Applying Quality Control to Data
Applying Quality Control To Data
An interactive guide to quality assurance, quality dimensions, rules and metrics, validation, verification, profiling, reconciliation, peer review, audits, automated checks, calculations, and visual validation.
Why Data Quality Matters
Quality data supports quality reporting. Inaccurate, incomplete, or inconsistent data can undermine analysis and reduce confidence in decisions.
Quality Assurance
The process of ensuring data used in analysis is high enough quality to give decision makers confidence in the findings. Quality assurance is also commonly described here as data cleaning.
Validation
Checks whether data meets required standards, structures, formats, systems, and business requirements.
Verification
Confirms whether data is accurate or true. An unusual value may be structurally valid but still require verification.
Quality Checks
Quality checks should be routine and should continue throughout the project, especially after manipulation, transformation, or transfer.
A value can pass a format rule and still be wrong. Quality control therefore needs both validation and verification.
Five Dimensions of Quality
Data integrity is reflected through accuracy, completeness, and consistency, with timeliness and uniqueness also recognized as increasingly relevant quality dimensions.
Completeness
Are all expected and required fields populated?
Accuracy
Are the values in the fields correct?
Consistency
Are values entered in the same intended way according to business rules?
Timeliness
Is the data current and available within the expected timeframe?
Uniqueness
Are records or values that should be distinct free from duplication?
| Problem | Example | Dimension |
|---|---|---|
| Missing email | A required email field is blank. | Completeness |
| Wrong product name | An old product name remains after a rename. | Accuracy |
| Mixed credentials | MD, M.D., PhD, PHD, Lpn, LPN | Consistency |
| Week-old sales extract | The decision requires current sales. | Timeliness |
| Repeated email | Multiple records use the same email where uniqueness is expected. | Uniqueness |
Why Quality Problems Appear
Quality issues can originate in people, system design, transformations, transfers, or the combination of independently maintained data.
Mergers & Acquisitions
Organizations may combine data built under different standards, structures, and systems.
Manual Manipulation
Manual entry and editing create opportunities for human error.
Transformations
Transformations should be checked to confirm they produced the intended result.
Human Error
Wrong entries, calculations, joins, or other mistakes can be difficult to detect in large datasets.
Data Transfers
Analysts need to confirm that the intended data reached the intended destination without missing or unwanted records.
Rules, Metrics & Quality Targets
Governance teams may define quality rules and measurable standards. These metrics can function as KPIs for evaluating quality and tracking improvement.
Business Rules
Constraints on how data is captured can reduce invalid entry. Required names, valid email structure, and standardized location codes are examples.
Quality Targets
Different fields can have different quality goals based on business importance.
Pass / Fail
Analysts may report how many rows pass all checks and how many fail the required standard.
Illustrative quality score
1,000
950
95%
When data fails a quality check, investigate whether the entry itself was bad or whether the data did not conform to the required rules or process.
Data Profiling
Profiling data early establishes reasonable expectations and creates reference points that help analysts recognize suspicious results later.
Record Counts
Know how many records exist before and after movement or transformation.
Expected Totals
Knowing an expected aggregate can reveal incorrect math or joins when a result suddenly changes dramatically.
Fields & Types
Inspect fields and data types before relying on them.
Nulls & Redundancy
Profiling can include quality metrics such as missing values and redundant data.
Reconciliation, Cross-Checks & Peer Review
Quality control becomes stronger when results are compared against independent sources and reviewed by another person.
If accounting reports total sales of 35,000 while marketing reports 50,000, do not assume one is simply wrong. Their definitions may differ—for example, actual sales versus projected sales.
Automated Validation
Software can enforce rules at entry or during pipelines, quickly processing large volumes and reporting which records passed or failed.
Email Rules
Validation can require an @ symbol and an acceptable ending before an email is saved.
Format Rules
Phone numbers and dates can be constrained to expected formats.
Transfer Checks
Validation tools can report successful and failed records and often provide reasons for failures.
Versioned Rules
Rules should be version-controlled because requirements can change over time, such as when an organization expands internationally.
Automated validation cannot guarantee truth. A birth date can have the correct format while containing the wrong year, so verification is still necessary.
Verification Methods
Verification asks whether the data is accurate. Analysts should examine suspicious field values, record counts, calculations, and other evidence that can confirm the data.
look outside the norm
detect possible loss
spot-check logic
source vs. report
Field Values
Investigate values that appear outside the norm and consult the team responsible for the data when useful.
Data Loss
Records can be lost, incomplete, poorly named, or dropped during ineffective processes or incompatible transfers.
Calculation Review
Spot-check intended outcomes and verify that formulas and equations received the correct inputs.
Test & Explore Visuals
Quality control continues after the data is calculated. Reports and visuals must also be validated against business requirements and user expectations.
Correct Labels
Check titles, labels, text, and spelling.
Correct Sorting
A visual titled “Top 5 Products” should actually display the top five rather than the bottom five.
Business Meaning
Confirm the visual meets the intended requirement and makes sense to users.
Consistent Encoding
Use visual encodings consistently across the project; the same category should retain the same meaning wherever it appears.
Quality Control Workflow
Quality is not a single final check. Validation and verification should occur throughout a data project rather than being treated as a single final check.
Knowledge Check — 20 Questions
Review the key quality-control concepts covered throughout these notes.