A numerical contribution to the classification of the Poaceae

A numerical classification was prepared for a widely representative sample of 72 grass genera with the use of 52 attributes. An index of similarity was employed which accorded the attributes weights inversely proportional to their frequency in the population studied; clusters were formed such that all the similarities between genera within a cluster exceeded a specified significance level. At each stage of the clustering process tests of significance for homogeneity were imposed. The analysis grouped the genera into three large and two small clusters, with an unclassified residue of 15 genera. All but five of these 15 genera showed evident affinities with one of the clusters. In decreasing order of size the clusters comprised the panicoid, pooid, bambusoid, and oryzoid grasses, with a further cluster composed of Lecomtella, Lygeum, Nardus, and Spartina. The five genera seemingly distinct from any clusters were Coleanthus, Diarrhena, Hubbardia, Uniola, and Micraira.