Data Quality Analysis and Determination: Dimensions, Methods, and Standards

Data quality analysis and determination is the systematic process of measuring, evaluating, and verifying datasets against defined quality dimensions such as accuracy, completeness, consistency, and timeliness. Guided by international standards including ISO 8000 and ISO/IEC 25012, it delivers quantifiable evidence that data is fit for its intended purpose. This article explains the core technical principles, assessment dimensions, testing methods, applicable data types, industry applications, and frequently asked questions.

Technical Overview

Data quality analysis and determination refers to the systematic measurement, evaluation, and verification of datasets against defined quality criteria in order to establish whether the data is fit for its intended use. The process quantifies quality through measurable dimensions, including accuracy, completeness, consistency, validity, timeliness, uniqueness, and integrity, and produces objective evidence in the form of metrics, composite scores, and defect records.

The discipline is underpinned by recognized standards frameworks. ISO 8000 defines fundamental concepts and management processes for data quality, while ISO/IEC 25012 establishes a formal quality model for data and ISO/IEC 25024 specifies how the characteristics of that model are measured. GB/T 36344 provides corresponding evaluation indicators for data quality assessment. In measurement and laboratory contexts, ISO/IEC 17025 governs the control of data and the monitoring of the validity of results, and ISO 5725 defines trueness and precision terminology for measurement methods and results.

Modern determination practice combines rule-based validation, statistical analysis, and data profiling, increasingly executed through automated pipelines that support continuous monitoring rather than one-off audits. The final output typically includes a data quality score, dimension-level metrics, a classified defect register, root-cause findings, and prioritized remediation recommendations.

Testing Samples

The objects of assessment span all major data categories encountered in enterprise, industrial, and laboratory environments. The most commonly evaluated data types are listed below.

Structured Data

Relational database tables, data warehouse records, and tabular files such as CSV and spreadsheet exports with defined schemas. Typical assessment targets include transaction records, customer master records, product catalogs, and financial reporting datasets.

Semi-Structured and Unstructured Data

JSON and XML documents, API response payloads, message-queue events, system log files, and free-form documents, where structure exists but the schema is flexible or evolving. Assessment focuses on parseability, schema conformity, and content completeness.

Instrument, Sensor, and Time-Series Data

Outputs generated by laboratory analytical instruments, industrial sensors, IoT devices, and online monitoring systems. These datasets are evaluated for signal integrity, timestamp continuity, calibration linkage, and associated measurement uncertainty.

Master and Reference Data

Canonical records such as material masters, supplier lists, code sets, and reference tables, which serve as authoritative sources against which transactional data is validated and reconciled.

Testing Items

Assessment is organized around quantifiable quality dimensions, each expressed through one or more measurable indicators and target thresholds agreed with data owners.

Accuracy

The degree of conformity between recorded values and verifiable reference values, determined through comparison against certified reference materials, source documents, or authoritative reference datasets.

Completeness

The proportion of required values, records, or fields that are actually present, measured as record completeness, field-level completeness, and population completeness against expected volumes.

Consistency

The agreement of data across records, systems, and time, encompassing cross-field consistency within a record and cross-system reconciliation of shared entities.

Validity and Conformity

Compliance with defined formats, data types, value ranges, units of measure, code lists, and business rules, covering both syntactic and semantic conformity.

Timeliness and Currency

The freshness of data relative to its intended use, measured through update latency, record-age distribution, and adherence to expected refresh cycles.

Uniqueness and Duplication

The absence of redundant records representing the same real-world entity, expressed as a duplicate rate after entity resolution and record matching.

Integrity and Traceability

Referential integrity between related tables, orphan-record rates, and the availability of lineage, provenance metadata, and complete audit trails.

Precision and Measurement Uncertainty

For instrument-generated data: repeatability and reproducibility evaluated per ISO 5725, control-chart status assessed per ISO 7870, and measurement uncertainty estimated per ISO/IEC Guide 98-3 (GUM).

