Effect of Weighting and the Size of the Attribute Set in Numerical Classification

Abstract

Three generations of plants from seed of 50 individuals of three annual species of wild oats (Avena spp.) were grown in a glasshouse during 1975-77. More than 50 characters, some morphological, some agronomic, were measured on duplicate plants. The morphological attributes were used to classify the individuals and their progeny by various numerical taxonomic strategies.

The effectiveness of the various strategies was assessed by the number of mixed species groups and the degree of amalgamation of duplicates in the final classification. The divisive polythetic method was found to be unsatisfactory. Both the agglomerative polythetic method and the method incorporating initial divisive monothetic splitting followed by maximum likelihood reallocation performed better, particularly when the more random attributes were removed from the data set.

The soundest classification was obtained with the agglomerative polythetic method incorporating individual attribute weighting. Weighting was objective and based on the variability of characters between and within duplicate pairs.