Evaluation of correlated bias approximations in experimental uncertainty analysis

Abstract

A new method to approximate the effect of correlated bias errors in experimental uncertainty analysis is presented. This new method is shown to be greatly superior to previously published and historical approximations, especially when bias errors are estimated in terms of a percentage of reading. To establish this method, the estimation of the bias limit for experimental results determined from measured variables containing hiases from several elemental sources that are partially or wholly correlated is investigated using a Monte Carlo simulation. For each of four sample data reduction equations, the percentage coverage of the uncertainty limits computed using the new and previous approximate methods is determined for various combinations of elemental bias source correlation. These coverage values are compared with the desired coverage of 95% to see which method is the most consistent. The new method is found to be by far the most consistent method to approximate the effect of correlated bias errors.