Sampling: Why a Few Vials Stand for a Whole Batch

No one tests an entire batch, because testing consumes what it measures. Every certificate therefore rests on an inference from a sample to a population, and the design of that sample is usually a larger source of uncertainty than the analysis performed on it.

The inference

Testing a sample and reporting the result for a batch is valid to the extent the sample is representative. If the batch is genuinely homogeneous, one vial is as good as another and a single sample suffices. If it is not, the sample describes itself.

Homogeneity is therefore the assumption doing the work, and it is established by how the material was produced rather than by the test. What a batch is covers where homogeneity holds and where it becomes an approximation.

What different sampling schemes protect against

• Random sampling gives every unit an equal chance of selection. It protects against systematic bias in which units get chosen and is the default where no particular failure mode is suspected.
• Stratified sampling deliberately draws from defined sub-populations — beginning, middle and end of a filling run, or one vial per lyophilisation cycle. It protects against a specific concern: that a property varies across the run in a way random selection might miss.
• Convenience sampling takes whatever is nearest. It is the weakest design and, in practice, the most common outside regulated settings.

The square-root-of-n-plus-one rule is a familiar heuristic for how many containers to sample. It is a convention rather than a statistical derivation, and it is widely used because it scales sensibly rather than because it guarantees anything.

Why sampling dominates the uncertainty

An HPLC purity determination on a well-controlled method carries an uncertainty of a few tenths of a percent. Measurement uncertainty works through where that comes from.

Between-vial variation in a batch that is not perfectly homogeneous can exceed that comfortably. Refining the analytical method while sampling one convenient vial improves the precision of a measurement whose representativeness is the weaker link — a familiar pattern wherever measurement is easier to improve than study design.

Who draws the sample

The sampling party is part of what a report establishes. A laboratory reports on the material it received; who selected that material, and how, sits outside the analysis and is rarely recorded on a customer-facing document. Third-party versus in-house testing covers what changes between the arrangements.

This is not a hidden problem so much as an unstated boundary. A certificate describes a sample accurately and says, by convention, that the sample stood for a batch. The strength of that convention is the strength of the sampling behind it.

All material is supplied for laboratory research use only. It is not a drug, not a supplement, and not for use in humans or animals.