A recent interlaboratory study that required individual analysts to estimate uncertainty intervals for their results revealed that some experienced chemical analysts have difficulty with measurement uncertainty calculations. To help validate assumptions and calculations implemented in a software tool intended to help these analysts calculate appropriate uncertainties for their measurements, a model problem was constructed and posed to a number of chemical and statistical analysts with interest and expertise in the evaluation of measurement uncertainty. Evaluation of the resulting 95 % level of confidence uncertainty intervals revealed three hidden assumptions that influence the estimates: 1) independence of replicated data, 2) nature of the distributions from which data are sampled, and 3) choice of coverage factor. The nature of data replicates is best decided by those most familiar with the experimental details and is no more relevant to the user of measurement results than other experimental details. However, it is recommended that the symbols used in reporting expanded uncertainties be augmented to specify whether intervals have been calculated using propagation of uncertainty or propagation of distribution methods and, if propagation of uncertainty is used, whether conventional or statistical coverage factors are employed.
Proceedings Title: Metrology, Reliable Measurements for the Development of Mexico Symposium 2006 | | Symposium of Metrology | Centro Nacional de Metrolog a
Conference Dates: October 1, 0025
Conference Location: Quer taro, -1
Conference Title: Metrology, Reliable Measurements for the Development of Mexico Symposium Proceedings
Pub Type: Conferences
measurement of uncertainty, propagation of distributions, propagation of uncertainty