A new model for the inference of population characteristics from experimental data using uncertainties. Application to interlaboratory studies

A new model to make inferences about population characteristics from experimental datasets is presented. It derives concepts and procedures from quantum chemistry. The model uses the observed values and the uncertainty estimates thereof. It provides the different modes of the distribution and for each mode the expectation value, the standard deviation and a percentage indicating the fraction of observations encompassed. An implementation of the model that does not require uncertainty estimates is provided too. In this paper, the model is elaborated and applied to the evaluation of interlaboratory studies. It has, however, a much wider generic application. It is demonstrated that the model can cope with asymmetric, strongly tailing and multimodal distributions and that it is superior to existing techniques (e.g. ISO 5725, robust statistics).