Lesson 3 — The Nine Common Data Quality Dimensions
A dimension is a quality characteristic. It gives you vocabulary for the kind of fitness problem, but it does not by itself define the exact test or acceptable level.
1. Validity — “Is this value allowed?”
Conformance to defined type, range, format, precision, allowed value/domain, or other formal rule.
Example: AGE=247 violates the permitted range.
Trap: valid does not mean accurate. john@example.com can match email format and still belong to the wrong customer.
2. Completeness — “Is required data present?”
Required values, records, or sets are present under applicable conditions.
Example: U.S. customers must have ZIP code; missing ZIP is completeness.
Completeness can be conditional: a debt-manager field may be mandatory only when account status is Debt Risk.
3. Consistency — “Do representations agree?”
Values/meanings are coherent across records, data sets, systems, or time where they should agree.
Example: source and warehouse balances differ when the same business definition requires equality.
4. Integrity — “Are relationships valid?”
Relationships among data are coherent, especially parent-child/reference relationships.
Example: every ORDER has CUSTOMER_ID populated, but some IDs have no Customer parent. The field is complete; the relationship lacks integrity.
5. Timeliness — “Did I get it when needed?”
Elapsed delay between capture/update and availability to a consumer.
Example: current balances arrive three hours after an 8:00 deadline.
6. Currency — “Is it still current enough?”
How recently the value was updated relative to its volatility and business need.
Example: a feed arrives exactly on time but contains two-year-old addresses that no longer satisfy currentness requirements.
7. Reasonableness — “Is this plausible?”
A value or aggregate is plausible against fixed limits, expected patterns, or statistical/benchmark ranges.
Example: a file normally has 5k–10k rows but suddenly has 2 million. That is evidence to investigate; it is not automatically proof of invalidity.
8. Uniqueness / Deduplication — “Is one entity represented once?”
The same real-world entity should not appear multiple times under the identification logic.
Example: two customer IDs share identity attributes and split one customer’s services.
9. Accuracy — “Does it match reality?”
The value correctly represents the real-world entity or event. Accuracy is often hardest to verify because it may require a trusted source, sampling, calibration, or direct real-world confirmation.
Four exam-critical pairs
- Validity vs Accuracy — allowed by rule vs true in reality.
- Timeliness vs Currency — delivery latency vs currentness/staleness.
- Completeness vs Integrity — required presence vs valid relationship.
- Consistency vs Uniqueness — agreement vs one-per-entity.
Additional trap: Reasonableness vs Validity — plausible vs formally allowed.
Source: pp. 429–432.