The Application of Simpson's Criterion of Consistency to Phenetic Classifications

Abstract

Simpson (1961), following Beckner (1959), suggested an elegant criterion of consistency between classification and phylogeny. A taxon is said to be monophyletic at rank j if all organisms referred to it are immediately derived by one or more phylogenetic lineages from organisms which are referred to not more than one taxon of rank j. A taxon of a given rank is minimally monophyletic if it is monophyletic at the same rank. A taxonomic hierarchy is consistent with phylogeny if all its component taxa of supraspecific rank are minimally monophyletic. This criterion of consistency is applicable only to classifications which cover a fossil record sufficient to establish, in part at least, the phylogenetic relations of the populations classified. It has been criticised by Hull (1964) on the grounds that the inapplicability of the definition of minimal monophyly to taxa of specific rank renders it incoherent, and that it is so weak that it allows no testable inferences from classification to phylogeny. The first criticism can be met as follows. It is true, as, Simpson adnuitted, that the iiinimal monophyly condition cannot be applied to taxa of specific rank, since there is no feasible way in which it can be ensured that the range of every species corresponds to a temporal segment of a single phylogenetic lineage. The taxa of specific rank which biologists recognise often range over several distinct phylogenetic lineages (in asexual organisms, for example); conversely a single phylogenetic lineage includes the ranges of several species whenever species whose ranges are not, or only partially, isolated from each other are recognised. But the consistency criterion does not allow the ranges of species to be indiscriminate assemblages of representatives from different lineages, since the ranges of species are indirectly limited by the requirement that the supraspecific taxa in which they are included be minimally monophyletic. The second objection is prima facie more serious. If classifications are entirely based on what is known of the phylogeny of organisms the consistency criterion is merely a convention about the way in which they should represent phylogeny. If, however, classifications are wholly or partially phenetic, that is, based on the relative dissimilarities of populations measured with respect to many attributes, then consistency with phylogeny implies some relationship between the relative dissimilarities of populations and their phylogeny. It can be shown that the postulation of consistency for the classifications obtained by a particular kind of phenetic method implies a precise and testable hypothesis about evolutionary rates. In Jardine et al. (1967) and Jardine (1969) it was pointed out that the single-link (nearest-neighbour) method of cluster analysis, first suggested for use in taxonomy by Florek et al. (1951), is the only hierarchic cluster method which satisfies certain intuitively plausible requirements about the way in which a hierarchic clustering should represent data in the form of a dissimilarity coefficient on a set of populations. These requirements are:-that well-marked clusters in the data should be preserved in the result; that the clustering should impose as little distortion as possible on the data subject to the cluster-preservation condition; and that small changes in the data should producecommensurately small changes in the result. The first two requirements are self-explanatory; the last is motivated by the need to ensure that small errors in the data should not have drastic consequences, and by the requirement that it be possible to make meaningful comparisons between the results obtained when a method is applied