In-depth practical guides

Read an SDMX series without losing its dimensions

Identify the structure, codes and attributes of a statistical extract before comparing values.

Updated :

Statistical analysis workstation with illustrative charts.

AI-generated illustration

Do two values actually describe the same series?

SDMX supports statistical data and metadata exchange. An isolated value does not identify its population, unit or observation status. The producer’s structure explains the codes; a response format cannot replace that reading.

  1. Find the flow and structure

    Record agency, flow identifier and structure version. An omitted version may select the latest; retain the structure actually used.

  2. Break down the key

    Read dimensions in their documented order and consult code lists. Frequency, territory, unit or adjustment can change a series’ meaning.

  3. Keep attributes

    Retain unit, multiplier and status at their attachment level. Do not turn missing or provisional observations into final results.

  4. Freeze the extract

    Record full request, filters, period, Accept header, download time and raw file. Compare keys before numbers.

Reading workshop

Four dimensions before reading the value

The DEMO flow and codes are invented. Its contract orders the key as FREQ.REF_AREA.UNIT.ADJ; this order is not universal.

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

Same area does not mean same series

M means monthly, A annual; COUNT counts events, INDEX is an index; NSA is unadjusted, SA seasonally adjusted. Status is an observation attribute retained outside the key.

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

Same area does not mean same series
Complete DEMO keyPeriodValue and statusPermitted comparison
M.DEMO-A.COUNT.NSA2026-0380; provisionalRetain this unadjusted monthly observation with its qualification.
M.DEMO-A.COUNT.SA2026-0378; provisionalDifferent adjustment: not a temporal fall from 80 to 78.
A.DEMO-A.COUNT.NSA2026900; estimatedAnnual frequency: do not append as another month.
M.DEMO-A.INDEX.NSA2026-03105; finalIndex unit: do not subtract it from an event count.

Which rows directly form one homogeneous timeline?

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

All, since each contains DEMO-A

No. Frequency, unit and adjustment change. Area alone does not identify a series.

80 then 78 show a fall from March to April

No. Both observations concern March; only adjustment differs.

No temporal sequence is established by this table

Yes. Select a complete key and obtain compatible periods, retaining each observation’s status.

Fictional structure record

Flow: DEMO; local structure retained
Order: FREQ, REF_AREA, UNIT, ADJ
Selected key: M.DEMO-A.COUNT.NSA
Period: 2026-03; value: 80
Status: provisional, attribute retained
Decision: no temporal change calculated

Period identifies an observation within the series. A provisional attribute does not become final because the key is complete.

Documentation of the rules, separate from fictional examples

Compare other situations and exercises

Different units

In a fictional example, unit and thousand-unit tables show 120 and 0.120. The multiplier explains the difference; no economic change is established.

Open frequency filter

An illustrative request returns annual and monthly series. Separate keys rather than treating every row as one timeline.

From observation to decision

The same period, two editions and two growth rates

A fictional index series with unchanged definition and unit is accessed twice. The 1 September 2026 extract reports 100 for 2023 and 104 for 2024. The 1 October extract retains 100 for 2023 but revises 2024 to 106.

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

Revised estimates for the same interval

Scale from 0 to 8%. Each bar describes 2023–2024 growth in a different edition.

  • September 2026 extract4%
  • October 2026 extract6%
Case reading grid
ObservationUseful checkNext action
Extract dated 2026-09-012023: 100; 2024: 104.2023–2024 change: (104 − 100) / 100 = +4%.
Extract dated 2026-10-012023: 100; 2024: 106.Revised 2023–2024 change: (106 − 100) / 100 = +6%.
Same observation periodsThe extraction edition changes, not the 2024 period.Retain series key, structure, unit, attributes and each extraction date.
Your decision — Did the 2024 value increase by 2% between readings?

Explained answer

The gap from 104 to 106 is 2 index points, about 1.92% relative to 104. It revises the estimate for the same period; it is not new growth between September and October 2026. Read status codes in the producer’s structure rather than assuming universal meanings.

