Data without documentation is data that cannot be fully trusted or reused. If nobody knows what a field means, how it was collected, or when it was last updated, every use of that data carries uncertainty. Documentation is what transforms raw data into usable information.
What to Document
Good data documentation covers several layers:
- Field definitions. What does each field mean? What are the valid values? What units are used? This is the core of a data dictionary.
- Collection method. How was the data collected? Survey, administrative record, sensor, manual entry? Collection method affects how the data should be interpreted.
- Collection date and frequency. When was the data collected? Is it a snapshot or an ongoing record? How often is it updated?
- Source. Where did the data come from? Who is responsible for it?
- Known limitations. What are the known gaps, biases, or quality issues? Documenting limitations is as important as documenting strengths.
- Change history. What has changed over time? If a field definition changed, when did it change and why?
A data dictionary is a document that defines every field in a dataset: its name, data type, allowed values, and meaning. It is the single most important piece of documentation for any dataset. Even a simple spreadsheet listing field names and definitions is far better than nothing. Start there and build from it.
Documentation as a Practice
Documentation is most useful when it is created at the time data is collected or a system is built, not reconstructed afterward. Retroactive documentation is better than none, but it is always incomplete and often inaccurate. Build documentation into your data collection and management processes from the start.
Keeping Documentation Current
Documentation that is out of date is almost as bad as no documentation — it can actively mislead users. Assign responsibility for keeping documentation current. When a field definition changes, update the data dictionary. When a new data source is added, document it immediately. Treat documentation as a living artifact, not a one-time deliverable.
Making Documentation Findable
Documentation only helps if people can find it. Store it in a consistent, accessible location — alongside the data it describes, in a shared wiki, or in a data catalog. Link to it from the data itself where possible. If people do not know the documentation exists, it will not be used.
Key Takeaways
- Undocumented data cannot be fully trusted or reused
- Document field definitions, collection methods, sources, and known limitations
- Create documentation at the time of collection, not retroactively
- Assign responsibility for keeping documentation current
- Store documentation where people can find it
Some of the data you hold requires special protection. Learn the core principles in Protect Sensitive Information.