Towards An Unbiased Phylogenetic Reconstruction of Taxonomic Evolution

Our perception of taxonomic richness and evolutionary rates from the fossil record is distorted by incomplete and variable preservation over time. One common approach to lessen this bias and infer true underlying biological patterns is to use model phylogenetic inferences to reconstruct unsampled lineages, or unpreserved segments of preserved lineages. Many workers have noted that this approach produces a biased adjustment of lineages by extending lineage durations only prior to their first occurrences in the fossil record, and failing to account for unsampled durations following their last occurrences. Furthermore, this approach only adjusts half of the first occurrences, essentially assuming the other half is known without error.

Here, I present a novel approach to infer the true durations of taxa from their observed stratigraphic ranges and a model phylogenetic estimate. This approach follows the same logic as previous methods, namely that phylogenies imply gaps in the fossil record. Using a model phylogeny and the gaps it implies, I borrow the general approach of gap analysis to estimate per-interval probabilities of preservation. These probabilities can then be used to calculate the expected proportion of taxa with last occurrences in an interval that cross into the following interval, but were not preserved there. This approach similarly adjusts all first occurrences, not just those implied by the phylogeny. Some strengths, weaknesses, and potential extensions of this approach are demonstrated on several previously published empirical data sets.

In principle, this approach can be used to provide phylogenetically adjusted estimates of origination and extinction rates. Alternatively, it can provide more accurate results if, as in recent non-phylogenetic approaches, evolutionary rates and preservation rates are estimated simultaneously. The model phylogeny can constrain the range of possible parameter values, and potentially reduce analysis time and the number of required data.