Frame the decision
Write down the question, population, geography, period and intended use. Separate internal exploration from a public claim or individual decision.
GUIDE / PROTOCOL
Assess a public dataset before reuse
A catalogue entry helps you find data; it does not establish the quality of every record or its fitness for your question.
Direct answer
When can a public dataset support a conclusion?
Define the target population and period first. Check the producer, documentation, statistical unit, coverage, definitions, actual downloaded resource and licence. Test missing values and breaks in series. Preserve the version used and keep the conclusion within the data’s scope.
01 / Review method
Review method
Read the method
Identify the producer, primary source, unit observed, definitions, sampling if any, granularity and methodological changes. Compare stated coverage with your actual need.
Inspect the resource
Open the file or API, not just its listing. Check dates, columns, units, missing values, duplicates, geographic codes, aggregates and possible revisions. Match sample values to the documentation.
Decide and record
Record licence and attribution duties, identifier, URL, resource date and checksum, transformations and checks. Narrow the inference or seek another source if scope or quality is insufficient.
02 / Evidence to keep
Evidence to keep
Scope
Question, population, time and geographic coverage, statistical unit and known exclusions.
Provenance
Producer, dataset ID, methodological documentation, licence and access date.
Processed version
Resource name and URL, retrieval date, format, schema, checksum and version if available.
Repeatable check
Cleaning rules, missing values, consistency test, external comparison and limits of the claim.
PRACTICAL CHECK
Before reaching a conclusion
A frequent misreading and the check that helps avoid it.
Premature conclusion
A complete listing or high score is mistaken for validation of the observations.
Decisive check
Download a sample, compare units and dates to the method, and quantify missing values before drawing the conclusion.
AFTER THE REVIEW
Decide from the evidence
Two dates differ
The listing changed but the file did not. Preserve both dates and verify which resource holds the relevant observations.
A definition changes
Separate periods on either side of the break, record the method and avoid claiming a continuous trend without a justified adjustment.
Decision and review trail
Decision and review trail
What the check must help decide and what to retain so the choice can be reviewed.
Sufficient evidence
Scope fits the question, the file was inspected, method and licence are known, and limitations can be stated clearly.
Next decision
Publish a bounded claim with version and transformations; otherwise pause, supplement or change source.
OSINT / Official references
Choose the catalogue for the question
The portal points to a resource; the evidence still depends on its method and actual file.
data.gouv.fr
Why start hereFind French producers, licences, resources and publication dates.
Limit to rememberCheck the chosen dataset’s publisher, coverage, licence, method and last update.
Read source entryEurostat · Data Browser
Why start hereCompare European series, definitions and reference metadata.
Limit to rememberCheck method, revisions and coverage; the live database may not retain past versions.
Read source entryInsee · Séries chronologiques
Why start hereCheck a French series and its methods before combining figures.
Limit to rememberKeep the series code, unit, coverage, method and revisions; different scopes are not directly comparable.
Read source entryDetailed inspection cases
Typical situations for preparing a check. They do not describe completed assignments or actual observations.
A missing value is treated as zero
What to inspect
Read the variable dictionary, missing-value codes and collection notes. Compare empty fields, null values and zeros on actual rows before calculating.
Decision and action
Preserve distinctions between absence, non-applicability and observed zero. Impute only under an explicit method and keep the non-imputed result for comparison.
Record to retain
Keep codes, transformation rules, counts by status and the effect on the denominator.
Import changes identifiers and decimals
What to inspect
Compare a sample of the raw file with imported and re-exported values. Check leading zeros, decimal separators, dates and non-Latin characters.
Decision and action
Treat identifiers as strings when their form is part of their identity. Define number-parsing rules separately and repeat the round trip on boundary cases.
Record to retain
Keep the raw sample, import settings, before-and-after values and anomalies corrected or awaiting resolution.
A change comes from scope
What to inspect
Compare filters, population, units and version before interpreting variation.
Decision and action
Calculate within comparable scope and describe extensions separately.
Record to retain
Versions, scopes, differences and comparison rule.
Review before deciding
These questions complement the method. They help establish what is actually supported while retaining qualifications in the record.
Like-for-like comparison
Do compared values use compatible definition, unit, population, area and period? Retain filters, missing values and series breaks rather than inventing continuity.
View a worked decision
Fictional example: the state applies only to the criterion and scope described.
- Fictional observation
- Two values are joined by year: one counts people, the other thousands of people in another territory.
- Declared state
- Observed discrepancy
- Decision and next check
- Harmonize unit and territory, then check definition and population. Unit conversion alone does not fix scope.
Recoverable extraction
Do the exact distribution, version, licence and transformations permit recalculation? Recheck a series revision before reusing an old chart.
View a worked decision
Fictional example: the state applies only to the criterion and scope described.
- Fictional observation
- The calculation retains its result, but resource version, filters and transformations are undocumented.
- Declared state
- Not checked
- Decision and next check
- Reproducibility remains unverified. Locate the exact distribution and rebuild steps before reuse.
Record and observations
Is the catalogue update date distinguished from the observation period and version of the resource used?
Documentation consulted : W3C · DCAT 3 ·
View a worked decision
Fictional example: the state applies only to the criterion and scope described.
- Fictional observation
- The catalog record was updated in 2024; the record retains 2021 observations and resource version V2.
