On the transferability of the GUM-S1 type A uncertainty

A key requirement for the evaluation of measurement uncertainty in metrology is its transferability. This is particularly relevant for ensuring traceability through a chain of measurements. The GUM [JCGM 100] requires that ' it should be possible to use directly the uncertainty evaluated for one result as a component in evaluating the uncertainty of another measurement in which the first result is used'. The GUM implements the transfer of uncertainty by applying variance propagation to linear models or models that can be well approximated through a linearization. GUM-S1 [JCGM 101] has been released to broaden the scope of uncertainty evaluation to cover also non-linear models. We demonstrate in terms of examples that the GUM-S1 type A evaluation does not satisfy the requirement of transferability, i.e. re-using the probability distribution produced by GUM-S1 in a subsequent uncertainty exercise can lead to inadequate results for non-linear models. Furthermore, already for linear models the type A evaluation of GUM-S1 is shown to produce unsatisfactory solutions. These findings are discussed and the underlying reason is identified, namely that the use of non-informative priors as in GUM-S1 does not lead to results that can always be reliably transferred. We discuss possible alternatives and finally draw some conclusions.