Reference: SDMX organises statistical data and metadata. This case compares fictional extracts, not an official series or code. SDMx — Statistical Data and Metadata

Worked exercise

Two open dimensions produce four series

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

Monthly frequency, two areas and two measures. Total and adjusted variants are not declared additive.

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

Starting information
AreaMeasureObservations
ATotal12
AAdjusted12
BTotal12
BAdjusted12

Your checks

  1. Expand dimension combinations.
  2. Build each complete series key.
  3. Check observations by key and period.
Read the answer

Answer and decision

There are 2 × 2 = 4 series and 48 observations, not a 48-month timeline. Separate all four keys. Total and adjusted are measure variants, not categories to sum without a definition allowing it.

If the case changes

Distinguish a nonexistent combination from an existing series with missing observations.

Record to retain

Keep structure, code lists, filters, full keys, attributes and value statuses.

DOCUMENTED REFERENCE

Eurostat slack: four components and an extended labour force

This historical extract fixes 2024, both sexes and two age bands for the EU, Spain, France and Finland. The 40 lfsi_sla_a observations were updated 10 September 2026 and retrieved 22 October 2026. They are annual data, not the latest available quarter. Published values and flags are retained.

Eurostat slack: four components and an extended labour force
ReferenceValue or formWhat it identifies
UNEUnemployed personsUnemployed people as a percentage of the extended labour force. This is not the usual unemployment rate, which uses a narrower denominator.
UEMP_PTUnderemployed part-timeUnderemployed part-time workers wanting and available for more hours under the series definition. They are already in the labour force.
SEEK_NAVLSeeking, unavailablePeople seeking work but not immediately available under Eurostat criteria, outside the usual labour force.
NSEEK_AVLAvailable, not seekingPeople available but not seeking work under the series criteria, outside the usual labour force.
SLACKPublished totalEurostat’s published slack total, retained separately from the sum of four rounded component values.
PC_ELF% extended labour forceThe denominator adds the two outside groups to employment and unemployment. Do not add underemployed part-time people a second time.
dDefinition differsDefinition differs: see metadata. All selected French and Spanish cells carry this flag; the response does not establish the national cause here.

Compare labour-market slack components in 2024

Retrieved · Response updated .

European Union (27) · 15–74 · 2024
Published total : 11.8 %
Component sum : 11.7 %
Gap: total minus sum (points) : 0.1

Percentage of extended labour force · 2024 · both sexes · European Union (27) · 15–74The bar sums four published components. Common 0–25% axis. Published total remains separate.0%5%10%15%20%25%Unemployment : 5.7 %Underemployed part-time : 2.4 %Seeking, unavailable : 0.9 %Available, not seeking : 2.7 %
Percentage of extended labour force · 2024 · both sexes UnemploymentUnderemployed part-timeSeeking, unavailableAvailable, not seeking
Percentage of extended labour force · 2024 · both sexes
TerritoryAgeUnemploymentUnderemployed part-timeSeeking, unavailableAvailable, not seekingPublished total
European Union (27)15–745.72.40.92.711.8
European Union (27)20–645.52.30.82.311.0
France15–747.2 d3.7 d1.6 d2.1 d14.7 d
France20–646.8 d3.7 d1.4 d1.9 d13.7 d
Finland15–747.94.43.32.317.9
Finland20–647.54.02.41.515.3
Spain15–7410.9 d4.4 d1.0 d3.0 d19.3 d
Spain20–6410.6 d4.4 d1.0 d2.8 d18.7 d

d: definition differs; see metadata. The flag accompanies every affected value.

The table retains all 40 observations and flags even without JavaScript. Gaps are not added as a bar category. No flag is not a comparability certificate.

Source: Eurostat. Sums, gaps, presentation and translations by this site; Eurostat is not responsible for these adaptations. Reuse conditions.

Implication for the record

For EU ages 15–74, the published total is 11.8%, while the four one-decimal components sum to 11.7%. The chart displays their sum without stretching it to the total. The gap is shown in points, not a fifth category. Rounded values alone cannot precisely explain every gap.

