Modelling uncertainty in a concentration range

Abstract

Combined uncertainty modelling in a concentration range is an important task for laboratories. Despite several models and regression methods reported, unsolved problems remain, including variability and type of distribution of combined uncertainty estimations and the influence of these on modelling. Intralaboratory data of eight trace elements in natural waters by flame atomic absorption spectrometry and interlaboratory data from some ISO standards were used as an experimental basis. Starting from these and applying the bootstrap technique, high relative variability of combined uncertainty estimations (second-order uncertainty) was found, but normal distributions or distributions with small deviations from normality were encountered. Linear and/or variance models are appropriate for modelling the analyzed data when ordinary weighted least-squares or repeat median robust regression methods are applied. The influences of combined uncertainty variability on modelling and the effect of points that do not follow the general trend of the remaining points are discussed. Some other guidelines are offered.