Abstract: | A new similarity measure is proposed, the ‘dilution factor spread’ or, derived from it, a ‘goodness of fit’ parameter. This has the advantage that raw data without any transformation can be used, diluted samples are recognized and errors of individual concentration values can easily be included. The use of this similarity coefficient in multivariate cluster analyses to construct dendrograms is shown and compared with the use of the well-known similarity measure of Euclidian distance in attribute space and of the cos θ measure. |