Comparative Analysis of MDB for Different Outliers Detection Methods

Quality control is described by internal and external reliability according to the Delft philosophy. Internal reliability describes the ability to find biases in observational data and is represented accordingly by the minimal detectable bias (MDB). The MDBs formulas of four popular outliers detection methods include data snooping, quasi-accurate detection of gross errors(QUAD), partly least squares (PLS) and the method of simultaneous locating and evaluating multiple gross errors(LEGE) are derived with the given significance level α₀ and given power of test β₀. The numerical example reveals that MDBs are same value when the observations have the same weight. If the clean data are not equally weighted, the MDBs of data snooping are equal to that of QUAD and PLS, but all are slightly smaller than the MDBs of LEGE. If the clean data are correlated observations, the MDBs of QUAD are equal to that of PLS, but both are larger than that of data snooping and smaller than that of LEGE.