The combination of clock models and substitution models for molecular and morphological data and a model of the process that generates dated phylogenetic trees with fossils comprises a full probabilistic model that generates all data used in the analysis.
This approach can utilize all available fossils as individual data points.
The tree prior distribution describes the diversification process where fossil and extant species are treated as samples from this process.
The placement of fossils and absolute branch times are determined in one joint inference rather than in separate analyses.
Node calibration using fossil constraints has two main drawbacks.
First, having identified fossils as belonging to a clade, a researcher needs to specify a prior distribution on the age of the common ancestor of the clade.
A typical node-dating analysis is sequential: it first uses morphological data from fossil and extant species to identify the topological location of the fossils within a given extant species tree topology, then uses fossil ages to construct calibration densities, and finally uses molecular data to estimate the dated phylogeny.
Treating the different types of data in this sequential manner implies an independence between the processes that produce the different types of data, which is statistically inaccurate and errors at any step can propagate to subsequent analyses.
Exactly how to incorporate information from the fossil record into a phylogenetic analysis remains an active area of research. Relaxed molecular clock models act as prior distributions on lineage-specific substitution rates and their introduction has greatly improved divergence dating methods (Thorne et al. 2006; Rannala and Yang 2007; Drummond and Suchard 2010; Heath et al. These models do not assume a strict molecular clock, instead they allow each branch in the tree to have its own rate of molecular evolution drawn from a prior distribution of rates across branches.However, this approach must be executed with caution and attention to the quality of the fossil record for the clade of interest, as posterior estimates of divergence times are very sensitive to prior calibration densities of selected nodes (Warnock et al. 2015) meaning that erroneous calibrations lead to erroneous results (Heads 2012).The second major concern about node calibration is that the fossilization process is modeled only indirectly and in isolation from other forms of data.Bayesian Markov chain Monte Carlo (MCMC) methods are now the major tool in phylogenetic inference (Yang and Rannala 1997; Mau et al. Stochastic branching models describing the diversification process that generated the tree are typically used as prior distributions for the tree topology and branching times (Yule 1924; Kendall 1948; Nee et al.1999; Huelsenbeck and Ronquist 2001) and are implemented in several widely used software packages (Lartillot et al. 1994; Rannala and Yang 1996; Yang and Rannala 1997; Gernhard 2008; Stadler 2009).