Showing papers by "Jens Enevold Thaulov Andersen published in 2015"

Response to Letter to the Editor: 'On the development of quality assurance in analytical chemistry'

Abstract: One of the main issues of quality assurance is the origin of uncertainty; is the uncertainty introduced by human error [1,2], or, is the uncertainty related to features of the accessible analytical methods? Heydorn clearly advocates for the former, whereas it was proposed in numerous publications [3] that the latter candidates should become the dominating providers of uncertainty. Various statistical tools, such as Q-test [4], Grubb’s test [5], Cochran’s test [5], Hampel’s test [6], t-test and T-test [7], are available for identification and elimination of outliers. For the analytical chemist, would it be reasonable to assume that all these tests were able to identify the same outliers, but this is unfortunately not the case; outliers highlighted by one type of test may pass unnoticed by another test. Such observations are very disturbing for scientists in analytical chemistry, who would like to perform firm decisions after overcoming lengthy and complicated statistical testing. Since testing leads to conflicting decisions, it is time to decide whether or not it is worthwhile to proceed with endless discussions about the quality of testing procedures, which, at the end of the day, is really a matter of statistical science. In addition, testing seems to be more or less superfluous because rejection of outliers ruins the uncertainty picture and hampers comparison of results [3]. Therefore, it is suggested to not perform such testing, which inevitably includes T-testing. The T-test relies on weighting of data with variance, and this would be a viable pathway if it were possible to provide reliable estimates of variance of data sets with a low number of repetitions. However, this is not possible. Variances are determined randomly with small data sets of short-term precision [8], which jeopardizes the validity of decisions and thus renders T-testing obsolete.