Interpreting LCA results: the ISO 14044 phase-four checks

Interpretation, the fourth phase of LCA, is the systematic process of evaluating the inventory and impact assessment results together with the goal and scope to produce conclusions, limitations and recommendations. ISO 14044 defines the phase in three steps: identifying significant issues based on the results of the earlier phases; an evaluation consisting of completeness, sensitivity and consistency checks; and reporting conclusions, limitations and recommendations. In most reports interpretation is the shortest section, yet the standard gives it equal weight with the other three phases and places it in an iterative relationship with them: interpretation may show that the scope needs narrowing or the data improving, and the study returns to that point. This article explains the three steps, what each check asks, and how to report results with their uncertainty.

The three steps of the interpretation phase

The first step is to find the issues that carry the result: which life cycle stage, which process, which flow and which impact category make up most of the total, and which methodological choice (allocation, system boundary, scenario) determines the result. The second step tests the reliability of those findings with three checks: completeness (is the necessary information present), sensitivity (how much does the result move when assumptions change) and consistency (are methods and data consistent with the goal and scope and between alternatives). The third step is to report the conclusions from those two steps together with the study's limitations, in language suited to the audience. The order matters: if sensitivity analysis is run before establishing what is significant, the study scans a hundred parameters and misses the three with leverage.

Identifying significant issues

ISO 14044 suggests three kinds of analysis for identifying significant issues. Contribution analysis splits the total result into processes, flows and modules and shows each one's share; in most product systems the top three to five processes carry the bulk of the total. Dominance analysis ranks the contributions, statistically or qualitatively. Anomaly assessment looks for unexpected results: a transport step overtaking production, a value with the wrong sign in one impact category, an unexplained gap between two similar alternatives. This third analysis is the step that catches the most modelling errors; a unit mistake (g instead of kg, kWh instead of MJ) usually shows up as an anomaly. Once identified, the significant issues are listed: ‘the result depends most on A3 electricity consumption, on the background data for the steel input and on the C3 recycling rate assumption’. That list defines what the next two checks will test.

The completeness check

The completeness check asks whether the necessary information and data are present and sufficient for each significant issue. In practice it is done as a table: for each significant process or flow, is data present, is its quality consistent with the goal, and if something is missing does the gap affect the result. A missing data item not on the significant issues list is noted and passed over; one on the list is either completed or the goal and scope are rewritten accordingly. Cut-off rules are re-evaluated here: a flow left out at the scoping stage under a one per cent threshold by mass or energy may turn out to be significant when viewed by impact (an additive in the toxicity category, for instance). The completeness check is the step that shows the cut-off decision also holds on an impact basis.

The sensitivity check

The sensitivity check asks to what extent data uncertainty, the allocation choice, the LCIA method and scenario assumptions affect the final result and the decision drawn from it. For each item on the significant issues list the parameter is changed by a set proportion (typically ten to twenty per cent) or to a plausible alternative value and the change in the result is recorded. For methodological choices an alternative is run: economic instead of mass allocation, system expansion instead of cut-off, EF v3.1 instead of CML. The output is a ranking: which assumption carries the result most. The ISO 14044 criterion is not whether the result itself changes but whether the decision drawn from it does; if the ranking of two alternatives reverses under a reasonable change of assumption, the study cannot make a reliable choice between them and must say so plainly.

The consistency check

The consistency check asks whether assumptions, methods and data are consistent with the goal and scope and between alternatives. In comparative studies it is the most critical check: if one alternative is modelled with site-specific primary data and the other with an industry average, the difference may come from the data rather than the products. The checklist is well established: data source and quality, geographical and temporal representativeness, technology level, system boundary, allocation rule, LCIA method and characterisation factor version, cut-off criteria. For each row the two alternatives are written side by side and any inconsistency is discussed for its effect. An inconsistency cannot always be removed; where no primary data exists for one alternative that is a limitation and must be named as such in the conclusions.

Reporting with uncertainty and the significant difference

Conclusions are reported not as a single number but with an uncertainty range. Monte Carlo simulation, using distributions derived from the pedigree matrix, produces P10, P50 and P90 values for each impact category; instead of ‘2.4 kg CO₂e per kg of product’ the report reads ‘median 2.4, 80 per cent interval 2.0 to 2.9 kg CO₂e’. In comparative studies the question of significant difference is answered with these ranges: if the distributions of the two alternatives overlap substantially, the difference disappears into the uncertainty of the data and the claim ‘A is lower-carbon than B’ cannot be made. That is why ISO 14044 makes uncertainty and sensitivity analysis mandatory in studies supporting comparative assertions. There is no fixed threshold, but a difference below ten per cent is rarely significant given typical LCA uncertainty and should be written up with corresponding caution.

Common interpretation errors

• Repeating the results: the interpretation section does not turn tables into prose, it questions them
• Interpreting on one impact category: the alternative that is smaller on GWP may be larger on water or toxicity
• Running sensitivity before identifying significant issues: a hundred parameters scanned, the three with leverage missed
• Reporting a difference that lies inside the uncertainty range as ‘better’
• Ignoring the uncertainty of background data: an ecoinvent activity is an estimate too
• Squeezing limitations into one sentence: every significant issue carries its own limitation
• Basing a comparative assertion on a weighted single score (ISO 14044 prohibits it)