Trendanalysen mit Mann-Kendall-Test und Theil-Schätzer. Teil 1: Berücksichtigung von Messunsicherheiten (German)
Helm, Dieter
Umweltwissenschaften und Schadstoff-Forschung 19 (2007), 2, 91-95
Abstract
Goal and Scope
The influence of the measurement uncertainty on the outcomes of statistical evaluations is shown.
Methods
Examplarily using data from the Environmental Specimen Bank the influence of measurement uncertainties on results of trned analyses (Mann Kendall test and Theil estimator) was tested. Both, the simple and the double measurement uncertainty were tested, as well as the standard deviation of repeated measurements.
Results
Considering measurement uncertainty means that not points but intervals are used. A value yj is then considered larger [smaller] than yi if applies: yj-u(yj>yi+u(yi) [yj+u(yj)<yi-u(yj)], where u(yj) and u(yi) are measurement uncertainties of yj and yi. Mann Kendall test of trend will react more conservatively (in sense of the maintenance of null hypothesis: no trend), when measurement uncertainties are considered. Thus, the resulting trend statements will be safer.
Outlook
For future application a decision is neccessary about the kind of the measurement uncertainty to be used (none, simple, double, standard deviation). Part 2 will deal with weighting of the data.