Analysis of international comparisons with the minimum variance method

The minimum variance (MV) method is applied to the analysis of international comparisons. Simple evaluation criteria for choosing a proper noise model are described. Allan variance study of the data from a comparison of Zener voltage standards indicates a nearly 1/f² type instability after removal of linear drift. When applied to the comparison data, the MV method with 1/f² noise model gives the closest agreement between participants. Almost identical results are given by a simple two-point nearest neighbor linear regression, whereas assumptions of 1/f and especially white noise give significantly larger scatter of results. Transportation clearly degrades the stability of the transfer standard. The instability can be taken into account by correct, time dependent weighting of the data.