Abstract
Analyses of evolutionary dynamics depend on how phylogenetic data are time-scaled. Most analyses of extant taxa assume a purely bifurcating model, where nodes are calibrated using the daughter lineage with the older first occurrence in the fossil record. This contrasts with budding, where nodes are calibrated using the younger first occurrence. Here, we use the extensive fossil record of bivalve molluscs for a large-scale evaluation of how branching models affect macroevolutionary analyses. We time-calibrated 91% of nodes, ranging in age from 2.59 to 485 Ma, in a phylogeny of 97 extant bivalve families. Allowing budding-based calibrations minimizes conflict between the tree and observed fossil record, and reduces the summed duration of inferred 'ghost lineages' from 6.76 billion years (Gyr; bifurcating model) to 1.00 Gyr (budding). Adding 31 extinct paraphyletic families raises ghost lineage totals to 7.86 Gyr (bifurcating) and 1.92 Gyr (budding), but incorporates more information to date divergences between lineages. Macroevolutionary analyses under a bifurcating model conflict with other palaeontological evidence on the magnitude of the end-Palaeozoic extinction, and strongly reduce Cenozoic diversification. Consideration of different branching models is essential when node-calibrating phylogenies, and for a major clade with a robust fossil record, a budding model appears more appropriate.