What is the Uncertainty Factor?

Significance of the topic

Measurement uncertainty (MU) is essential for interpreting analytical results and for making decisions such as regulatory compliance. Standard symmetric uncertainty intervals (x ± U) are adequate when measurement variability is low and data follow a normal distribution. However, when relative uncertainty is large or when replicate measurements show positive skew (log-normal behaviour), a symmetric interval can misrepresent the true confidence limits. The uncertainty factor (FU) provides a convenient, realistic, and asymmetric uncertainty interval in these situations.

Objectives and overview of the study

The leaflet explains the concept of the uncertainty factor and demonstrates how FU can be calculated and communicated. It contrasts the traditional expanded uncertainty (U or U') with FU and presents case studies where FU is more appropriate, especially when sampling heterogeneity or analytical processes produce log-normal result distributions.

Methodology and analytical approach

The key methodological points are:

• Traditional expanded relative uncertainty U' is derived from the standard deviation of replicate measurements (s_meas) and a coverage factor k (commonly 2 for ~95 % confidence), and assumes an approximately normal distribution.
• If replicate results are positively skewed, a log-transformation (natural logarithm) of results can reveal an approximately normal distribution on the log scale, indicating multiplicative (log-normal) dispersion.
• For log-normal data, FU is derived from the standard deviation of the log-transformed replicates (s_L,meas). Practically, FU can be computed as the exponential of k times this log-scale standard deviation (i.e., FU ≈ exp(k · s_L,meas) for a chosen k, typically 2), yielding an asymmetric interval [x/FU, x·FU].
• The duplicate method (sampling duplicates and repeated analyses) is an effective empirical approach to capture repeatability and sampling-induced heterogeneity; analysis of variance (ANOVA) on the log-transformed results can be used to estimate s_L,meas when data structure warrants it.

Used instrumentation

The case study and examples in the leaflet rely on common sampling and laboratory techniques and software:

• Acid digestion of soil samples prior to analysis.
• Inductively Coupled Plasma - Atomic Emission Spectroscopy (ICP-AES) for Pb determination.
• Use of Certified Reference Materials (CRMs) to check analytical bias.
• Statistical tools such as ANOVA software (example: RANOVA3) to process replicate or duplicate measurements and to estimate log-scale variance components.

Main results and discussion

Highlights from the examples discussed in the leaflet:

• Soil lead example: 100 sampling targets were collected; 10 targets yielded duplicates that were each analysed twice (40 results). Replicate results were positively skewed on the original scale but near-normal after natural-log transformation, indicating log-normal variability driven largely by sampling heterogeneity.
• Computed FU for the duplicated soil Pb dataset was 2.62. For a reported concentration of 300 mg kg-1 this corresponds to an approximate 95 % confidence interval from 300/2.62 ≈ 115 mg kg-1 up to 300×2.62 ≈ 786 mg kg-1, an asymmetric and wide interval reflecting true uncertainty dominated by sampling heterogeneity.
• Analytical examples can also show log-normal behaviour: proficiency-test results for GMO content in soy exhibited a log-normal distribution, where FU would likewise be the appropriate MU expression.
• FU better represents multiplicative uncertainty and avoids implausible negative lower bounds that can arise when using symmetric intervals for highly variable or skewed data.

Benefits and practical applications of the uncertainty factor

The uncertainty factor offers several practical advantages:

• Provides asymmetric confidence intervals that align with multiplicative variability commonly encountered in environmental sampling and some analytical measurements.
• Maintains interpretability where conventional ±U would either be misleading or yield impractical negative lower limits.
• Is straightforward to compute from log-transformed replicate data and can be implemented with standard statistical software or ANOVA routines.
• Helps convey realistic ranges for decision-making in regulatory, environmental, and quality-assurance contexts where sample heterogeneity or skewed analytical responses dominate uncertainty.

Future trends and potential applications

Prospective directions and uses include:

• Wider adoption of FU reporting in areas with strong multiplicative variability: environmental site characterisation, contaminant hot-spot assessment, biological assays, and some molecular or PCR-based analyses.
• Integration of FU calculation into laboratory information systems and proficiency testing reporting, so that users routinely receive asymmetric uncertainty intervals when appropriate.
• Refinement of sampling design and duplicate/replicate strategies to better characterise log-scale variance components, improving the robustness of FU estimates.
• Development of guidance and training materials to improve communication of FU-based intervals to non-specialist stakeholders and regulators.

Conclusion

The uncertainty factor is a practical, scientifically justified way to express measurement uncertainty when data are log-normally distributed or when relative uncertainty is large. FU yields asymmetric confidence intervals that reflect multiplicative sources of variability—most notably sampling heterogeneity—and can be derived from log-transformed replicate data or via ANOVA on duplicate sampling schemes. Clear communication of FU-based results is important to ensure users correctly interpret asymmetric uncertainty ranges for decision-making.

References

1. Ramsey M. H., Ellison S. L. R. and Rostron P., editors. Measurement uncertainty arising from sampling: a guide to methods and approaches. Second Edition. Example A2, pages 44–52. Eurachem, 2019. ISBN 978 0 948926-35-8.
2. RANOVA3 software. Royal Society of Chemistry Analytical Methods Committee (example ANOVA tool used to estimate log-scale variance components).
3. AMC. GMO Proficiency testing: Interpreting z-scores derived from log-transformed data. Technical Brief No 18. Analytical Methods Committee, 2004.