- Declared state
- Verified within this scope
- Decision and next check
- The three time dimensions are separated. Do not label the data as 2024 observations; recheck version after revision.
TOOLS / REGISTERS
Information to record
Define population, units and deduplication rules before counting. A missing value is not zero; retain the used extract to make the calculation reproducible.
- Access date and time zone
- Exact URL
- Dataset / series code
- Producer / publisher
- Coverage / exclusions
- Unit / calculation basis
- Method / definition
- Version / revision
- Licence / reuse rights
- Decision / next action
- Review owner
- Next review
03 / FAQ
Frequently asked questions
Does a catalogue quality score validate the figures?
No. It covers documentation, licence and other metadata. Check the file and method separately.
Is a recent update date enough?
No. It does not tell you which periods are covered or whether older observations were revised.
Can two producers’ series be combined?
Only after comparing units, definitions, geography, periods and methods; document mappings and breaks.
Which version should be cited?
Cite the exact dataset and resource, retrieval date, stable ID or URL, and transformations.
What if the licence is missing?
Do not assume unrestricted reuse. Check the producer’s conditions before the intended publication.
04 / Official references
Official references
Describe a context
Related checks
- Distinguish a dataset, distribution and service
- SCL: a class code is not a probability
- Wikidata: truthy selects a rank, not certified truth
- Outer joins: ON and WHERE do not retain the same rows
- Keys and constraints guarantee different things
- PROV provenance: connect version, process and responsibility
- World Bank API: distinguish three dates and recover the unit
- Repeated keys: predict row counts before joining
- Connectivity: distinguish people, fixed subscriptions and mobile lines
- Incomplete judgments: preserve unknowns and scope
- Several OCR documents: sum errors and reference lengths
- A corrected version and a new period are different relationships
DEEPER READING / PRACTICE
Assess whether a dataset fits the question
A dataset is available to download, but its coverage or definitions remain unclear. Assess whether it can answer the question, beyond simply opening the file.
Updated on .
Priority checks
Examine producers, dates, fields, units, coverage and missing values. Release method breaks and rules that limit comparisons.
Interpretation pitfalls
A large number of lines does not compensate for an inadequate scope. The presence of a digital column does not guarantee a comparable measurement between observations.
Working record to retain
The evaluation sheet combines question, coverage, useful variables, defects and usage conclusion. The necessary transformations are listed before the calculation.
PRACTICAL CASE / WORKED EXAMPLE
A large file covers a different population
The pedagogical example concerns the preparation or interpretation of data. It shows why a technically possible calculation should remain linked to verified definitions, versions and transformation rules.
Example context
A dataset contains many observations and seems suitable. However, his documentation shows that he excludes an important category for the question studied.
Suggested approach
Read coverage, definitions and collection rules. Quantify the available elements and identify the exclusions. Search for a supplement or reformulate the question according to the actually covered scope.
Conclusion and next step
The evaluation describes the ability of the dataset to a specific use. The volume of data does not hide the population limit or the consequences on interpretation.
Questions for deeper practice
Can we use an imperfect dataset?
Yes, if its scope responds to a delimited question and the limits are explicit. Defects that prevent interpretation must lead to supplementing the data or reducing the scope of the result.
DETAILED WORKSHOP / APPLICATION
Test whether the dataset fits the question before calculating
01 / Prepare the necessary information
Prepare the question, population, period and necessary variables. Assemble files, documentation, versions and conditions associated with the resource.
02 / Carry out the approach
Examine coverage, fields, units, absent values and ruptures. Test some observations and identify the necessary transformations. Compare these possibilities to the truly sought scope.
03 / Examine an alternative situation
If the dataset covers only a part of the question, formulate a limited result or look for a supplement. Its volume does not compensate for an inappropriate definition.
04 / Decide on the next step
The dataset is set for use when definitions and coverage allow an interpretation. The known defects and treatment assumptions accompany the decision of use.
05 / Hand over the verification records
Transmitting qualifications, documentation and controls. The subsequent calculations may be based on already explicit rules and an identified version.
ANNOTATED WORKING DOCUMENT / REVIEW GRID
A list of data capacity to the issue
The file combines need, coverage, variables, version and defects. It indicates the necessary transformations and the limits that condition the use of the dataset before any calculation.
On a small screen, scroll the grid horizontally to read all three columns.
| Observation | Supported interpretation | Next step |
|---|---|---|
| The definitions and coverage meet the need. | The dataset can be retained for this use with its documented reserves. | Test the transformations and keep the version used. |
| A large number of people is absent. | The available volume does not cover the whole issue. | Reduce the scope or look for an appropriate supplement. |
| The fields are present but their meaning remains uncertain. | The format is not enough to make the values interpretable. | Find documentation and definitions before processing. |
Reviewing the working document
The final verification examines the suitability of the dataset if needed, not only its availability. Blocking defects must be distinguished from acceptable imperfections for a delimited and explicitly formulated question.
REPORTING / EXAMPLE WORDING
Presentation of results: Evaluation of the ability of a dataset to a question
Point to clarify
The ability of the dataset depends on the question and not only on its volume.
Before handing over
Read coverage, definitions and transformations before holding the dataset.
Example of a qualified conclusion
The resource covers the scope described with the indicated variables and limits. Larger uses require additional data or verification.