Distances, Assumptions and Social Wasps

The use of distance data in phylogenetic inference retains adherents in molecular systematics, despite the well-established cladistic critique of distances. Molecular distance matrices have been supposed to measure evolutionary divergence; however, trees based upon analyses of such matrices are generally uninterpretable as doing so-under realistic conditions (Farris, 1981, 1985, 1986). In addition, conversion of character data (for example, molecular sequences) to distances discards information (e.g. Farris, 1981). Adherents of distances have usually defended the approach by invoking restrictive assumptions (for example, additivity in expectation, so that the distances may be interpreted as statistical estimates of actual evolutionary divergence; Felsenstein, 1984, 1986). It has also been claimed that conversion of character data to distances may be preferable under certain circumstances (Swofford and Olsen, 1990: 422, 425), because: (1) sequences are not intuitively meaningful: (2) p resuppositions about evolutionary processes can be invoked to estimate unseen events (i.e. correct for non-additivity); and (3) numerous methods for distance analyses exist. A superficially more critical “reason” for preferring distances is the argument that parsimony uses only a small fraction of the data, by which is meant the informative characters (e.g. Li and Graur, 1991: 114). Redolent as this argument is of syncretist “criticism” of cladistic classification, it is actually based upon simulations “demonstrating” that parsimony may give incorrect results, relative to distance methods predicated upon the simulated conditions. That is to say, the argument is a conclusion following from the assumptions of the simulations All of these defenses of distances are thus alike, in relying upon a p%ri assumptions about evolutionary processes. But which assumptions should be made? And suppose that the assumptions are violated-how may this be determined? Most studies using distance analyses do not consider these questions at all. Examination of the assumptions of a given distance analysis is instructive, for the data themselves frequently provide evidence that the assumptions are violated. The analysis is then internally contradictory, a point which is emphasized when the results are contrasted with the results from a distance analysis that is not predicated on the same assumptions. Such cases do not only support the preference for character analysis over distances, they argue against a prioti assumptions in phylogenetic inference in general. I will illustrate this with an example from a recent study on social wasps.