Zur multivariaten Auswertung von Ringversuchen

The advantages and drawbacks of some graphical methods as interpreting aids for cooperative tests are demonstrated using results of slag sample analyses (5 laboratories, 7 components). In cases of low dimensions (components) the method of regular polygons has been found most informative. For higher dimensions modern data analysis techniques, such as principal components analysis (pca), correspondence analysis and discriminant analysis are to be preferred. From pca of correlation matrices one can also derive best choices of feature subsets to construct good polygons. The sensitivity of partial correlation coefficients due to deviating laboratory manners can be assessed by leaving-one-out comparisons.