ISO 14044 Data Quality Requirements: A Practitioner's Complete Guide
ISO 14044 Section 4.2.3 specifies seven data quality indicators that every LCA study must address. This guide explains each indicator, the pedigree matrix scoring method, primary vs. secondary data hierarchy, and how to defend data quality decisions during critical review.
Why Data Quality Determines LCA Credibility
An LCA study can use best-practice modelling, a rigorous LCIA method, and a well-defined system boundary — and still produce misleading results if the underlying inventory data is unreliable. ISO 14044:2006 (confirmed 2016) recognises this risk and dedicates Section 4.2.3 to specifying the data quality requirements that must be documented in every LCA study.
Data quality is not binary. A single LCA model typically draws on dozens of datasets with varying levels of temporal relevance, geographic specificity, and technological representativeness. The practitioner's job is to characterise that variation honestly, understand its effect on conclusions, and communicate uncertainty in a way that decision-makers can use.
ISO 14044 Section 4.2.3: The Seven Data Quality Indicators
1. Time-Related Coverage
Definition:** The age of the data and the minimum length of time over which data should be collected.
Requirement:** Data should represent conditions during the time period relevant to the study. ISO 14044 does not prescribe a maximum age, but the standard states that data should be "representative of the time period covered by the study."
Practical implication:** Primary production data older than 5 years should be flagged for review, particularly in industries with rapidly changing energy mixes (e.g., electricity grids, aluminium smelting) or significant process efficiency improvements. Ecoinvent 3.x datasets carry a reference year that practitioners must compare to the study period.
Critical review trigger:** A reviewer may challenge a study that uses electricity grid emission factors from 2015 for a 2024 EPD of an EU product, because European grid GWP has changed materially due to renewable energy growth.
2. Geographical Coverage
Definition:** The geographic area from which data for unit processes should be collected to satisfy the goal of the study.
Requirement:** Data should represent the actual geographic context of the processes modelled. A European steel producer using a global average steel dataset when country-specific data is available represents a geographical quality gap.
Practical implication:** Production geography matters most for electricity (grid emission factors vary by a factor of 10 or more between coal-heavy grids and renewable-dominant grids), for agricultural processes (climate and yield vary strongly by region), and for waste treatment (landfill methane recovery rates differ by country).
Hierarchy for practitioners:** - Site-specific primary data (best) - Country-specific industry average - Regional average (e.g., EU-28, North America) - Global average (least preferred for geographically sensitive processes)
3. Technology Coverage
Definition:** The specific technology or mix of technologies required.
Requirement:** Data should represent the actual technology used in the production or process being modelled, not a market average if a specific technology is known and relevant.
Practical implication:** If a manufacturer uses electric arc furnace (EAF) steel production rather than basic oxygen furnace (BOF) steel, using a market-average steel dataset that blends the two will materially understate the environmental benefit of EAF production (EAF GWP is typically 60-75% lower than BOF when powered by low-carbon electricity). Technology specificity is a primary data advantage.
4. Completeness
Definition:** The percentage of flow data that is quantified for a unit process out of the total that should be quantified.
Requirement:** Data gaps must be identified and their potential impact evaluated. Cut-off decisions must be documented against the PCR's cut-off rules (typically 1% of mass, energy, or environmental relevance).
Practical implication:** Completeness failures are the most common reason EPDs fail critical review. Common gaps include: capital goods (machinery, buildings) excluded without justification, packaging of intermediate products omitted, and emissions from ancillary chemicals (cleaning agents, surface treatments) overlooked.
5. Consistency
Definition:** Whether the same study methodology is applied uniformly to the various components of the analysis.
Requirement:** Allocation methods, cut-off criteria, impact characterisation factors, and data vintage should be applied consistently across all processes within the system boundary. Mixed allocation approaches (e.g., economic allocation for one co-product, physical allocation for another without justification) violate consistency.
Practical implication:** When combining primary foreground data with ecoinvent background datasets, consistency requires using the same characterisation method (e.g., do not mix CML 2016 and ReCiPe characterisation factors in the same study). EN 15804+A2 mandates EN 15804+A2 characterisation factors throughout.