Testing Methods

Determination combines several complementary techniques, selected according to data type, volume, criticality, and the availability of reference sources.

Data Profiling

Automated statistical scanning of the dataset to establish baseline characteristics, including value distributions, null rates, minimum and maximum values, cardinality, pattern frequencies, and inferred data types. Profiling typically precedes rule definition and reveals latent defects such as format drift, hidden anomalies, and undocumented duplicates.

Rule-Based Validation

Execution of a defined rule set covering syntax checks (format, type, length), domain checks (allowed values and code lists), relational checks (referential integrity), and business-rule checks (logical and temporal constraints). Each rule produces pass and fail counts together with row-level defect records for root-cause tracing.

Statistical Analysis and Outlier Determination

Distribution analysis, trend analysis, and outlier detection using established statistical criteria. ASTM E178 provides standard practice for dealing with outlying observations, complemented by z-score analysis, interquartile-range screening, and Grubbs-type tests. Multivariate and machine-learning approaches extend anomaly detection to large, high-dimensional datasets.

Sampling-Based Determination

Where full-population scanning is impractical, statistical sampling plans based on ISO 2859-1 attribute inspection are applied, and quality levels are expressed through acceptance quality limits and estimated defect rates with stated confidence bounds.

Cross-System Reconciliation

Record matching and balancing between source and target systems to quantify discrepancies in record counts, aggregated sums, and attribute values. This is a core determination technique for data migration, system integration, and regulatory reporting scenarios.

Continuous Monitoring and Scoring

Repeated execution of metric sets through automated pipelines, with dimension-level metrics normalized and aggregated into a composite data quality score. Results are trended over time and compared against defined alerting thresholds, enabling early detection of quality degradation.

Applications

Data quality analysis and determination is applied across sectors where decisions, compliance obligations, or automated processes depend on reliable data.

• Financial services: regulatory reporting, risk model inputs, anti-money-laundering screening, and customer due-diligence records.
• Healthcare and life sciences: clinical trial datasets, electronic patient records, and pharmacovigilance case data.
• Manufacturing: statistical process control, industrial IoT sensor streams, and supply chain traceability data.
• Analytical laboratories: control of test data and monitoring of result validity in accordance with ISO/IEC 17025.
• Retail and e-commerce: product catalog data, pricing records, and customer master data.
• Government and public sector: open data portals, census outputs, and statistical reporting.
• Data governance programs: master data management, data migration, and system consolidation projects.

FAQ

What is the difference between data validation and data quality determination?

Data validation typically checks individual records against rules at the point of entry, whereas determination provides an overall quantified assessment of a dataset across multiple quality dimensions, often including statistical inference about defect levels across the full population.

Which standards govern data quality analysis?

The principal frameworks are ISO 8000 for data quality concepts and management, ISO/IEC 25012 for the data quality model, ISO/IEC 25024 for its measurement, and GB/T 36344 for evaluation indicators. For measurement data, ISO 5725, ISO/IEC Guide 98-3, ISO 7870, and ISO/IEC 17025 are additionally applied.

How is a data quality score calculated?

Dimension-level metrics such as completeness rate, duplicate rate, and conformity rate are normalized to a common scale and aggregated through weighted averaging, with weights assigned according to the business criticality of each dimension for the intended use of the data.

What deliverables does a determination project produce?

A typical report includes profiling statistics, dimension-level metrics with quality scores, a classified defect register with concrete examples, root-cause analysis, confidence or uncertainty statements for sampled results, and prioritized remediation recommendations.

How often should data quality be assessed?

High-criticality data streams warrant continuous automated monitoring supplemented by periodic formal audits, for example quarterly. Static or archival datasets are commonly assessed at milestone events such as migration, major integration, or system decommissioning.

Can determination be performed by an external provider?

Yes. Accredited third-party laboratories and specialized service providers perform independent data quality assessments, which is particularly valuable for regulatory submissions, dispute resolution, and audits requiring impartial evidence of data fitness.