COMBINING DATA FROM INDEPENDENT CHEMICAL-ANALYSIS METHODS

Data from several different chemical analysis methods must often be combined to form an overall mean value and uncertainty. However, the data from these different methods frequently do not "agree" when a statistical comparison is done, so that taking a simple average of all of the individual observations would be inappropriate. Computation of a weighted mean and corresponding uncertainty that take into consideration the random measurement error within each method as well as the disagreement between methods will be described. An example of this approach, applied to the certification of NIST's Standard Reference Materials, will be given.