FEW MEASUREMENTS, NON-NORMALITY: A STATEMENT ON THE EXPANDED UNCERTAINTY

In the current GUM approach the law of propagation of uncertainty (LPU) is not used consistently for type A input quantities. This may cause understatement of measurement uncertainty in the case of few repeated measurements. The Welch-Satterthwaite formula only partially solves this issue and is hardly utilized in practice. Performing Monte Carlo simulations (GUM-S1) might be a solution, but many metrologists prefer the use of simply rules of thumb in daily practice. Therefore we suggest a more consistent approach of type A input quantities by using a full Bayesian framework. Our method remains an approximation but we believe it to be more correct and straightforward than the current GUM approach. Also, we allow for the treatment of type A input quantities which arise from non-normal repeated measurements.