In-depth practical guides

Check a statistical series break

Separate real change, definition changes and revisions before calculating growth.

Updated :

Statistical analysis workstation with illustrative charts.

AI-generated illustration

Check a statistical series break — method

Two values with the same label may measure different populations. Classification, coverage, collection or method changes may create a break. Metadata and flags are part of the result.

  1. Describe the series

    Fix code, unit, frequency, geography, population and seasonal adjustment. Record base period for an index and denominator for a rate.

  2. Read flags

    Download notes and observation status: provisional, estimated, break or missing. Missing is not zero. Check flag definitions with the producer.

  3. Choose comparison

    Compare within a homogeneous segment or use a producer-published backcast series. Do not splice segments with an improvised factor; explain excluded periods.

Reading workshop

When totals and coverage change together

The DEMO register is fictional. Its observed total grows while more units contribute to collection.

Fictional data and records for learning: none of these results comes from a live service query.

100 then 110: what change is being measured?

Period A covers 80 units and counts 100 events. Period B covers 100 units and counts 110 events. No common-scope subtotal is available. Events may repeat within one unit.

If the table exceeds the screen, scroll horizontally; arrow keys work when the region is focused.

100 then 110: what change is being measured?
CheckPeriod APeriod BConsequence
Published total100 events110 eventsThe raw change is +10%.
Collection coverage80 units100 unitsCoverage grows by 25%; units are not events.
Common scopeSubtotal not suppliedSubtotal not suppliedChange at constant coverage remains unknown.
Linking segmentsNo bridge tableNo backcast suppliedKeep two segments rather than invent a coefficient.

Which conclusion respects the records?

Open each proposition to compare its diagnosis. More checks may still be needed.

The phenomenon grew 10% at constant scope

No. The 10% calculation concerns published totals with different coverage.

The published total grows 10%, but a comparable trend remains undetermined

Yes. Retain the coverage break and request common-scope subtotals.

Dividing by the unit count automatically reconstructs the series

No. That creates another indicator and assumes an interpretation of composition. It is not documented linking.

Fictional comparability record

Series: DEMO register, repeatable events
A: total=100; covered units=80
B: total=110; covered units=100
Raw change: +10%; coverage: +25%
Common scope: subtotals absent
Decision: no adjusted change published

The two percentages refer to different objects. They cannot be added or used to decompose change without further evidence.

Documentation of the rules, separate from fictional examples

Compare other situations and exercises

Rebased index

Check whether only the base changes or method and scope change too.

Revised latest quarter

Compare extracts from one version and retain the old release when analysing revisions.

From observation to decision

An apparent 20% increase that needs qualification

A series is 100 before a definition change and 120 after it. The producer flags a break but supplies neither an overlapping year nor a backcast series.

Fictional learning case: counts and dates illustrate the method and do not describe a real assignment.

  1. 100old definition
  2. 120new definition
  3. ?comparable change unknown
Case reading grid
ObservationUseful checkNext action
Same title, changed definitionRead the note and break date.Separate the two segments visually.
Raw calculation: (120 − 100) / 100 = 20%Arithmetic does not correct a method change.Do not present 20% as a comparable change in the phenomenon.
No published linkSeek overlap or a homogeneous series from the producer.Restrict comparisons to compatible periods.
Your decision — Does dividing every new point by 1.2 fix the break?

Explained answer

No. It would force the two levels to match without evidence. A coefficient needs methodological justification and suitable overlap data.

Reference: Eurostat’s legend distinguishes a series break, provisional status and unavailable values. Read the legend of the dataset used; the case figures are fictional. Eurostat — Database: flags and coding conventions

Worked exercise

A new definition adds 30 units

Fictional example: the values and documents teach the method and do not describe a real organisation.

In B the definition includes a previously excluded category. Its count is known only for B.

If the table exceeds the screen, scroll horizontally; arrow keys work when the region is focused.

Starting information
PeriodItemCount
APublished total100
BPublished total140
BNewly included category30

Your checks

  1. Calculate the raw change.
  2. Remove the new category for a comparable B scope.
  3. State what this adjustment cannot establish.
Read the answer

Answer and decision

Raw growth is 40%. On the old scope in B, the level is 140 − 30 = 110, or +10% against A. The definition change accounts for 30 units in B; it does not prove that this category had 30 in A or allow reconstruction of the entire series.

If the case changes

If the subtotal of 30 is unavailable, mark the break without inventing an adjusted series.

Record to retain

Keep both definitions, the note, documented subtotal and raw/comparable results.

Record to retain

  • Code, dimensions, unit and frequency.
  • Extract version, flags and methodological notes.
  • Comparable segment, calculation and justified exclusions.

Official references

Related source profiles

Related protocols

Checks before reaching a conclusion

  • Check whether only the base changes or method and scope change too.
  • Compare extracts from one version and retain the old release when analysing revisions.

Frequently asked questions

Can growth cross a break?

Only with established comparability and an explicit method; otherwise separate periods.

Can a new download change past values?

Yes, revisions may change old observations. Date and version the extract.

When can two series segments be connected?

When a approach method is justified by data and documentation. The connection must describe its hypotheses and limits; a visual continuity is not enough to establish comparability.

TOOLS / REGISTERS

Information to record

Compare within a homogeneous segment or use a producer-published backcast series. Do not splice segments with an improvised factor; explain excluded periods.

