Data Quality Dimensions
In short A data quality dimension is a measurable aspect of data, such as completeness or accuracy. Choosing the right dimensions and measuring them regularly turns vague complaints about “bad data” into numbers you can track and improve.
Based on: DAMA International, DAMA-DMBOK: Data Management Body of Knowledge, 2nd ed., ch. 13 “Data Quality”, which summarizes L. English (1999) and DAMA UK (2013).
On this page
Key ideas#
Do not confuse these dimensions with quality attributes of requirements or software. Those describe how well a requirement is written or how a system behaves. Data quality dimensions describe the data itself. DMBOK admits there is no single agreed set and presents several.
English: inherent and pragmatic characteristics#
Larry English (1999) divides data quality characteristics into two groups: - Inherent: independent of how data is used. They include conformance to definitions, completeness of values, conformance to business rules, accuracy against the source and against reality, precision, absence of duplicates, and agreement between redundant or distributed copies. - Pragmatic: dependent on how data is presented and used. They include accessibility, timeliness, contextual clarity, usability, integrity of derived data, and correctness of the facts.
DAMA UK: six core dimensions#
In 2013 DAMA UK proposed six core dimensions: completeness, uniqueness, timeliness, validity, accuracy and consistency. It also named related qualities, such as usability, flexibility, confidence and value, without calling them dimensions.
A practical working set#
DMBOK condenses the versions into a working set. The measures below show one practical way to assess each dimension.
| Dimension | Question it answers | How to measure it |
|---|---|---|
| Completeness | Are required values present? | Share of records with every mandatory field filled. Set separate rules for mandatory, conditionally mandatory and optional fields. |
| Accuracy | Does the value match reality? | Compare a sample with an authoritative reference (an official register, a trusted directory, a physical count). Report the share that matches. |
| Consistency | Do values agree with each other? | Rule checks at three levels: within a record (end date after start date), across records (totals match details), and over time (no unexplained jumps). Report the share passing. |
| Timeliness | Is the data current enough? | Whether the copy is synchronized with its source (the only fully objective check), plus the delay between a real-world change and its recording, compared with an agreed limit. |
| Validity | Does the value fit its format, type and allowed values? | Share of values passing format, range and code-list rules. A valid value can still be wrong. |
| Reasonableness | Does the data make sense? | Compare distributions with history or a control set, such as sales by region against known demand. Flag results outside agreed tolerances. |
| Uniqueness | Is each real-world thing recorded once? | 1 − (duplicate records ÷ total records), using agreed matching rules. |
Two principles from DMBOK make these measures work in practice: - Set timeliness limits by volatility. Country codes may not change for years; share prices change by the second. One standard for all data makes no sense. - Calibrate reasonableness over time. Start with tolerances based on history or a control set, then refine them after the first rounds of measurement and track the trend. Some judgment is unavoidable, so agree the limits with the people who use the data.
How to apply it#
- Pick the critical data. Start with the few data elements whose errors hurt most.
- Choose dimensions per element. Not every dimension matters for every element.
- Write each as a rule with a threshold, such as “at least 98% of active customers have a valid mobile number”.
- Measure a baseline, then measure regularly and watch the trend.
- Find root causes of failures, typically at data entry, in integrations, or in missing rules.
- Give each rule an owner, usually a data steward.
- Turn thresholds into requirements for new systems, such as validation at entry or duplicate checks.
Template#
| Data element | Dimension | Rule | Threshold | How measured | How often | Owner |
|--------------|-----------|------|-----------|--------------|-----------|-------|
| | | | | | | |
Example#
The Workshop, a fictional bike-repair chain, needs reliable repair-status notifications. The analyst defines rules (all targets are illustrative): - Completeness: active customers with a mobile number, target 98%. - Validity: mobile numbers in the agreed national format, target 99%. - Uniqueness: duplicate customers across the 12 workshops, matched on name plus phone, target under 1%. - Timeliness: repair status updated within 15 minutes of the mechanic finishing, because customers act on the notification. - Accuracy: a monthly sample of 50 parts stock records checked against a physical count. - Reasonableness: daily bookings per workshop compared with the same weekday over the past eight weeks; a drop of more than half triggers a check of the booking system.
The baseline shows completeness at 85%: front-desk staff skip the phone field for walk-in customers. The fix becomes a requirement: a mobile number is mandatory for any booking that asks for notifications.
Pitfalls#
- Measuring everything → focus on the data that drives important decisions or processes.
- Treating validity as accuracy → a well-formed phone number can still belong to someone else.
- Fixing data without fixing the cause → trace failures back to where the data is created.
Sources and further reading#
- DAMA International. DAMA-DMBOK: Data Management Body of Knowledge, 2nd ed. Technics Publications, 2017, ch. 13 “Data Quality” (including its summaries of L. English, 1999, and the DAMA UK dimensions, 2013).