Uncertainty Evaluation in Quantitative Spectral Analysis: Principles, Methods, and Applications

Measurement uncertainty evaluation provides a quantified, statistically defensible confidence interval around every concentration result produced by quantitative spectral analysis techniques such as ICP-OES, ICP-MS, AAS, XRF, and spark OES. Built on the ISO/IEC Guide 98-3 (GUM) framework and the accreditation requirements of ISO/IEC 17025, a rigorous uncertainty budget enables accredited third-party laboratories to demonstrate metrological traceability, support conformity decisions, and continuously control data quality. This article explains the key uncertainty sources, calculation methods, verification strategies, and practical applications of uncertainty evaluation in spectroscopic quantitative analysis.

Technical Overview

Measurement uncertainty is a non-negative parameter characterizing the dispersion of the quantity values attributed to a measurand, based on the information used. In quantitative spectral analysis, the full measurement chain typically includes sampling, sample preparation (digestion, fusion, or dilution), preparation of calibration standards, instrumental measurement, and mathematical data reduction. Each of these steps introduces variability, and the evaluated uncertainty expresses the doubt that remains about the validity of the final reported concentration.

Uncertainty must be clearly distinguished from error: an error is the deviation of a single result from a true value, whereas uncertainty describes a range of values that could reasonably be attributed to the measurand. The internationally recognized methodology is defined in the Guide to the Expression of Uncertainty in Measurement (GUM), published as ISO/IEC Guide 98-3, and further elaborated for analytical chemistry in the Eurachem/CITAC guide on quantifying uncertainty in analytical measurement and in national documents such as GB/T 27418. The core workflow consists of four steps:

• Specify the measurand and define the mathematical model relating input quantities to the analytical result.
• Identify all significant uncertainty sources and quantify each component using Type A (statistical) or Type B (other means) evaluation.
• Combine the components through the law of propagation of uncertainty to obtain the combined standard uncertainty.
• Multiply by a coverage factor, typically k = 2, to report the expanded uncertainty corresponding to approximately 95% coverage probability.

Laboratories accredited under ISO/IEC 17025 are required to identify the contributions to measurement uncertainty in their quantitative spectral methods and to apply the evaluation when reporting results, particularly where conformity with specifications or regulatory limits must be decided. Traceability of calibration standards to certified reference materials certified in accordance with ISO Guide 33 and ISO Guide 35 forms the foundation of a defensible uncertainty budget.

Testing Samples

Metallic Materials and Alloys

Plain carbon steels, low-alloy and stainless steels, aluminum, copper, zinc, magnesium, and titanium alloys, nickel-base superalloys, and foundry products are routinely analyzed by spark discharge optical emission spectrometry, X-ray fluorescence spectrometry, and glow discharge OES. Uncertainty evaluation for these samples must account for reference material matching, metallurgical structure, surface preparation, and calibration curve quality.

Environmental and Geological Samples

Surface water, groundwater, wastewater, soil, sediment, fly ash, and ores are digested by acid or alkaline fusion and measured by ICP-OES, ICP-MS, or AAS for major, minor, and trace elements. Uncertainty components from digestion recovery, dilution volumes, and matrix effects are frequently significant for these matrices.

Food, Agricultural, and Pharmaceutical Matrices

Plant tissues, animal feed, beverages, dietary supplements, and drug products are analyzed for regulated elements such as lead, cadmium, arsenic, and mercury, typically after microwave digestion. Because regulatory limits are often close to detection limits, uncertainty evaluation is essential for defensible pass or fail decisions.

Petrochemicals, Chemicals, and Polymers

Fuels, lubricants, process catalysts, catalyst residues in polymers, and inorganic chemicals are analyzed by XRF, ICP-OES, and ICP-MS after appropriate dilution, ashing, or digestion. Viscous and volatile matrices add preparation-related uncertainty components that must be included in the budget.

Testing Items

Uncertainty evaluation in spectral quantitative analysis is typically delivered as a documented uncertainty budget accompanying elemental or molecular concentration results. Typical deliverables include:

• Combined and expanded uncertainty for each reported element concentration, expressed in concentration units and as a relative percentage.
• A component-by-component uncertainty budget identifying the relative contribution of each source.
• Calibration function uncertainty derived from least-squares regression residuals, slope, and intercept.
• Contributions from certified reference material certificates, standard stock solutions, and gravimetric or volumetric dilution steps.
• Repeatability and intermediate precision components from replicate and inter-run measurements.
• Sample preparation contributions, including sample mass, final volume, dilution factor, and digestion or recovery correction.
• Drift correction, matrix effect, spectral interference, and background correction contributions where applicable.
• Expanded uncertainty U calculated with a coverage factor k = 2 for approximately 95% coverage probability, together with conformity statements and guard bands where decision rules apply.

Testing Methods

Defining the Measurand and the Measurement Model

The evaluation begins by writing the functional relationship between the analytical result and its input quantities. For a typical digestion-based method, the mass fraction of an element equals the concentration measured in the digest multiplied by the solution volume and dilution factor, divided by the sample mass, further multiplied by recovery or interference correction factors. Every quantity appearing in this model becomes a candidate uncertainty source.

