Contact

GlossaryPublic records, privacy, and data quality

Data dictionary

A data dictionary explains fields, codes, units, and missing values so public-safety records can be interpreted correctly.

A data dictionary is documentation that explains the fields in a dataset and how their values should be interpreted. It can identify what each column means, what a row represents, which codes are allowed, what units are used, and how missing or inapplicable values are recorded. It is the reference needed to turn a table of values into understandable information.

A dictionary belongs to a particular dataset, system, or version. A familiar column name is not enough to establish its meaning across agencies. “Date,” “priority,” “type,” and “status” can describe different events or classifications in different files.

What useful documentation answers

Start with the unit represented by one record. A row might be an incident, offense, victim, patient, dispatched unit, or update. The answer determines whether counting rows gives a meaningful incident total or duplicates events that have multiple associated records.

Next, examine identifiers and relationships. An incident identifier may repeat across unit records without being an error. A supposedly unique value may be unique only within an agency or year. The dictionary should explain enough to connect records without assuming that identical-looking identifiers always refer to the same event.

Field definitions should also identify time zones, measurement units, allowed codes, and special values. An empty field, zero, “unknown,” and “not applicable” can have materially different meanings.

San Diego's police calls-for-service page illustrates this documentation in practice: it provides field descriptions and separate references for call types, dispositions, and priorities. Reading only the column list would miss some of the code meanings. San Diego calls-for-service documentation

Example

Hypothetical: An analyst receives a file with columns called incident_id, unit_id, dispatch_time, and arrival_time. The dictionary explains that each row represents a unit assignment, not a distinct incident.

Three rows share one incident identifier because three units responded. Counting rows would yield three assignments; counting distinct incident identifiers would yield one incident. The dictionary changes the interpretation without changing a single data value.

The documentation also explains that blank arrival times include canceled assignments. Treating each blank as zero travel time would create artificially fast responses. The analyst instead separates eligible completed responses from records that cannot support the timing calculation.

Common confusion

A data dictionary is not merely a list of column names. A list can show structure while leaving meaning unresolved. Definitions, code meanings, and exceptions are what make the fields usable.

A codebook often emphasizes coded values and categories; a data dictionary often covers the broader field structure. Publishers sometimes use the names interchangeably, so examine the contents rather than relying on the title.

Documentation is also not a guarantee that every record is correct. It states how information is intended to be represented. Validation is still needed to find impossible dates, unexpected values, duplicate records, and other departures from that design.

Match the documentation to the file

Record the dictionary's version and the date range of the data. Systems evolve, and a code can be added, retired, or redefined. The FBI's NIBRS User Manual, for example, is versioned and includes detailed definitions and reporting rules for particular data elements. This page uses version 2025.0 as a documented example, without claiming it is the latest edition. FBI NIBRS User Manual, version 2025.0

If a field remains ambiguous, name the unresolved question in the analysis. Do not silently substitute another city's definition. A reliable explanation should let readers trace a result back to the exact fields, code meanings, and counting rules used.

Sources