answers none of them should be used only for broad description, with the revision break stated by the reader rather than by the publisher. Trust here is not about the reputation of the source alone. It is about the transparency of the bridge.

How should a revision be described in a report or article?

Describe the revision in the same place as the finding, not in an appendix that most readers will skip. State the switch year, name both editions, and say whether the comparison crosses the break. If it does, say how the break was handled, and what uncertainty remains. If a chart crosses the break, draw the break visibly and label it, so that a reader scanning the chart sees the join before reading the numbers. That practice may make a trend look less smooth, but smoothness bought by hiding a revision is false precision. A long series that carries its break openly still tells a powerful story about decades of work. It simply tells that story with the honesty that a classification change demands.

Handled openly, a revision does not destroy a long series. It divides the series into comparable segments, joined where the evidence allows, and separated where it does not, and that division is itself part of the history of work.

Sources checked

1. ILO ISCO-08

Back to the From ISCO-88 to ISCO-08, and revision group or browse all occupations.