1. Definition of use

Question: What counts as use of the data?

Rationale

"Use" covers very different things. A paper builds its whole analysis on a survey. A strategy sets a target against an indicator. An article quotes a release. A report lists a dataset among its sources without drawing on it. An AI answer states a figure. Counting all of these together gives a number without a meaning, and a number without a meaning cannot be compared with last year's or argued with at budget time.

Two things have to be settled before any tool runs. The first is a definition that separates these kinds of use, so that a count says what it counts. The second is a table that says which dataset each mention refers to, because documents name the same survey in a dozen ways and none of them is the catalog's name. This chapter writes both, and it ends with the unit and the scope of counting, so that one period's report can be compared with the next.

Target state

• A written typology distinguishes mention from use, and primary use from secondary and background use.
• Every dataset has one canonical name, one acronym, one identifier, and the list of variants people write, in a table the catalog and the citations share.
• The unit of counting (documents, by default) and the scope (which sources, which period, whether the organization's own publications are included) are stated.

Maturity levels

Level	Steps
Foundational	Write the typology. Build the canonical names table for the most-used datasets. Decide the unit and scope.
AI-ready	Extend the table to the whole catalog with identifiers and variants found in documents. Publish the recommended citation on every dataset page. Record the typology in the extraction's usage fields.
AI-native	Maintain the table from the monitoring results, with new variants reviewed and added, and expose it through the catalog so that other systems match the same names.

Running example

The example organization's table has six datasets with their acronyms and the variants found so far, including "the labour survey" and "unemployment rate from the LFS". The typology marks a mention as used when the document builds on the data and as background when it cites the data as context.

Recipes

Recipe

1.1 Adopt a typology of use

Level

Foundational

Skills

Dissemination lead

Indicative time

Half a day

Result: the organization has a one-page definition of what counts as a use of its data, and every count and every report in this guide refers to it.

Without a typology, a document that names a survey in passing and a document whose whole analysis rests on it are counted the same. The typology separates them.

1. Distinguish a mention (the dataset is named or referred to) from use (the document's analysis, decision, or statement depends on the data). The program's extractor records this as the is_used flag.
2. Within use, distinguish primary (the main source of the analysis), secondary (combined with another main source), and background (cited as context). The extractor records it as usage_context.
3. Treat named mentions ("Household Budget Survey 2021") and unnamed mentions ("the labour survey") separately. Unnamed mentions need harmonization and are less certain.
4. Add the uses that documents do not show: downloads and API calls, which access statistics cover (chapter 2), and AI systems' answers, which the AI visibility check covers (chapter 8).

Worked example. The example organization's typology fits on one page. It gives one example of each category from its own collection. The 2023 journal article that estimates poverty from the household budget survey is a primary use. The regional food security outlook, which cites the organization's food security report in its introduction and draws nothing from it, is a mention with the context "background". The news article that quotes the unemployment rate in its first paragraph is a use, because the article's statement depends on the figure, and it is marked "secondary" because the article also draws on an interview. The page ends with the rule that downloads, API calls, and AI answers are counted separately and never added to the document counts.

What this does not do. A typology states categories. Applying them is a judgement that the extractor makes with a confidence and that a person confirms on a sample (chapter 6).

Recipe

1.2 Build the canonical names table

Level

Foundational

Skills

Dissemination officer

Indicative time

Half a day for the main datasets; ongoing additions

Result: the organization has one list of the names people use for each of its datasets, and detection, harmonization, and the catalog all read the same list.

Documents name the same survey in many ways: "Household Budget Survey 2021", "HBS 2021", "the budget survey". The names table maps all of them to one identifier. It is a CSV with one row per dataset:

Column	Meaning
dataset_id	The catalog's identifier (idno) or the DOI, so that a matched mention links to the record
canonical_name	The official name, used in the recommended citation
acronym	The acronym, where one exists
producer	The organization that produced the dataset
variants	Other forms found in documents, separated by ;

dataset_id,canonical_name,acronym,producer,variants
EX-HBS-2021,Household Budget Survey 2021,HBS,National Statistics Office (example),household budget survey;HBS 2021;budget survey 2021;household budget survey (HBS) 2021
EX-LFS,Labour Force Survey,LFS,National Statistics Office (example),labour survey;labor force survey;quarterly labour force survey
FS_UNDERNOURISH_PCT,Prevalence of undernourishment,,National Statistics Office (example),undernourishment;prevalence of undernourishment series;undernourishment rate
LF_UNEMP_PCT,Unemployment rate,,National Statistics Office (example),unemployment rate from the LFS;national unemployment rate
EX-REP-2025-03,Food security report 2025,,National Statistics Office (example),food security report;national statistics office food security report 2025
EX-CEN-2014,Population and Housing Census 2014,,National Statistics Office (example),2014 census;population census 2014

1. Start from the datasets with the most downloads and the flagship products (surveys, censuses, main indicators, main reports).
2. Use the catalog's identifiers as dataset_id.
3. Add variants as they are found: acronyms with and without years, informal names, translations, and misspellings that recur.
4. Publish the recommended citation on each dataset page in the catalog (chapter 5 of the dissemination cookbook). The citation uses the canonical name, which makes future mentions matchable.

