Accurate Modelling of Translational Bias and its Application to the Reduction of Thermodynamic Data Series

A modelling procedure is introduced for improvement of the reduction of data with the least-squares method, when the data are subdivided in series and cannot be considered as replicated. It combines both the advantage of preserving the individuality of each series and that of improving the variance estimate which arises from fitting the overall data. The procedure assumes - and includes in the model - a translational bias for every series. The method has been applied to a thermal expansion database of copper below 300 K and to high-accuracy low-temperature gas thermometry data, using several least-squares solvers.