Audit a Research Data Dictionary Before Analysis

Compare the dictionary with the actual file

A data dictionary should explain the dataset that the analyst actually receives. Start by comparing the column names in a dated extract with the dictionary's variable list. Record undocumented columns, documented variables that are absent, and duplicate names. A document can look complete while describing an earlier export.

The UK Data Archive's quality-control procedures include dataset dimensions, variable and value labels, and checks for out-of-range values. These are useful foundations for an audit, but a study also needs definitions specific to its questions and collection process.

Give each field an operational definition

For each variable, record its name, readable label, unit, storage type, allowed values, relevant population, and measurement time. Distinguish the question asked from the answer code stored. If a field is derived, specify its inputs, formula, rounding, and handling of missing information.

Use a concrete definition rather than a familiar abbreviation alone. “Age” might mean age at consent, randomization, diagnosis, or extraction. A timestamp might represent specimen collection or result authorization. Resolving these distinctions early prevents apparently reasonable calculations from using the wrong time origin.

Test the codes against fictional observations

Imagine a fictional treatment-status field documented as 0 = not started and 1 = started. Its export also contains 9. Do not assume that 9 means missing: inspect the collection form or ask the data owner. If clarification establishes 9 = unknown, preserve that meaning in the dictionary and transformation rule.

Next, check whether a blank means unknown, not applicable, or an omitted response. Those categories should not automatically become the same analytic missing value without a recorded decision. A measured zero must remain distinguishable from all of them.

Review relationships between variables

Test rules that span fields, such as whether a follow-up date precedes baseline or whether a reported measurement has a compatible unit. Treat failures as queries, not automatic instructions to overwrite data. An unusual combination can reflect a legitimate exception or a misunderstanding of the dictionary.

Record the observed value, expected rule, source, proposed resolution, and responsible reviewer. Preserve the original extract so corrected records can be traced back to what was supplied. The variable-name audit guide addresses identifiers separately; a clean name does not guarantee a clear definition.

Finish with a versioned dictionary and a short unresolved-items list. Ask a colleague unfamiliar with the study to interpret several important fields using only the dictionary and source documentation. If they cannot reproduce the intended meaning, refine the definition before relying on that field in an analysis table.

Source checked on 1 November 2026; example records are fictional.