Discussion on the Range of Cut-Off Values in Robust Algorithm A

Abstract

ISO 13528-2005 “Statistical methods for use in proficiency testing by interlaboratory comparisons” introduced a robust method for calculating the mean and standard deviation—Algorithm A. In cases where it is not possible to use the classical statistical method to eliminate outliers, Algorithm A can be used without deleting the abnormal data and give a good estimate of population mean and population standard deviation. This method is widely used in proficiency testing to calculate assigned value and standard deviation for proficiency assessment. The iterative calculation of Algorithm A has a parameter c, called cut-off values, which usually takes between 1 and 2. In the actual application process, the cut-off values is often misused, and is taken to more than 2 which is not within the recommended range. By using simulation, this paper shows that the reasonable value of parameter c should be between 1 and 2. When the value exceeds 2, it does not satisfy the actual assumption.