Abstract
Outliers are always frequent in all analytical laboratories and ever source of trouble, mainly in food owing to the heterogeneity in the matrixes. Although there are several formal statistical tests to identify those data, publications lack enough information about their practical results. Box plots, stem-and-leaf and other graphics are not included in a routine of these laboratories and their results are not known, therefore these and other graphics, in addition to Dixon, Grubbs and Hampel tests, being the last one more recent than the other, and the robust Z-score (Zr) were employed when revisiting previously published interlaboratory experiments. The results showed some disagreement among the conclusions from the graphics, the same among the numerical methods and among graphics and numeric methods. In addition to the box plots, the stem-and-leaf could provide an effective visualization of outliers. The Zr score showed the highest number of potential aberrant values and the Hampel test was the most appropriated test. The results of outliers in one of the four revisited experiments were in agreement with the results of this research, however, those regarding the other three revisited interlaboratories were not. This research suggested to use the experimental design model to study the measures, when possible, and to examine the cases with more than 1% potential outliers.
Keywords:
Metrology; Quality in laboratories; Single validation; Proficiency tests; Collaborative studies; Box plot; Stem-and-leaf