On weighting characters differently in different parts of the cladogram

GOLOBOFF, PABLO.; MIRANDE, MARCOS; ARIAS, J. SALVADOR

San Miguel de Tucumán, Argentina

Simposio; XXVII Annual Meeting of the Willi Hennig Society; 2008

Willi Hennig Society

  One commonly heard criticism of weighting methodssuch as successive or implied weighting is that theyassume uniformity of weights throughout the clado-gram. A variant of the same criticism is that currentweighting methods rely on the dubious assumptionthat instances of homoplasy in distant clades are asrelevant as instances of homoplasy in more closelyrelated clades. While the assumption of uniformity canhardly be part of a relevant criticism of weighting (sinceequal weighting also assumes it), developing a methodwhich does not assume it may be desirable. Although itis of course possible to run separate analyses, compart-mentalizing the data and using different weights for eachpart of the tree, the ideal method should consider all theevidence simultaneously and provide a globally optimalsolution to the problem. This paper describes a modi-fication of the method of auto-weighted optimization(Goloboff, 1997) to downweight the cost of eachcharacter-state transformation according to the numberand distance of other transformations between the samestates. The method was experimentally implemented,and tested with simulations to confirm that it performsas expected. However, hypothetical examples show thatsuch type of weighting cannot be considered to be aform of parsimony—that is, a form of the criterion thatcompares trees based on the degree to which similaritiescan be explained as resulting from common ancestry. Ifperhaps realistic, the variable-weights criterion neces-sarily implies that other additional considerations—pos-sibly in conflict with parsimony—come into play.Apparently such undesirable consequences would be anecessary part of any method which weights charactersdifferentially in different parts of the tree. Therefore,uniformity of weights throughout the cladogram seemsto be a sine qua non requirement of the parsimonycriterion, and parsimony-based weighting methods can-not be criticized for assuming uniformity of weights.