Can coverage factor 2 be interpreted as an equivalent to 95% coverage level in uncertainty estimation? Two case studies

The GUM modelling, its Bayesian modification and the Monte Carlo method (MCM) to estimate the uncertainty are compared in two practical measurement situations (finding reference value of relative humidity and a generic chemical instrumental analysis procedure). The results of the three approaches agree very well when there are no dominant input quantities with type A evaluated uncertainty estimated from small number of repeated measurements. In the opposite case the GUM gives underestimated expanded uncertainties (by up to 20–25%), compared to both other approaches. Analysis of the practical measurement situations reveals that even in the case of several dominating input quantities of similar uncertainty contributions, if one of them is distributed according to the t-distribution and has a low number (3–4) of degrees of freedom, the output quantity cannot be safely assumed Normally distributed and in such a case coverage factor 2 is not an equivalent to 95% coverage level.