Measurement Uncertainty Review: Build an Error-Source Inventory

Overview

Short answer: look for and document separate contributors across sampling, calibration, repeatability, interferences, equipment performance, reference materials/standards, method validation, and random variability — then assess how the study quantified or bounded each. Below I explain the typical error sources, how analysts try to measure them, and practical questions to ask when reading a paper or report so you can judge how complete the uncertainty inventory is for an analytical result. This is research-source literacy, not medical advice.

1. Sampling and sample handling

Sampling is often the largest single contributor to uncertainty because the sample sent to the lab may not represent the population of interest. EURACHEM’s guide focuses on the many ways sampling generates variability and bias: sample heterogeneity, subsampling, sample mass, and handling (transport, storage, preservation) each introduce uncertainty and can be quantified or bounded with designed experiments or revealed by measurement of field replicates .

Practical evidence-reading questions

What the literature shows

• Did investigators describe the sampling plan (random, composite, stratified) and the physical heterogeneity of the matrix?
• Were field or batch replicates taken to estimate sampling variability, and did the authors report variance components from those replicates ?
• Is there a documented chain-of-custody, preservation method, and time window between collection and analysis, and was any stability testing reported?
• EURACHEM provides structured approaches to partition sampling uncertainty from analytical uncertainty and suggests using replicate sampling and nested designs to estimate variance components . Sampling uncertainty is distinct from instrument repeatability and must be quantified separately.

2. Calibration and standards

Calibration translates instrument response into quantities. Uncertainty arises from the calibration model (fit errors, range, nonlinearity), the uncertainty of standard values, and standard preparation (weighing, dilution). Peer-reviewed discussions stress that calibration uncertainty can be a dominant term unless handled explicitly .

Practical evidence-reading questions

What the literature shows

• What calibration materials were used (primary standards, certified reference materials, in-house standards)? Are their stated uncertainties provided?
• Was the calibration model described (linear, weighted, non-linear)? Are residuals, R², or goodness-of-fit statistics reported?
• Were multiple calibration levels used to cover the measurement range, and were calibration checks or bracketing standards run during the sample batch?
• Calibration should be coupled with an uncertainty budget that includes standard value uncertainty, volume/weighting errors in standard preparation, and model-fitting residuals; peer-reviewed guidance documents recommend explicit propagation of these terms into the final uncertainty .

3. Repeatability and intermediate precision

Repeatability (same conditions, short time) and intermediate precision (different days, operators, instruments) capture within-lab variability. Papers that separate these components allow better attribution of random variation in results [1,3].

Practical evidence-reading questions

What the literature shows

• Are repeatability and intermediate precision reported as standard deviations or coefficients of variation? Over how many replicates/days?
• Were different operators, instruments, or reagent lots intentionally used to estimate intermediate precision?
• Did the authors use ANOVA or variance-component analysis to allocate variability to repeatability versus other sources?
• Method validation studies often report repeatability and intermediate precision separately; variance-component analysis is the standard approach to partition these terms and feed them into an overall uncertainty budget [1,3].

4. Interferences and matrix effects

Matrix interferences alter signal independent of analyte concentration: suppression/enhancement in spectrometry, co-elution in chromatography, or chemical reactions during sample prep. Studies can evaluate these by spike-recovery, standard-addition, matrix-matched calibration, or interference panels [1,3].

Practical evidence-reading questions

What the literature shows

• Did the study assess matrix effects (e.g., recovery tests, standard-addition experiments, or matrix-matched calibration)?
• Were possible chemical or spectral interferences identified and tested, and were limits of quantification adjusted accordingly?
• If internal standards were used, is their behavior in different matrices documented?
• Assessing and reporting matrix effects is essential; failure to do so leaves an important unresolved source of bias and uncertainty [1,3].

5. Equipment, consumables, and maintenance

Instrument performance (stability, drift, resolution), consumables (columns, vials, filters), and maintenance affect both random and systematic errors. Performance checks and quality control charts can document these contributors over time .

Practical evidence-reading questions

What the literature shows

• Are instrument performance checks reported (blanks, system suitability, drift tests) and over what frequency?
• Did the authors report maintenance/consumable changes and any associated shifts in results?
• Were quality control charts or control sample results presented to show ongoing performance?
• Routine performance verification and documentation enable estimation of long-term uncertainty and detection of systematic shifts .

6. Reference materials and traceability

Certified reference materials (CRMs) provide anchors for traceability. Their stated uncertainties, matrix compatibility, and commutability to the sample matrix should be documented. If CRMs are absent, authors must justify alternative approaches .

Practical evidence-reading questions

What the literature shows

• Were CRMs or certified standards used, and are their certificate uncertainties quoted?
• If no CRM was available, how did the authors establish traceability or validate accuracy?
• Are limitations of CRM commutability to the sample matrix discussed?
• Use of CRMs with documented uncertainty strengthens an uncertainty assessment; lack of appropriate CRMs requires transparent justification and introduces unresolved limitations .

7. Method validation and bias assessment

Method validation examines trueness (bias), precision, selectivity, limits of detection/quantification, and range. Validation studies should connect these elements to the uncertainty budget using documented experiments rather than assumptions [1,3].

Practical evidence-reading questions

What the literature shows

• Was validation performed for the specific matrix and concentration range? Are bias and precision data provided?
• Were limits of detection/quantification and their uncertainty determination methods described?
• How were outliers handled and how might that affect uncertainty estimates?
• Proper validation quantifies bias and precision components; these are necessary inputs to an honest uncertainty budget [1,3].

8. Random variability and uncertainty propagation

Random variability is captured statistically and propagated through the measurement model to produce an overall uncertainty. Authors should state whether they used frequentist propagation (variance arithmetic, law of propagation of uncertainty) or Monte Carlo/simulation approaches, and list assumptions .

Practical evidence-reading questions

What the literature shows

Summary checklist for readers

A careful paper will not simply state a single uncertainty number without documenting these contributors. Where the evidence is incomplete, the unresolved items should be explicitly listed so users can judge risk and decide whether additional experiments (field replicates, matrix-matched recovery, CRM checks) are needed before relying on the reported values .

• Did the paper present an uncertainty budget or at least the dominant quantified terms and their propagation method?
• Were correlation among input quantities considered (non-independence can change propagated uncertainty)?
• If Monte Carlo methods were used, are the input distributions and number of iterations reported?
• Both analytical propagation and Monte Carlo approaches are accepted; transparency about the chosen method and assumptions is crucial to evaluate completeness .
• Is sampling variability quantified and separated from analytical variability ?
• Are calibration standards, their uncertainties, and model-fit residuals reported ?
• Are repeatability and intermediate precision estimated and partitioned [1,3]?
• Have matrix effects and interferences been tested and documented [1,3]?
• Are instrument performance checks and quality-control records presented ?
• Are CRMs used or traceability justified, and are their uncertainties given ?
• Does a validation study provide bias and limit data relevant to the matrix and range [1,3]?
• Is the method for uncertainty propagation reported, and are assumptions/correlations addressed ?