Why is the unemployment component not a national unemployment rate?

Its denominator is the extended labour force rather than the labour force alone. Comparing the percentages as equal-base measures creates an artificial difference. Keep PC_ELF and age in the caption.

Primary documentation

Documentation consulted on .

DOCUMENTED REFERENCE

JSON-stat: locate a cell before interpreting its value

In a JSON-stat cube, dimension and category order define the meaning of a flattened value array. The official order.json example contains 24 demonstration cells, not economic observations. The last dimension changes fastest.

A · 3× 8B · 2× 4C · 4× 1(1, 0, 3) → 1 × 8 + 0 × 4 + 3 = 11
JSON-stat: locate a cell before interpreting its value
ReferenceValue or formWhat it identifies
id[A, B, C]Dimension order A, B, C is part of the contract. Alphabetically sorting dimension names can relocate observations in another dataset.
size[3, 2, 4]Three A categories, two B categories and four C categories give 24 cells, even if a sparse value object lists fewer entries.
category.index0 … n−1Positions start at zero. Category code, display label and position are separate properties.
stride[8, 4, 1]Strides use subsequent dimension sizes: 2 × 4, then 4, then 1. They are not statistical units.
offsetiA × 8 + iB × 4 + iCPositions (1, 0, 3) produce flattened index 11. Reconstruct this address before pairing value and status.
value / statusindexMissing or null is not zero. Status information refers to cells and can also use a compact representation.

Explore the official file’s 24 cells

Choose one category per dimension. Positions start at zero; category codes happen to start at 1 here. The table preserves all 24 published values.

A1 · B1 · C1 → index 0 = 0 × 8 + 0 × 4 + 0; published value: A1B1C1.

Explore the official file’s 24 cells
Flattened indexCategory addressLiteral source value
0A1 · B1 · C1A1B1C1
1A1 · B1 · C2A1B1C2
2A1 · B1 · C3A1B1C3
3A1 · B1 · C4A1B1C4
4A1 · B2 · C1A1B2C1
5A1 · B2 · C2A1B2C2
6A1 · B2 · C3A1B2C3
7A1 · B2 · C4A1B2C4
8A2 · B1 · C1A2B1C1
9A2 · B1 · C2A2B1C2
10A2 · B1 · C3A2B1C3
11A2 · B1 · C4A1B1C4
12A2 · B2 · C1A2B2C1
13A2 · B2 · C2A2B2C2
14A2 · B2 · C3A2B2C3
15A2 · B2 · C4A2B2C4
16A3 · B1 · C1A3B1C1
17A3 · B1 · C2A3B1C2
18A3 · B1 · C3A3B1C3
19A3 · B1 · C4A3B1C4
20A3 · B2 · C1A3B2C1
21A3 · B2 · C2A3B2C2
22A3 · B2 · C3A3B2C3
23A3 · B2 · C4A3B2C4

Implication for the record

The official file consulted on 2026-10-22 has an illustrative inconsistency: index 11 contains the literal A1B1C4, although its category address is A2, B1, C4. The explorer preserves the source text. A value’s label must not replace dimension metadata.

Primary documentation

Documentation consulted on .

Record to retain

  • Flow, agency and version
  • Structure and code lists
  • Full keys and filters
  • Attributes, raw file and extraction date

Official references

Related source profiles

Related protocols

Checks before reaching a conclusion

  • Are all dimensions retained?
  • Are units and multipliers explicit?
  • Are observation attributes preserved?
  • Can the period filter be reproduced?

An extract linking each observation to its key, unit and status.

Frequently asked questions

Can a key be interpreted from its segments alone?

No. Their order and meaning come from the flow structure. The same short code can occur in different lists.

Does a CSV archive the request?

It retains values but may omit parameters or metadata. Store the request and relevant structures alongside the file.

Why keep the codes and their labels?

Codes stabilize identification and textbooks make it easier to read. Their dictionary and its version enable you to understand the categories and to find out the series despite an evolution of presentation.

TOOLS / REGISTERS

Information to record

Record full request, filters, period, Accept header, download time and raw file. Compare keys before numbers.

