‘ROBUST STATISTICS’: A USEFUL TOOL TO HANDLE OUTLYING OBSERVATIONS IN THE LABORATORY
The curricula defined by the Faculty of Sciences of the University of Burgos for the Degree in Chemistry and for the Degree in CTA (Food Science and Technology) established several competences about the capacity of relating Chemistry with other disciplines, for example, to apply mathematical and statistical methods to validate models fit from experimental data. It is a fact known by any analyst that, when analysing data, outlying observations cause problems because they may strongly influence the results. Robust statistics aims at detecting the outliers by searching for the model fitted by the majority of the data.
In this work we use several robust methods and outlier detection tools for two different goals.
The first one is related to objectively evaluate the skills achieved during the experiments at the lab of Analytical Chemistry, during the development of student’s practices in two courses with a large quantity of practical training (4 out of 6 ECTS are devoted to practices). Both courses are in the third semester in the degrees in Chemistry and CTA.
A second goal refers to the introduction of some concepts about ‘robust statistics’ and their usefulness in chemical analysis. This is done in the course “Chemometrics and experimentation in Analytical Chemistry” which is in the sixth semester of the degree in Chemistry. Concretely, statistical procedures for univariate data (estimators of location and scatter) are revised, and other procedures for detecting outliers are introduced by using an interlaboratory trial, that would need of a study of z-scores when analysing the data. In this way, practices are used to introduce some alternatives that appear in part 5 of the ISO-5725 [Accuracy, Trueness and precision of measurement methods and results, Genève, 1994]. Part 5 of the norm provides alternative designs that may be more valuable in some situations than the basic procedures proposed in ISO-5725-2, suggesting robust methods of analysis that give estimates of location and scatter that are less dependent on the data analyst’s judgement than those given by mean and standard deviation. Box and whisker plots, median, winsorized mean, winsorized sigma, MAD-scores, etc. are among the topics handled, which serve both, student and teacher, to perform a more proper analysis of the data and an objective evaluation of the skills acquired by students.
To achieve these goals, all the practices are performed under the supervision of a group of teachers (teachers of Analytical Chemistry and teachers of Statistics) working together in a shared course, so that students perceive the transverse character of the acquired knowledge and how this knowledge is used in practice.