Statistical Character Weighting and Similarity Stability

Abstract

Terpenoid data from seven species ofJuniperus were used to examine:1) the effect of using different character weights upon the same set of OTU’s; 2) the effect of the organization of the initial data base by populations, individuals, and averages on statistical weights and the resulting classification; and3) the effect of the use of exemplars on statistical weights and similarity measures. Cophenetic correlation and numerical taxonomy, along with certain relationships well documented for these species, were used to examine these questions. Equal weighting of characters, advocated by most numerical taxonomists, gave the most distorted results. Thiswas followed by the 1/coefficient of variation, then the weighting of {ie305-01} and finallyF andF-1 weighting producing the highest fidelity to the known similarities. The use of either all individuals or random individuals appeared to be better than the use of population averages for some OTU’s and individuals for other OTU’s and much better than the use of population averages only. The use of exemplars has small effects, mostly causing the OTU’s to cluster more loosely when exemplars were included in the computation of the statistical weights. The use of equal weighting is discussed and strongly discouraged in numerical taxonomy in favor of statistically derived weights.