Type A Evaluation by Statistical Analysis

Replicate measurements of a homogeneous test sample under repeatability conditions provide an experimental standard deviation; the uncertainty of the mean is obtained by dividing this standard deviation by the square root of the number of replicates. Intermediate precision data collected over different days, analysts, and instrument calibrations, often analyzed by analysis of variance, capture longer-term variability that should not be omitted from the budget.

Type B Evaluation from Prior Knowledge

Components that cannot be observed statistically are evaluated from existing information: the expanded uncertainty and coverage factor stated on certified reference material certificates, the calibration uncertainty of the analytical balance, the manufacturer tolerances of volumetric flasks and pipettes, reagent purity specifications, and data processing resolution. Quantities bounded by limits with unknown internal distribution, such as a flask tolerance or a recovery window, are modeled as rectangular distributions and divided by the square root of three to obtain standard uncertainties.

Uncertainty of the Calibration Function

For the linear regression of instrument signal against standard concentration, the uncertainty of a predicted concentration is computed from the residual standard deviation of the regression, the slope, the number of calibration points, the number of replicate readings of the unknown, and the distance of the predicted concentration from the mean of the calibration standards. Residual plots are inspected to confirm linearity; where heteroscedasticity is evident, particularly near detection limits, weighted regression is applied. Covariance between the slope and intercept is considered for small calibration data sets.

Combining Components and Calculating Expanded Uncertainty

Relative standard uncertainties of independent components are summed in quadrature to obtain the combined relative uncertainty, which is multiplied by the result to give the combined standard uncertainty. The expanded uncertainty is then calculated as U = k × u, with k = 2 providing a level of confidence of approximately 95% for a normal distribution. A typical reporting format is: lead content 0.512 mg/kg ± 0.036 mg/kg, where the stated uncertainty is the expanded uncertainty using a coverage factor of k = 2.

Alternative and Complementary Approaches

ISO 21748 allows uncertainty to be estimated from collaborative study data by combining the repeatability, between-laboratory reproducibility, and trueness estimates obtained in accordance with ISO 5725. For strongly non-linear models, dominant non-rectangular input distributions, or non-normal output distributions, the Monte Carlo method described in ISO/IEC Guide 98-3 Supplement 1 provides a numerical propagation alternative to the classical GUM law.

Verification, Quality Control, and Documentation

The uncertainty budget is validated by analyzing certified reference materials and confirming that the absolute difference between the measured and certified values does not exceed twice the root sum of squares of the laboratory and certified uncertainties. Control charts of quality control samples, participation in proficiency testing schemes, and zeta score evaluation confirm the continuing validity of the budget, which is reviewed after any change of instrument, calibration procedure, or preparation scheme.

Applications

Uncertainty evaluation is not a purely academic exercise; it directly supports regulatory, commercial, and quality decisions in every sector that relies on spectral quantitative data:

• Conformity assessment against specification or regulatory limits using documented decision rules and guard bands, as required by ISO/IEC 17025 when statements of conformity are issued.
• Accreditation assessments, where a complete and defensible uncertainty budget demonstrates technical competence of the laboratory.
• Method validation and method transfer, where the evaluated uncertainty is compared with the target, or fitness-for-purpose, uncertainty defined by the customer or regulator.
• Manufacturing quality control, for example release of steel heats by spark OES or verification of restricted substances in consumer products by XRF and ICP-MS.
• Environmental permitting and discharge compliance reporting, where concentrations near statutory limits must be interpreted with a known confidence interval.
• Interlaboratory comparisons and research studies, where uncertainty enables statistically sound comparison of results from different laboratories and techniques.

FAQ

What is the difference between measurement error and measurement uncertainty?

Error is the difference between an individual measured value and the true value, and it is a single signed quantity that can, in principle, be corrected. Uncertainty is a positive parameter describing the dispersion of all values that could reasonably be attributed to the measurand; it quantifies the reliability of the result rather than correcting it.

Why is a coverage factor of k = 2 so widely used?

For an approximately normal distribution of output values, multiplying the combined standard uncertainty by k = 2 yields an interval with a coverage probability of approximately 95%, which is the conventional confidence level expected in commercial and regulatory reporting. A k factor close to 2.0 is also obtained from the t-distribution for the effective degrees of freedom typical of most spectral analysis budgets.

Which uncertainty components usually dominate in spectral quantitative analysis?

In most budgets, the calibration regression component and the analytical repeatability dominate, followed by the certified uncertainty of the calibration standards. At trace levels, contributions from blank variability, background correction, and spectral interferences grow rapidly and may become controlling.

How should a result with uncertainty be reported?

The result is reported together with the expanded uncertainty and the coverage factor, for example 12.34 ± 0.28 mg/kg, with a statement that the reported uncertainty is the expanded uncertainty using a coverage factor of 2, corresponding to a coverage probability of approximately 95%.

How often should an uncertainty budget be reviewed?

The budget should be reviewed whenever the instrument, calibration procedure, preparation scheme, or reference materials change, and at defined intervals as part of the quality management system. Continuous verification through control charts and periodic certified reference material checks confirms that the documented budget remains representative of routine performance.

What is the relationship between uncertainty and the limit of detection?

The limit of detection describes the smallest concentration distinguishable from the blank with a stated statistical confidence, while uncertainty describes the dispersion of the measured value itself. Near the detection limit the relative uncertainty becomes very large, which is why measurement capability at trace levels is judged jointly by detection capability and uncertainty performance.