The taxonomy of Salvia: a test of two radically different numerical methods

Forty-one variations in calyx morphology and venation, stamen form, trichomes, style shape and aspects of habit and gross vegetative morphology were recorded for a cosmopolitan sample of 100 Salvia species. Comparative data were subjected to computation under two fundamentally different programs, multclas (a mixed-data method) and centclas (taxonomically crude, strictly qualitative). The resulting classifications were essentially similar, but not identical; and inspection suggests that the differences result largely from taxonomic distortion inherent in the centclas method. A classification of these Salvias is suggested, taken from the multclas hierarchy, which is fundamentally different from anything yet proposed by taxonomists. Of the five subgenera, only Calosphace has gained support; and of the twelve sections only Hymenosphace is homogeneous. The new scheme seems taxonomically reasonable, and is informative in that the groupings are consistent with observations other than those on which it was based. The main subdivision represents a distinction between Central and South American Salvias and the rest, and reflects variation in stylar morphology, though neither geography nor stylar form were subjected to computation. It is concluded that even OENTCLAS, used alone, could have greatly enhanced the taxonomic understanding of this well-known genus.