Proficiency testing and interlaboratory comparisons

Uncertainty budgets are models; comparisons are the experiment that tests them. In an interlaboratory comparison a stable artefact — a reference PRT, a set of loggers — is circulated and each laboratory reports its value and uncertainty. The results are compared with a reference value, usually from a pilot laboratory or an NMI.

Interpreting E_n

• |E_n| ≤ 1: the difference from the reference is within the combined expanded uncertainties — consistent.
• |E_n| > 1: inconsistent; investigate the measurement or the uncertainty claim.
• |E_n| ≪ 1 for all points: possibly an over-estimated uncertainty.

For customers, asking whether a laboratory participates in comparisons for the quantity you need is a reasonable due-diligence question.

The E_n number, worked

E_n compares a laboratory's difference from the reference value against the two uncertainties combined. The formula is E_n = (x_lab − x_ref) / √(U_lab² + U_ref²), and the whole point is that it tests the uncertainty claim, not just the measurement.

The last two rows are the same measurement. The difference is the uncertainty claimed, and it illustrates what E_n actually tests: a laboratory can always pass by claiming a larger uncertainty, and a string of E_n values clustered near zero with a large U is not a good result — it is an over-conservative budget that will lose work on capability.

Reading the pattern, not the pass

An |E_n| ≤ 1 is satisfactory and an |E_n| > 1 requires investigation, but the useful signal is the sequence. Three consecutive results at +0.7, +0.8, +0.6 are all passes and all pointing the same way, which is a systematic effect the budget has not accounted for — a reference correction not applied, an immersion difference, an ambient assumption. Treating each round as an independent pass or fail throws that information away, which is why management review is asked to look at comparison results as an input rather than as a certificate.

For customers and for laboratories

• Uncertainty budgets are models, and a comparison is the experiment that tests the model. It is the only external check on whether a budget describes reality.
• Participation is expected under clause 7.7 as part of monitoring the validity of results, proportionate to the work — not for every scope line every year, but on a planned basis with a rationale.
• Trace results back to the individuals who did the work: a comparison outcome is also competence-monitoring evidence under clause 6.2.
• As a customer, asking whether a laboratory participates for the quantity you need is a fair due-diligence question, and asking how they responded to their last unsatisfactory result is a better one.
• An investigation into a poor E_n should end in either a corrected measurement or a corrected budget. Ending it by widening the uncertainty without a cause is how the fourth row above happens.