6. Reproducibility
Definition:** The extent to which information about the methodology and data values will allow an independent practitioner to reproduce the results reported in the study.
Requirement:** Sufficient documentation must be provided for an independent reviewer to replicate the key results. This does not mean publishing all proprietary data, but it does mean documenting the modelling choices, data sources, and calculation methods clearly enough that the review can be conducted.
Practical implication:** In practice, reproducibility is evaluated during critical review. The reviewer should be able to reconstruct the major inventory contributions and verify that reported GWP results are arithmetically consistent with the disclosed inventory.
7. Sources of Data and Their Representativeness
Definition:** Whether the data sources used are fit for purpose given the goal and scope.
Requirement:** Data sources must be cited and their representativeness assessed. Using a dataset labelled "European plastic injection moulding" for a process conducted in Southeast Asia fails the representativeness test.
The Pedigree Matrix: Quantifying Data Quality
The pedigree matrix (Weidema & Wesnæs, 1996, later formalised in the ecoinvent methodology) provides a semi-quantitative framework for scoring data quality across five indicators and translating those scores into additional uncertainty (variance) applied to each dataset.
The Five Pedigree Indicators
| Indicator | Score 1 (Best) | Score 2 | Score 3 | Score 4 | Score 5 (Worst) |
|---|---|---|---|---|---|
| Reliability | Verified data based on measurements | Verified data partly based on assumptions | Non-verified data based on measurements | Qualified estimate | Non-qualified estimate |
| Completeness | Representative sample, > 50% of sites, over adequate period | Representative sample, < 50% of sites, or shorter period | Representative sample of smaller sites | Incomplete data from only some sites | Unknown or very incomplete data |
| Temporal correlation | Less than 3 years difference | Less than 6 years | Less than 10 years | Less than 15 years | Age unknown or more than 15 years |
| Geographical correlation | Data from area under study | Average data from larger area in which area under study is included | Data from area with similar production conditions | Data from area with slightly similar conditions | Data from unknown or clearly different area |
| Technology correlation | Data from enterprises/processes under study | Data from processes produced with same technology | Data from processes produced with different technology in same sector | Data from related processes or materials | Data from unrelated processes |
Translating Pedigree Scores to Uncertainty
In the ecoinvent 3.x DQR (Data Quality Rating) approach, each pedigree score maps to an uncertainty factor (U2, the squared geometric standard deviation) that is added to the basic uncertainty of the dataset. The combined uncertainty is used in Monte Carlo sensitivity analysis to propagate uncertainty through the system model.
For a GWP result, this means: if a key input dataset has high pedigree uncertainty (e.g., temporal score of 5 because data is 20 years old), the Monte Carlo output will show a wide confidence interval on the final GWP, and the study's conclusions should be qualified accordingly.
Practical threshold:** Ecoinvent recommends that datasets with a DQR average above 3.0 across all five indicators should be prioritised for replacement with better data. In EPD contexts, critical reviewers often flag datasets with pedigree scores of 4 or 5 on geographical or technology correlation if those datasets represent more than 5% of total GWP.
Primary vs. Secondary Data: The Hierarchy for EPD Purposes
Primary Data
Primary data is collected directly from the processes being studied — site-specific, measured or calculated from metered inputs and outputs. It includes:
- Energy meter readings (electricity, natural gas, diesel consumption)
- Raw material purchase records (mass of inputs)
- Production volume records (units or mass of output)
- Waste transfer notes (waste quantities and treatment routes)
- Water meter readings
Primary data has the highest data quality for foreground processes because it is geographically, technologically, and temporally specific by definition. Under EN 15804+A2, primary data is required for modules A1-A3 (product manufacture) when the EPD declares specific product performance.
Secondary Data
Secondary data comes from external databases, literature, industry association reports, or other LCA studies. It is used to model background processes (e.g., raw material extraction, energy production, transport infrastructure) where site-specific data collection is not feasible.
