A comparative study of ordination techniques in numerical taxonomy in relation to racial variation in the ringed snake Natrix natrix (L.)

Abstract

The iniluence ot various types of standardization, similarity coefficient and ordination technique (i.e. non-metric multidimensional scaling, canonical analysis, principal coordinate analysis, Q_and R mode principal component analysis) are tested in relation to the numerical taxonomic analysis of the racial differentiation in the ringed snake, N. natrix. The a priori preferences and various criteria, including portrayal of a ‘known taxonomic model’, indicate that canonical analysis and R mode principal components analysis on column-standardized data (or its dual principal coordinate analysis) are to be recommended. Non-metric multidimensional scaling and lack of appropriate character standardization gave inferior results. SUMMARY Thirty-four numerical taxonomic analyses were run varying the method of data standardization, similarity coefficient and ordination technique (i.e. non-metric multidimensional scaling, canonical analysis, principal coordinate analysis and Q and R mode principal component analysis). These analyses were based on sample populations of both sexes representing the marked racial differentiation within the ringed snake N. natrix. Several a priori preferences are stated and various criteria are discussed for judging the taxonomic performance of the methods. (The methods are compared on the basis of taxonomic performance rather than empirical similarity.) On the basis of the a priori preferences, the similarity of the taxonomic conclusions (robustness) and how well they portrayed a ‘known taxonomic model’, the canonical analysis and principal component-coordinate analyses on data standardized by columns (characters) are to be preferred. The principal component/coordinate techniques on unstandardized data is not recommended neither is the use of non-metric multidimensional scaling.