REGRESSION-MODELS IN EVOLUTIONARY ALLOMETRY

The traditional problem of choosing the best regression model or estimation method in evolutionary allometry requires an assumption about error structure. This question has never been adequately resolved because error structure in evolutionary allometry depends on the underlying biological model and cannot be determined from statistical criteria alone. Possible approaches to this problem include genetic, ontogenetic, and ecological models, and phylogenetic information. While estimates of within-taxon sampling variance may provide information about a purely statistical part of the error structure, a more interesting part of the "error" is due to real differences among taxa. This part of the error structure is not artifactual. It requires biological, not statistical, explanation. Resolution of the statistical problem of regression model choice in evolutionary allometry will probably require explicit biological models analyzed within a phylogenetic framework.