Towards a unified model of errors in analytical measurementBased on papers read by the author at SAC99 in Dublin, July 1999, and at FACSS in Vancouver, October 1999.

Abstract

At present there is no single model of errors that defines the relationship between the various quality assurance methods that have evolved in analytical chemistry. As a result, difficulties are encountered in ensuring that these methods are individually necessary and jointly sufficient. A model is suggested here that supplies that need. This model apportions error between method bias, laboratory bias, run bias and repeatability variation (the so called ‘ladder of errors’), together with matrix variability. The model allows us to see our current practices in context and to reassess our priorities. In particular, it demonstrates that high-level errors are usually of a magnitude comparable with the more accessible ones but are likely to be overlooked, and suggests ways in which current practices can be improved.