  • Access date and time zone
  • Exact URL
  • Dataset / series code
  • Dimensions / filters
  • Reference period
  • Unit / calculation basis
  • Method / definition
  • Version / revision
  • Observed status
  • Observed passage / value
  • Limit / uncertainty
  • Decision / next action
  • Review owner
  • Next review

DOCUMENTED REFERENCE

HICP 2026: distinguish reference-period and classification changes

Eurostat introduced ECOICOP version 2 with January 2026 data and the 2025 = 100 index reference. Its note updated on 3 March 2026 also describes coverage changes. This is more than replacing a year in a table title.

HICP 2026: distinguish reference-period and classification changes
ReferenceValue or formWhat it identifies
2015 → 2025100Moving from 2015 = 100 to 2025 = 100 rescales the index. Pure rescaling preserves relative changes before rounding; it does not mean prices fell when the reference changed.
2026-01ECOICOP ver. 2Replaces the classification used until December 2025 and aligns with COICOP 2018 down to five digits.
12Insurance and financial servicesDivision 12 now covers insurance and financial services, not the former miscellaneous goods and services scope.
13Personal care, social protection and miscellaneous goods and servicesNew division 13 covers personal care, social protection and miscellaneous goods and services. Check detailed mappings rather than joining on division number.
09.4.7Games of chanceGames of chance enter HICP coverage in this class in 2026, separately from rescaling to 100.
prc_hicp_ecoicop11996–2025Old-classification datasets for 1996–2025 remain as frozen archives. Their last observation is not a failure to update the new system.
prc_hicp_ecoicop2ECOICOP ver. 2The new folder uses new codes for available series. Retain dataset identifier, classification and unit when replacing old extracts.

Implication for the record

Separate index levels, monthly rates, annual rates and annual averages. The reform combines reference, classification and coverage changes; check breaks and rounding before attributing every difference to prices. Identify backcast series.

May similar labels link old and new categories?

Not automatically. Use the official correspondence table and examine content changes at every level. If scope cannot be reconciled, show the break rather than misleading continuity.

Primary documentation

Documentation consulted on .

DOCUMENTED REFERENCE

France Travail: seven administrative categories, not unemployment rates

Insee’s definition updated 7 April 2026 describes A–G. Historical A–E alone no longer describe the current classification: F and G were created in January 2025. These classify registered people, not the entire labour-force survey population.

France Travail: seven administrative categories, not unemployment rates
ReferenceValue or formWhat it identifies
ANo employmentNo employment, required to take active job-search steps. A alone is not a complete ILO unemployment test.
B≤ 78 h / monthShort reduced activity with required job search: 78 hours or less during the month. Exactly 78 belongs here, not C.
C> 78 h / monthLong reduced activity above 78 monthly hours, with required job search. These people performed work.
DNo employmentNo employment and no active-search obligation, including defined training, placement or illness situations.
EIn employmentIn employment without active-search obligation; examples include subsidised contracts and business creators.
FSocial pathway · 2025People on a social pathway. Created January 2025; do not automatically relabel as unemployment.
GAwaiting orientation · 2025People awaiting orientation. A stock category alone does not establish an inflow or duration.

Implication for the record

Record the published grouping: A, ABC, all categories or another scope. Across January 2025 document system changes and any backcast series. Registered stocks and survey unemployment rates differ in scope, period and criteria; disagreement does not prove either is wrong.

Is ABC the ILO unemployed count?

No. B and C include people who worked, and surveys apply their own reference, search and availability criteria. Keep both definitions rather than convert categories individually.

Primary documentation

Documentation consulted on .

DOCUMENTED REFERENCE

COG events: before-after links are not just name replacement

The event file describes changes since 1943 and may use several rows for one transformation. A row is a before-after relationship, not necessarily an independent event. Historical examples in the 2026 documentation demonstrate lost information in one-to-one matching.

COG events: before-after links are not just name replacement
ReferenceValue or formWhat it identifies
DATE_EFFYYYY-MM-DDEffective date, separate from file publication and access. Keep annual geography vintage separately.
MOD=10Name changeName change may retain the code; preserve historical labels.
MOD=32New commune creationNew-commune creation may link both to the new whole and to a delegated successor.
MOD=33Merger-associationMerger-association: Belmont/Luthézieu on 1 November 1974 has three links, not three new communes.
MOD=35Delegated commune removalDelegated-commune removal does not automatically remove the parent or all territorial population.
Before / afterTYPECOM_AV + COM_AV → TYPECOM_AP + COM_APBefore and after types accompany codes. Same-code COM-to-COMD is meaningful, not a disposable loop.
Arbignieu example2016-01-01 · COM 01015 → COMD 01015Official 1 January 2016 example links Arbignieu to COM 01015 and COMD 01015; unchanged digits do not preserve territorial role.

Implication for the record

Administrative links alone do not supply population or area allocation weights. Choose target geography, track multiple links and check backcasting. Name changes need not change boundaries; mergers are more than relabelling.

May old codes simply be replaced with current codes and summed?

Only after checking measurement unit and overlaps. Multiple successors can duplicate a measure; stocks, flows and ratios require different aggregation rules.

Primary documentation

Documentation consulted on .

DEEPER READING / PRACTICE

Differentiate Real Change and Method Change

A series has a sudden rise or decline. The reading must search for an event in the data as well as a change of scope, collection or definition.

Updated on .

Priority checks

View notes, metadata and versions. Compare the associated variables and identify the exact time of change before calculating a trend.

Interpretation pitfalls

A variation may result from a methodological breakdown. Sliding the series or linking periods without justification can hide the impossibility of a direct comparison.

Working record to retain

The breakdown note indicates date, documented cause or hypothesis, comparable segments and limits. Calculations distinguish periods whose definition remains stable.