A Comparative Method

The main methodological problem of comparative biology is to distinguish between effects attributable to ancestry and those attributable to function. The nested analysis of variance addresses this problem by partitioning variance and covariance between taxonomic levels. The estimation of variance components and the derived intraclass correlation coefficient can be used to describe the overall increase in variation through time and the relation between variances at different taxonomic levels. Applying this procedure to data about the body mass of eutherian mammals showed a steep increase in variance associated with the adaptive radiation of the group in early Tertiary times, with much lower rates of increase thereafter; chromosome number showed a similar but shallower increase, implying that chromosome number varies more between taxa at lower levels of classification than does body mass. The variance of species within genera increased with the geologic age of the genus, permitting a quantitative description of the rate of evolutionary divergence. A straightforward extension of the technique yields estimates of taxonomic components of covariance and the associated interclass correlation coefficients. It is argued that covariance between taxa is attributable to ancestry and covariance within taxa to function, and that adaptive hypotheses should be tested by the within-taxon correlation. This technique is illustrated by analyzing the covariance of litter mass with gestation period, independent of body mass, among eutherians. This reveals substantial correlation within species, attributed to function, and between suborders and orders, where it was attributed to ancestry, with little or no correlation at intermediate taxonomic levels. The components of variance and covariance can be used to describe how the correlation changes through time, in this case declining continuously from the time of divergence of suborders onward. Two sources of phylogenetic constraint are identified: the taxonomic components of covariance themselves, which may reflect the past operation of selection; and differences among correlation or regression coefficients among taxa at a given level of classification, representing an interaction between phylogeny and function, which may arise when a third variable alters the pattern of selection.