Interpretation, Uncertainty and Sensitivity Analysis
What interpretation must deliver
ISO 14044 §4.5 requires three steps: identify significant issues (contribution, dominance and influence analysis), evaluate completeness, consistency and sensitivity, then draw conclusions, limitations and recommendations. Contribution analysis ranks processes or flows by their share of an indicator; a 60% share for one process is a significant issue that must be discussed, not merely tabulated.
Methods
- Scenario analysis explores alternative model structures or futures - different grid mix, different end-of-life route.
- Sensitivity analysis perturbs one parameter or methodological choice and measures the resulting change.
- Perturbation analysis does this across all exchanges automatically to find influential data points.
- Uncertainty propagation (Monte Carlo, Latin hypercube) samples parameter distributions, usually lognormal with a geometric mean and GSD, producing a distribution of results.
- Pedigree/DQI scoring gives a semi-quantitative data-quality statement when full propagation is impractical.
Worked example
A cradle-to-gate result of 100 kg CO2e has a 90% interval of 80–125 kg CO2e from 10,000 Monte Carlo runs. Electricity contributes 60%. Switching to a marginal grid dataset shifts the result by +12%, which is smaller than the gap between the two options compared. Conclusion: the ranking is robust; the absolute value is not suitable for an offset claim, and the report says so.
Traps
Reporting a mean with no interval; conflating parameter uncertainty with natural variability across a product fleet; presenting a sensitivity as though it were a scenario; assuming normality for lognormal data; omitting run count and random seed; using an uncertainty range to imply precision when model-structure choices dominate.