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

DOCUMENTED REFERENCE

Statistical adjustment: SA, CA and SCA are not interchangeable

Eurostat’s S_ADJ response checked on 22 October 2026 distinguishes six categories. In the diagram S means seasonal and C calendar adjustment: 1 present, 0 absent. The code describes treatment, not overall series quality.

NSAS=0 · C=0SAS=1 · C=0CAS=0 · C=1SCAS=1 · C=1
Statistical adjustment: SA, CA and SCA are not interchangeable
ReferenceValue or formWhat it identifies
NSAS=0 · C=0Unadjusted: neither seasonal nor calendar adjustment. Preserve the status when comparing with adjusted results.
SAS=1 · C=0Seasonally adjusted but not calendar adjusted in this list. Do not expand SA to mean every seasonal and calendar effect removed.
CAS=0 · C=1Calendar adjusted but not seasonally adjusted. Working-day treatment does not imply all recurring annual variation is removed.
SCAS=1 · C=1Both seasonal and calendar adjustment. Still check method, period, model and data vintage across producers.
TCTCTrend-cycle component, not a raw observation or forecast. Recent points can change as information and methods evolve.
NAPNAPAdjustment not applicable. Neither measured zero nor an automatic synonym of NSA.

Implication for the record

Compare identical categories and retain raw series when available. Re-estimated seasonal factors may revise earlier periods without raw measurements changing. Monthly adjusted and annual unadjusted rates do not share the same treatment.

Is it possible to remove s_adj from the key after selecting a series?

Keep its value in the record and chart metadata. Removing every trace makes differences between extracts of the same variable and period harder to understand.

Primary documentation

Documentation consulted on .

DOCUMENTED REFERENCE

ECB: nine historical rates to check a conversion

These nine observations cover three selected dates in 2025 and three currencies. Extracted on 22 October 2026, this sample is fixed: it is neither a current quote, an annual average nor every day of 2025.

Rates are currency units per euro. Divide a foreign-currency amount by the rate to obtain euros; multiply in the other direction. This calculation does not adjust inflation or include exchange fees or margins.

Fixed extract, consulted on .

Rate table · currency units per EUR 1
Observation dateUSD/EURGBP/EURJPY/EUR
1.03210.83118162.04
1.14190.8434162.98
1.1750.8726184.09

Method: M in currency ÷ T in currency/EUR = M/T in EUR. Reverse: M in EUR × T = M×T in currency.

Use a dot or comma as the decimal separator and spaces for thousands. Negative amounts are accepted to examine a correction.

1 000.00 USD gives 968.90 EUR. Observation on 2025-01-02: 1 EUR = 1.0321 USD.

Arithmetic illustration: the entered amount is an exercise, not observed expenditure. Results use two decimal places; calculations use the stored rate. The ECB publishes reference rates for information and discourages their use for transactions.

Series keys and extraction method

Frequency D: daily. Three separate dates, no interpolation of missing days. Quote is currency per EUR, with unit multiplier 0 in the responses. All nine values retain source precision; derived calculations are not additional official rates.

  • EXR.D.GBP.EUR.SP00.A
  • EXR.D.JPY.EUR.SP00.A
  • EXR.D.USD.EUR.SP00.A

Source: European Central Bank. Nine observations reproduced unchanged; conversions and presentation calculated by this site. The data are also freely available from the ECB.

Open ECB source · ECB reuse conditions

DEEPER READING / PRACTICE

Read the dimensions before extracting a series

A SDMX response contains values, dimensions and attributes. Reading must reconstruct the meaning of each observation before presenting it as a chronological series.

Updated on .

Priority checks

Identify structure, codes, frequency, unit and attributes. Check the filters and keep the constant dimensions that define the extracted series.

Interpretation pitfalls

A single value is not enough to know territory, unit or status. A filter change can select a different series without changing the name presented.

Working record to retain

The extraction sheet retains query, dimensions, dictionaries and dates. The values are accompanied by the attributes necessary for their interpretation.