Sampling Is Part of the Measurement

Overview

Yes — sampling uncertainty can matter as much as the analytical instrument. Measurement results are only as reliable as the whole process that produced them. When you read or use a reported concentration or content value, treat the sample itself as one more part of the measurement system: how it was selected, how variable it was to begin with, how it was subdivided, how it moved, and how it was prepared for analysis can change the result as much as the balance or instrument used. This article unpacks those five linked steps in plain language and gives a practical way to judge evidence about measurements. Keep in mind: conclusions are sample-, method-, matrix-, and date-specific; no single result proves sterility, safety, efficacy, Canadian authorization, chain of custody, or batch-wide uniformity.

Why the model matters The EURACHEM model treats sampling and analysis as a chain of processes that together determine uncertainty. If any link is weak, uncertainty increases. Published analysis precision down to many decimal places only addresses the analytical step unless the report explicitly quantifies sampling uncertainty too . Peer-reviewed studies show sampling often dominates total uncertainty, especially for heterogeneous materials or small true differences between items . You cannot assume the measurement equals the sample’s true value without evidence about how that sample was obtained and handled.

Five linked steps that form one evidence chain

1) Selection: who or what was chosen, and why it matters Selection is the rule that decides which physical items become "the sample." Random, stratified, or convenience choices produce different kinds of evidence. If selection is non-random or not documented, reported results may not represent the population of interest. EURACHEM emphasizes that selection bias is a major source of uncertainty when the selection method is poorly specified or unsuitable for the product’s variability . Look for explicit statements about sampling plans, inclusion/exclusion rules, and whether the plan was followed.

2) Heterogeneity: inherent variation inside and between items Many materials are heterogeneous at scales relevant to the measurement: ingredients, particle size, moisture pockets, coatings, or distribution of active components. Heterogeneity increases the variability between different samples and within subsamples taken from the same unit. Laboratory precision cannot compensate for a sample that, by chance, over- or under-represents high- or low-concentration regions. Studies in analytical chemistry report that when heterogeneity is large relative to analytical precision, it becomes the dominant uncertainty term . Good reports quantify heterogeneity or at least describe expected variability and how the sampling plan addresses it.

3) Subdivision: reducing a large sample to the portion analysed After selection, most measurements require subdividing a bulk sample into aliquots for analysis. Subdivision can introduce or amplify bias: failing to adequately mix, to grind or homogenize, or to use proper coning-and-quartering techniques causes subsamples to be non-representative. EURACHEM provides methods to estimate uncertainty arising from splitting and recommends procedures to reduce it . When reading results, check whether the report explains how the sample was reduced and whether methods for homogenization or representative splitting were applied and validated.

4) Transport and storage: change between collection and analysis Time, temperature, light, and handling can alter what is being measured. Some analytes degrade, sorb to containers, or migrate within a sample matrix. Transport without suitable controls can therefore change the value measured in the lab relative to the in situ condition. A trustworthy report documents chain-of-custody, conditions of transport and storage, preservatives or containers used, and any tests for stability. Without that information, sampling uncertainty increases because change during transit becomes an extra unknown.

5) Analysis: the instrument is necessary but not sufficient Analytical instruments and methods have well-defined precision and bias characteristics, usually estimated by validation, calibration, and quality control. However, those numbers only address the analytical step, not upstream sampling uncertainty. EURACHEM’s framework insists measurements must combine analytical uncertainty with sampling uncertainty to reflect total uncertainty of the final reported value . Peer-reviewed work confirms that ignoring sampling error leads to overconfident conclusions, especially when sample heterogeneity or poor sampling practice is present .

A practical evidence-reading approach

What remains unresolved without data Even with careful reading, some questions can only be resolved by additional information: the exact degree of heterogeneity across the whole lot, whether selection rules were applied every time, whether degradation occurred during transport for specific analytes, and whether subsampling introduced systematic bias. Those are sample-, method-, matrix-, and date-specific issues; a single lab value cannot answer them for other samples or lots. Reports that omit sampling uncertainty leave readers unable to know how much of the total uncertainty is due to the sample chain rather than the instrument.

Bottom line Treat sampling as part of the measurement. Ask for explicit sampling plans, evidence about heterogeneity, clear subdivision protocols, transport and storage documentation, and a combined uncertainty statement. When those elements are missing or under-documented, sampling uncertainty can be as large or larger than analytical uncertainty, and reported results should be interpreted with caution . Remember the permitted boundaries: any conclusion you draw from a single result depends on the sample, method, matrix, and date; no isolated measurement proves sterility, safety, efficacy, authorization, chain-of-custody integrity, or lot-wide uniformity .

• Ask about the sampling plan first. Is the sampling design described (random, stratified, targeted)? Does the description match the question the data are claimed to answer? If not stated, the result’s applicability is limited .
• Look for heterogeneity information. Does the report quantify within-unit or between-unit variability, or at least describe expected heterogeneity in the matrix? High heterogeneity relative to reported analytical precision signals likely large sampling uncertainty .
• Check subdivision methods. Are homogenization, splitting, or subsampling procedures described and appropriate for the matrix? Were procedures validated or repeated to assess repeatability?
• Inspect transport and storage documentation. Were conditions controlled or tested for stability? Is chain-of-custody recorded? Missing information here enlarges the uncertainty envelope.
• Confirm total uncertainty reporting. Does the report combine sampling and analytical uncertainties into a single estimate or state them separately? EURACHEM recommends combined estimates; absence of sampling uncertainty means stated confidence may be misleading .