There was no obvious pattern to indicate that the valence of traits was decisive in determining the locus of activation.
For random taxon coverage patterns, such as might be seen in data sets derived from EST or BAC libraries, upper and lower bounds on the number of loci needed for decisive tree inference were obtained.
The result of this direct comparison is confirmed by the "decisive" [ 47] Bayes factor (>6,000 for all loci, Table 4).
A posteriori the pattern of taxon coverage is trivially decisive for all trees because there are 6 loci (out of 9481) that have complete coverage which automatically satisfies the four-way condition described earlier.
For the most likely scenario (Overlapping) we have computed the Bayes factors between the selection models SOS and LOS and we found a decisive support to SOS over LOS for all loci (Bayes factors > 2,700).
Log Bayes factor difference between the discrete and continuous model for COI was −669 and for ZF −215, thus indicating decisive support for the continuous model for both loci.
Bayes factors for global outliers were high, with seven loci having log10 values above 2 (decisive) and four of these had a log10 Bayes factor of 5, which corresponds to a posterior probability of one.
A pattern of taxon coverage is decisive for some tree on all the taxa in a multi-locus data set if that tree is uniquely defined by combining the (correctly inferred and perfectly concordant) trees from the separate loci.
If we were to continue sampling loci with the same coverage density (14%) in the hopes of constructing a data set with high probability of being decisive for the true tree (Theorem 2, with perfect resolution), 2932 loci would be minimally necessary, and 23,891 would suffice.
For fixed coverage patterns we have determined lower bounds on the number of loci needed for a coverage pattern to be phylogenetically decisive for all trees that is to allow reconstruction of a unique tree on all taxa irrespective of what the tree is.
