Robust and traditional statistical methods in the establishment of immunoglobulin E target values in external quality assessment program

Background In an EQA scheme, peer group mean is conventionally regarded as the target after removing values exceeding the mean ± 3SD. However, this computation fails if there are some outliers. Methods The outliers in each original data set were separately deleted by traditional and robust statistical methods. The data distributions of original and trimmed data sets were separately tested by Shapiro–Wilks or Kolmogorov–Smirnov Z test. The means of peer groups were used to set IgE targets. The difference between targets set by both methods was tested by 1-sample t test. Results The original data sets were not all Gaussian distributed. After deletion of outliers using robust statistical method, the modified data sets of all tested groups were normally distributed. IgE targets established by robust and traditional methods were significantly different in some groups. The incompatible IgE targets derived all from the original and trimmed data sets which were not Gaussian distributed. Conclusions The premise for choosing a traditional method to delete outliers and using peer group means as targets was that the data sets must be Gaussian distributed. The reasons that caused the targets to be incompatible possibly were the 2 rules to be broken. Robust statistical method was used for deletion of outliers due to the distorted distribution.