Worked example. The table of the running example has six rows. The row for the household budget survey has the identifier EX-HBS-2021, the canonical name "Household Budget Survey 2021", the acronym "HBS", the example organization as producer, and the variants "HBS 2021; the budget survey; household survey 2021". The row for the unemployment rate has the identifier LF_UNEMP_PCT, the canonical name "Unemployment rate", no acronym, and the variants "jobless rate; unemployment rate from the LFS". The variant "the labour survey", found in a news article, is added to the labour force survey's row after the review of chapter 7. Each dataset page in the catalog now shows the recommended citation with the canonical name and the DOI.

What this does not do. The table lists the organization's own data. Other organizations' datasets that appear alongside (the Demographic and Health Survey, the World Development Indicators) are tracked as co-use in chapter 7, without a row here.

Recipe

1.3 Decide the unit and the scope

Level

Foundational

Skills

Dissemination lead

Indicative time

One hour

Result: the unit of counting and the scope are written down, so that reports can be compared from one period to the next.

1. Count documents as the default unit, one per document per dataset. A document that mentions a survey five times is one use. Keep the mention count as a secondary figure.
2. State the sources in scope (chapters 4 and 5 list them), the period, and whether the organization's own publications are included. They are usually excluded from use counts and tracked separately.
3. State the languages covered. A source in the national language needs the extractor tested in that language.
4. Record the decisions with the typology. A change in scope is a break in the series of reports.

Worked example. The example organization counts documents and keeps the mention count as a secondary figure. Its scope statement lists the four open scholarly indexes of chapter 4, the publication pages of six ministries and the parliament, the national news archive, and the country reports of three international organizations. It covers documents dated 2023 onwards, in English and French. Its own publications are excluded from the use counts and listed separately. The statement is dated and filed with the typology. When the national research repository is added a year later, the report of that period marks the change as a break and shows the counts with and without the new source.

What this does not do. The unit and scope define what is counted. The coverage of the sources (chapter 4) and the recall of the extractor (chapter 6) decide what is found, and both are reported with the counts.

Implementation options

• Typology: the extractor's fields (is_used, usage_context, dataset_tag) as the recorded form; a one-page definition as the human form.
• Names table: a CSV under version control, or the catalog's own identifiers and aliases exported; the series record's aliases field in the World Bank schema holds variants.
• Identifiers: catalog identifiers and DataCite DOIs; a DOI makes citations countable through DataCite as well as through text mining.

World Bank examples

• Monitoring of Data Use records the usage flag and context per mention and harmonizes name variants ("WDI", "World Development Indicators", "World Bank Development Data") to one identifier.
• The dissemination cookbook places the canonical names table and the recommended citation at the foundational level of its monitoring chapter.

Resources

• Monitoring of Data Use: the extractor's fields (is_used, usage_context, dataset_tag) that record the typology.
• World Bank metadata schemas: the aliases field of the series record, where name variants belong.
• DataCite: DOI registration, which makes a dataset's citations countable through identifiers (chapter 3).
• Make Data Count: the research data community's definitions of usage and citation metrics, which the typology aligns with.
• Canonical names table: the template with dataset identifier, canonical name, acronym, producer, and variants.

Common mistakes

• Counting mentions as use. A dataset listed in a bibliography and a dataset the analysis depends on are different uses.
• A table without identifiers. Matched mentions cannot be linked to records or counted with DataCite citations.
• Scope that moves. Adding a source without saying so makes a rise in use look like growth.
• One language. Mentions in the national language go unfound if the table and the extractor are English-only.

Verification

• Typology. Two staff members classify twenty mentions independently and agree on most of them, and the disagreements have been used to refine the definitions.
• Names table. Every flagship dataset has a row with an identifier and at least three variants.
• Citations. Every dataset page shows the recommended citation with the canonical name.

Checklist

0 of 6 done

• Typology written: mention and use; primary, secondary, background
• Canonical names table with identifiers, acronyms, and variants
• Recommended citation on every dataset page
• Unit of counting stated (documents) and mention count kept
• Sources, period, languages, and own publications stated
• Decisions recorded; changes treated as breaks

Ticks are saved in this browser only.