Is correct sampling identical to taking a large sample?

No; size does not remove a systematic selection preference.

Can a correct sample differ from the lot average?

Yes, because finite-sample variability remains.

Does the term apply only to a laboratory instrument?

No; it concerns selection and preservation across the sampling process, including stages before instrumental analysis.

AI Glossary / Statistics and probability / Advanced

Correct sampling

Definition

Correct sampling is a technical principle in Pierre Gy's theory of sampling for particulate materials: every constituent particle of the lot should have an equal probability of being selected and retained in the sample. It concerns the physical selection and handling process, including whether the sampling device can include all relevant particles. Correctness does not mean that every individual sample exactly reproduces the lot's composition.

What it means

A scoop taken from the top of a container may favor particles that settle, rise, or remain easy to reach. A narrow opening may exclude larger particles. Even if the operator chooses a location casually, the physical mechanism can systematically favor some material.
Gy's principle draws attention to that mechanism. The term is narrower than the everyday phrase sampling correctly and is not a universal rule that every valid statistical survey must use equal inclusion probabilities. Survey designs can deliberately use unequal probabilities with appropriate estimation; that is a different framework from this particulate-sampling definition.

How it works

Define the lot to be represented and the property to be measured. Examine particle sizes, composition, spatial distribution, segregation, and other features that can affect selection. Specify how primary increments, combined samples, and analytical subsamples will be obtained.
Design the increment boundaries and extraction process so constituent particles receive equal opportunities for inclusion. A geometrically suitable cut can still fail if the device loses part of the selected material or excludes certain particles. Delimitation and extraction are therefore separate practical concerns.
Preserve the sample's relevant composition through transfer, storage, preparation, and subsampling. Losses, contamination, or alteration can undermine a sound initial selection. A small analytical portion must represent the material delivered to the laboratory; precision in the final instrument does not repair an unrepresentative subsample.
Correct sampling still leaves variability associated with the material's heterogeneity and the finite amount collected. In Gy's framework, fundamental sampling error remains even under ideal selection conditions. The sample mass, particle characteristics, and sampling design affect the achievable precision. Multiple increments can help address distributional heterogeneity, but their number alone does not guarantee correctness.
Assess equipment and procedures, document departures, and use suitable replicate sampling or other quality studies to understand uncertainty. The required approach depends on the material and objective. A general glossary definition cannot replace a validated sampling protocol for a particular substance or process.

Three practical examples

A mineral-processing team assesses whether a sampler cuts a full material stream without excluding larger particles.
A laboratory reviews how it obtains a small analytical portion from a heterogeneous particulate sample.
An AI quality project investigates whether its reference chemical measurements came from representative physical samples before treating those measurements as reliable training labels.

Think of it this way

Think of a lottery in which every particle needs an equal chance of a ticket and must remain in the winning group after selection. A device that cannot admit large particles is like a lottery that refuses some entrants. Careful measurement of the winners cannot repair that exclusion.

Worked example

Consider a simplified lot containing one thousand equal-mass particles. Eight hundred have analyte mass fraction 1%, and two hundred have analyte mass fraction 9%. The lot's mass fraction is (800 x 0.01 + 200 x 0.09)/1,000 = 0.026, or 2.6%.
Suppose a biased device selects exactly eighty particles from the first type and four from the second, uniformly within each type. Individual inclusion probabilities are 80/800 = 0.10 and 4/200 = 0.02. They are unequal, so the mechanism fails the stated correct-sampling principle.
The resulting sample fraction is (80 x 0.01 + 4 x 0.09)/84, approximately 0.01381, or 1.381%. It underrepresents the analyte-rich particles despite collecting eighty-four observations.
By comparison, a simple random sample of one hundred particles gives every particle inclusion probability 0.10. Its expected composition is eighty particles of the first type and twenty of the second, and its mean analyte fraction is unbiased for 2.6% in this equal-mass toy setting. An individual sample can still differ from that expected composition.
Equal particle masses make the arithmetic transparent. Real materials can have different masses, shapes, concentrations, and spatial structure, so physical sampling and mass-based estimation require additional care.

Illustrative case study

Fictional case: Granite Materials sees inconsistent laboratory results for a powder blend. The team initially suspects the analytical instrument, but an investigation finds that the subsampling scoop favors material near the surface after transport.
Analysts review the material's segregation behavior and the extraction method with sampling specialists. They test a revised procedure, track sample handling, and compare replicate subsamples. Instrument calibration remains necessary, but it addresses a different source of error. The team documents uncertainty in the physical sampling process before using the measurements to train an automated quality-prediction model.

Common misconceptions and mistakes

Correct sampling does not eliminate all sampling error, ensure perfect representativeness in every draw, or make any sample mass adequate. A random-looking operator action is not proof of equal particle inclusion. Mixing does not automatically solve every extraction problem, and repeated analysis of the same biased sample only studies analytical repeatability. Do not generalize the equal-probability definition to declare all unequal-probability survey designs invalid.

Business use

For businesses working with heterogeneous materials, sampling quality can determine whether expensive analysis supports a sound decision. Audit the full path from lot definition to the final analytical portion. When physical measurements become AI labels, preserve sampling metadata and uncertainty so a model is not evaluated against misleading reference values. Investing in analytical precision alone may leave the dominant sampling problem untouched.

Check your understanding

Which type is underrepresented by the toy device?

Answer: The particles with 9% analyte content.

Why does the mechanism fail correctness?

Answer: The two types have different individual inclusion probabilities.

Would measuring the eighty-four selected particles more precisely fix the selection problem?

Answer: No.

Related terms

Gy's sampling theory
particulate sampling
fundamental sampling error
increment delimitation
increment extraction
segregation
sample preparation
representativeness

Learn next

Trace every selection and reduction step in a material-sampling workflow. Then study how particle size, heterogeneity, sample mass, and sampler geometry influence uncertainty and the feasibility of correct extraction.

Sources

Robert W. Gerlach and John M. Nocerino, U.S. EPA, Guidance for Obtaining Representative Laboratory Analytical Subsamples from Particulate Laboratory Samples, EPA/600/R-03/027: https://nepis.epa.gov/Exe/ZyPURL.cgi?Dockey=2000GTWM.TXT Petersen, Minkkinen and Esbensen, Grab vs. composite sampling of particulate materials with significant spatial heterogeneity, original research abstract: https://pubmed.ncbi.nlm.nih.gov/19800475/

Previous: Corpus Next: Correlated sampling