Secondary data hierarchy for EPD purposes:**
- Ecoinvent 3.x — the most widely used background database, audited and documented
- GaBi Sphera datasets — commonly used in automotive and chemical sectors
- Industry association LCI databases (e.g., World Steel Association LCI data, Plastics Europe Eco-profiles)
- Literature-based datasets with documented sources
- Estimated or modelled data without published peer review (least preferred)
Critical review note:** Using an unverified estimated dataset for a process that contributes more than 10% of total GWP is likely to generate a reviewer comment. Always prioritise documented, reviewed secondary data sources.
Documenting Data Quality in Your LCA Study
ISO 14044 Annex B provides guidance on documenting data quality. A compliant data quality statement should include:
- A data quality table listing each significant unit process, its data source, the data year, geography, technology description, and pedigree scores across the five indicators
- A summary statement characterising overall study data quality
- Identification of data gaps and the method used to fill them (substitution, proxy, exclusion with justification)
- Uncertainty analysis results — at minimum a sensitivity check on the top 3 contributors to GWP; Monte Carlo preferred for comparative assertions
Common Data Quality Failures in Critical Review
Based on typical critical review findings under ISO 14044 Section 6 and EN 15804+A2:
- Outdated electricity factors: Using national grid emission factors more than 3 years old for electricity-intensive processes
- Wrong geography for transport: Using European road freight emission factors for Asian supply chain transport
- Missing upstream processes: Omitting chemical synthesis steps for specialty additives because "they are a small fraction of the product mass" without checking environmental relevance
- Inconsistent allocation: Applying economic allocation to one multi-output process and mass allocation to another in the same study without documented justification
- Unverified primary data: Reporting energy consumption figures from a single production run rather than a 12-month average, without noting the limitation
- Circular data sources: Using an EPD as a data source in another EPD without tracing back to the original primary or secondary data
Handling Missing Data Without Invalidating Results
When data is genuinely unavailable, ISO 14044 Section 4.5.3 permits the use of estimates and assumptions, provided they are:
- Clearly documented in the LCA report
- Justified as reasonable for the goal and scope
- Subject to sensitivity analysis (if the missing data affects a significant contributor)
- Disclosed in the EPD (for EPD-specific requirements)
The standard approach is to use a proxy dataset — a dataset from a similar process, geography, or technology — and apply a pedigree score of 3-4 on the relevant indicator to capture the additional uncertainty. The sensitivity analysis then tests whether replacing the proxy with the best available alternative (high estimate) and worst available alternative (low estimate) changes the study's conclusions. If conclusions are robust to this range, the missing data does not invalidate the result.
Practical Data Quality Workflow for a Manufacturing Company
For a company undertaking its first product LCA for EPD purposes, this sequence delivers a defensible data quality baseline:
- Identify foreground processes: Map every production step from incoming raw material to factory gate (modules A1-A3).
- Collect 12-month primary data: Pull energy bills, raw material invoices, waste manifests, and water bills for the most recent full calendar year. Normalise to your functional unit (e.g., per kg of product, per unit).
- Match to ecoinvent datasets: For each raw material and energy input, identify the corresponding ecoinvent 3.x dataset. Document the dataset name, version, and geography.
- Score pedigree for each significant dataset: Apply the five-indicator scoring for any dataset contributing more than 2% of GWP.
- Identify and document gaps: List any processes with no available dataset and document your proxy choices.
- Run sensitivity analysis: Test GWP sensitivity to the top 5 contributors by varying inputs +/- 20%.
- Draft data quality statement: Summarise findings in the format required by your PCR.
ClimatePoint and Data Quality Automation
ClimatePoint automates pedigree scoring for ecoinvent-backed datasets in the LCI compilation workflow, and the data quality results go into the LCA project report generated in the platform, by your team or ClimatePoint analysts, for the EPD verifier. Primary data entry is validated against plausibility ranges to catch data entry errors before they propagate into results.
If you are preparing an LCA study for third-party EPD critical review, book a consultation call at climatepoint.com to discuss your data collection plan.
