A data dictionary is structured documentation for a dataset: it names fields and records their meanings, formats, and permitted values. Read it before exploring or transforming data, because it states the assumptions attached to each column. section 2.1.1 section 3.2.2
Treat the dictionary as a starting hypothesis, not proof. Compare its definitions with actual records: check whether identifiers join as expected, whether claimed date ranges exist, and whether required fields contain missing values. When documentation is absent, first note the available data consistently across files, then create a dictionary for the fields used. section 3.2.2 section 5.2.2 section 6.4.1
A useful dictionary can support practical analytical decisions by documenting:
- Identifiers and relationships, such as customer, product, session, transaction, and location IDs. section 3.2.2 section 8.2.2
- Time fields and their granularity, such as an event timestamp, count date, or hourly interval. section 4.2.2 section 8.2.2
- Categorical codes and their meanings, such as event types, property types, tenure, and new-build status. section 4.2.2 section 10.2.2
- How collection instruments map questions to answer columns, including questions represented across several columns. section 6.2.2
Restoring headings to transaction files
Yearly property files without headers can be loaded using names derived from the dictionary, including transaction ID, sale price, sale date, postcode, property type, and tenure. Parsing the date then makes it possible to verify that the combined data covers the expected years before analysis. section 10.2.2 section 10.4.1
Table 2.1 Data dictionary for the customer address dataset
|
Column
|
Definition
|
|---|---|
company_id
|
A unique identifier for each customer company in the dataset
|
address
|
A single field to store the customer address
|
total_spend
|
The total amount this customer has spent to date (in GBP)
|
This compact dictionary links a customer identifier, free-form address, and cumulative spend to the fields available for geographic analysis. section 2.1.1
